EditorRichard J. Bucala, MD, PhDYale University School of Medicine, New Haven

Deputy EditorDaniel H. Solomon, MD, MPH, Boston

Co-EditorsJoseph E. Craft, MD, New HavenDavid T. Felson, MD, MPH, BostonRichard F. Loeser Jr., MD, Chapel HillPeter A. Nigrovic, MD, BostonJanet E. Pope, MD, MPH, FRCPC, London, OntarioChristopher T. Ritchlin, MD, MPH, RochesterBetty P. Tsao, PhD, CharlestonJohn Varga, MD, Chicago

Co-Editor and Review Article EditorRobert Terkeltaub, MD, San Diego

Clinical Trials AdvisorMichael E. Weinblatt, MD, Boston

Social Media EditorPaul H. Sufka, MD, St. Paul

Journal Publications CommitteeShervin Assassi, MD, MS, Chair, HoustonVivian Bykerk, MD, FRCPC, New YorkCecilia P. Chung, MD, MPH, NashvilleMeenakshi Jolly, MD, MS, ChicagoKim D. Jones, RN, PhD, FNP, PortlandMaximilian Konig, MD, BaltimoreLinda C. Li, PT, MSc, PhD, VancouverUyen-Sa Nguyen, MPH, DSc, Fort Worth

Editorial Staff Jane S. Diamond, MPH, Managing Editor, AtlantaMaggie Parry, Assistant Managing Editor, AtlantaLesley W. Allen, Senior Manuscript Editor, AtlantaKelly Barraza, Manuscript Editor, AtlantaJessica Hamilton, Manuscript Editor, AtlantaIlani S. Lorber, MA, Manuscript Editor, AtlantaEmily W. Wehby, MA, Manuscript Editor, AtlantaSara Omer, Editorial Coordinator, AtlantaBrittany Swett, Assistant Editor, New HavenCarolyn Roth, Senior Production Editor, Boston

Associate EditorsDaniel Aletaha, MD, MS, ViennaHeather G. Allore, PhD, New HavenDaniel J. Clauw, MD, Ann ArborRobert A. Colbert, MD, PhD, BethesdaKaren H. Costenbader, MD, MPH, BostonNicola Dalbeth, MD, FRACP, AucklandKevin D. Deane, MD, DenverMark C. Genovese, MD, Palo Alto

Insoo Kang, MD, New HavenWan-Uk Kim, MD, PhD, SeoulCarol Langford, MD, MHS, ClevelandKatherine Liao, MD, MPH, BostonS. Sam Lim, MD, MPH, AtlantaAnne-Marie Malfait, MD, PhD, ChicagoPaul A. Monach, MD, PhD, BostonChester V. Oddis, MD, Pittsburgh

Andras Perl, MD, PhD, SyracuseJack Porrino, MD, New HavenTimothy R. D. J. Radstake, MD, PhD, UtrechtWilliam Robinson, MD, PhD, Palo AltoGeorg Schett, MD, ErlangenNan Shen, MD, ShanghaiRonald van Vollenhoven, MD, PhD, AmsterdamFredrick M. Wigley, MD, Baltimore

Advisory EditorsAbhishek Abhishek, MD, PhD, NottinghamTom Appleton, MD, PhD, London,

Ontario Bonnie Bermas, MD, DallasLiana Fraenkel, MD, MPH, New HavenMonica Guma, MD, PhD, La Jolla Nigil Haroon, MD, PhD, Toronto

Erica Herzog, MD, PhD, New HavenHui-Chen Hsu, PhD, BirminghamJ. Michelle Kahlenberg, MD, PhD, Ann ArborMariana J. Kaplan, MD, BethesdaJonathan Kay, MD, WorcesterFrancis Lee, MD, PhD, New HavenSang-Il Lee, MD, PhD, Jinju

Rik Lories, MD, PhD, Leuven Bing Lu, PhD, BostonSuresh Mahalingam, PhD, Southport,

Queensland Aridaman Pandit, PhD, Utrecht Kevin Winthrop, MD, MPH, PortlandKazuki Yoshida, MD, MPH, MS, Boston

AMERICAN COLLEGE OF RHEUMATOLOGY

Paula Marchetta, MD, MBA, New York, PresidentEllen M. Gravallese, MD, Worcester, President-ElectDavid R. Karp, MD, PhD, Dallas, Treasurer

Kenneth G. Saag, MD, MSc, Birmingham, SecretarySteven Echard, IOM, CAE, Atlanta, Executive Vice-President

© 2019 American College of Rheumatology. All rights reserved. No part of this publication may be reproduced, stored or transmitted in any form or by any means without the prior permission in writing from the copyright holder. Authorization to copy items for internal and personal use is granted by the copyright holder for libraries and other users registered with their local Reproduction Rights Organization (RRO), e.g. Copyright Clearance Center (CCC), 222 Rosewood Drive, Danvers, MA 01923, USA (www.copyright.com), provided the ap-propriate fee is paid directly to the RRO. This consent does not extend to other kinds of copying such as copying for general distribution, for advertising or promotional purposes, for creating new collective works or for resale. Special requests should be addressed to: permissions@wiley.com

Access Policy: Subject to restrictions on certain backfi les, access to the online version of this issue is available to all registered Wiley Online Library users12 months after publication. Subscribers and eligible users at subscribing institutions have immediate access in accordance with the relevant subscriptiontype. Please go to onlinelibrary.wiley.com for details.

The views and recommendations expressed in articles, letters, and other communications published in Arthritis & Rheumatology are those of the authors and do not necessar-ily refl ect the opinions of the editors, publisher, or American College of Rheumatology. The publisher and the American College of Rheumatology do not investigate the informa-tion contained in the classifi ed advertisements in this journal and assume no responsibility concerning them. Further, the publisher and the American College of Rheumatology do not guarantee, warrant, or endorse any product or service advertised in this journal.

Cover design: Todd Machen

This journal is printed on acid-free paper.∞

Arthritis & RheumatologyAn Offi cial Journal of the American College of Rheumatology

www.arthritisrheum.org and wileyonlinelibrary.com

ART_v71_i8_Ed-board.indd 1ART_v71_i8_Ed-board.indd 1 23-07-2019 11:26:0023-07-2019 11:26:00

Arthritis & RheumatologyAn Offi cial Journal of the American College of Rheumatology

www.arthritisrheum.org and wileyonlinelibrary.com

In This Issue .............................................................................................................................................................................. A15 Clinical Connections ............................................................................................................................................................... A17

Special Articles Editorial: Evolving Use of Molecular Imaging in Research and in Practice

Ahmed Tawakol, Sebastian Unizony, Michael T. Osborne, Elena Massarotti, and Jon T. Giles ....................................................... 1207 Editorial: Picturing Giant Cell Arteritis: Projecting Into the Future

Sara K. Tedeschi and Ayaz Aghayev ....................................................................................................................................................... 1211 In Memoriam: Shaun Ruddy, MD, 1935–2019

Michael E. Weinblatt and William N. Kelley .......................................................................................................................................... 1115

Rheumatoid Arthritis Investigating Asthma, Allergic Disease, Passive Smoke Exposure, and Risk of Rheumatoid Arthritis

Vanessa L. Kronzer, Cynthia S. Crowson, Jeff rey A. Sparks, Robert Vassallo, and John M. Davis III ............................................... 1217 Derivation and Validation of a Major Toxicity Risk Score Among Nonsteroidal Antiinfl ammatory Drug Users Based on Data From a Randomized Controlled Trial

Daniel H. Solomon, Ming Shao, Kathy Wolski, Steven Nissen, M. Elaine Husni, and Nina Paynter ................................................ 1225 Development and Validation of an 18 F-Fluorodeoxyglucose–Positron Emission Tomography With Computed Tomography–Based Tool for the Evaluation of Joint Counts and Disease Activity in Patients With Rheumatoid Arthritis

Sang Jin Lee, Ju Hye Jeong, Chang-Hee Lee, Byeong-Cheol Ahn, Jung Su Eun, Na Ri Kim, Jong Whan Kang, Eon Jeong Nam, and Young Mo Kang .................................................................................................................................................... 1232

C itrullinated Inhibitor of DNA Binding 1 Is a Novel Autoantigen in Rheumatoid Arthritis Ray A. Ohara, Gautam Edhayan, Stephanie M. Rasmussen, Takeo Isozaki, Henriette A. Remmer, Thomas M. Lanigan, Phillip L. Campbell, Andrew G. Urquhart, Jeff rey N. Lawton, Kevin C. Chung, David A. Fox, and Jeff rey H. Ruth ......................... 1241

Activation of the Peroxisome Proliferator–Activated Receptor γ Coactivator 1β/NFATc1 Pathway in Circulating Osteoclast Precursors Associated With Bone Destruction in Rheumatoid Arthriti s

Jian-Da Ma, Jun Jing, Jun-Wei Wang, Ying-Qian Mo, Qian-Hua Li, Jian-Zi Lin, Le-Feng Chen, Lan Shao, Pierre Miossec, and Lie Dai ................................................................................................................................................................................................ 1252

Amelioration of Autoimmune Arthritis in Mice Treated With the DNA Methyltransferase Inhibitor 5′-Azacytidine Dániel M. Tóth, Timea Ocskó, Attila Balog, Adrienn Markovics, Katalin Mikecz, László Kovács, Meenakshi Jolly, Aleksandra A. Bukiej, Andrew D. Ruthberg, András Vida, Joel A. Block, Tibor T. Glant, and Tibor A. Rauch ................................. 1265

Osteoarthritis Disease Burden in Osteoarthritis Is Similar to That of Rheumatoid Arthritis at Initial Rheumatology Visit and Signifi cantly Greater Six Months Late r

Jacquelin R. Chua, Shakeel Jamal, Mariam Riad, Isabel Castrejon, Anne-Marie Malfait, Joel A. Block, and Theodore Pincus ....................................................................................................................................................................................... 1276

Prioritization of PLEC and GRINA as Osteoarthritis Risk Genes Through the Identifi cation and Characterization of Novel Methylation Quantitative Trait Loci

Sarah J. Rice, Maria Tselepi, Antony K. Sorial, Guillaume Aubourg, Colin Shepherd, David Almarza, Andrew J. Skelton, Ioanna Pangou, David Deehan, Louise N. Reynard, and John Loughlin ............................................................ 1285

Systemic Lupus Erythematosus Evaluating the Properties of a Frailty Index and Its Association With Mortality Risk Among Patients With Systemic Lupus Erythematosus

Alexandra Legge, Susan Kirkland, Kenneth Rockwood, Pantelis Andreou, Sang-Cheol Bae, Caroline Gordon, Juanita Romero-Diaz, Jorge Sanchez-Guerrero, Daniel J. Wallace, Sasha Bernatsky, Ann E. Clarke, Joan T. Merrill, Ellen M. Ginzler, Paul Fortin, Dafna D. Gladman, Murray B. Urowitz, Ian N. Bruce, David A. Isenberg, Anisur Rahman, Graciela S. Alarcón, Michelle Petri, Munther A. Khamashta, M. A. Dooley, Rosalind Ramsey-Goldman, Susan Manzi, Kristjan Steinsson, Asad A. Zoma, Cynthia Aranow, Meggan Mackay, Guillermo Ruiz-Irastorza, S. Sam Lim, Murat Inanc, Ronald F. van Vollenhoven, Andreas Jonsen, Ola Nived, Manuel Ramos-Casals, Diane L. Kamen, Kenneth C. Kalunian, Soren Jacobsen, Christine A. Peschken, Anca Askanase, and John G. Hanly ................................................ 1297

Pim-1 as a Therapeutic Target in Lupus Nephritis Rong Fu, Yong Xia, Meirong Li, Renxiang Mao, Chaohuan Guo, Mianjing Zhou, Hechang Tan, Meiling Liu, Shuang Wang, Niansheng Yang, and Jijun Zhao .................................................................................................................................. 1308

Clinical Images Tracheobronchial Cobblestone in Relapsing Polychondritis

Eun Bong Lee and Jin Kyun Park ............................................................................................................................................................ 1318

VOLUME 71 • August 2019 • NO. 8

ART_v71_i8_Toc.indd 1ART_v71_i8_Toc.indd 1 23-07-2019 11:30:3423-07-2019 11:30:34

Vasculitis Diagnostic Accuracy of Positron Emission Tomography/Computed Tomography of the Head, Neck, and Chest for Giant Cell Arteritis: A Prospective, Double-Blind, Cross-Sectional Study

Anthony M. Sammel, Edward Hsiao, Geoff rey Schembri, Katherine Nguyen, Janice Brewer, Leslie Schrieber, Beatrice Janssen, Peter Youssef, Clare L. Fraser, Elizabeth Bailey, Dale L. Bailey, Paul Roach, and Rodger Laurent ................... 1319

Glucocorticoid Dosages and Acute-Phase Reactant Levels at Giant Cell Arteritis Flare in a Randomized Trial of Tocilizumab

John H. Stone, Katie Tuckwell, Sophie Dimonaco, Micki Klearman, Martin Aringer, Daniel Blockmans, Elisabeth Brouwer, Maria C. Cid, Bhaskar Dasgupta, Juergen Rech, Carlo Salvarani, Hendrik Schulze-Koops, Georg Schett, Robert Spiera, Sebastian H. Unizony, and Neil Collinson ........................................................................................... 1329

Systemic Sclerosis Prevalence, Treatment, and Outcomes of Coexistent Pulmonary Hypertension and Interstitial Lung Disease in Systemic Sclerosis

Amber Young, Dharshan Vummidi, Scott Visovatti, Kate Homer, Holly Wilhalme, Eric S. White, Kevin Flaherty, Vallerie McLaughlin, and Dinesh Khanna ............................................................................................................................................. 1339

I dentifi cation of Cysteine-Rich Angiogenic Inducer 61 as a Potential Antifi brotic and Proangiogenic Mediator in Scleroderma

Pei-Suen Tsou, Dinesh Khanna, and Amr H. Sawalha ......................................................................................................................... 1350

Myositis The IgG2 Isotype of Anti–Transcription Intermediary Factor 1γ Autoantibodies Is a Biomarker of Cancer and Mortality in Adult Dermatomyositis

Audrey Aussy, Manuel Fréret, Laure Gallay, Didier Bessis, Thierry Vincent, Denis Jullien, Laurent Drouot, Fabienne Jouen, Pascal Joly, Isabelle Marie, Alain Meyer, Jean Sibilia, Brigitte Bader-Meunier, Eric Hachulla, Mohammed Hamidou, Sophie Huë, Jean-Luc Charuel, Nicole Fabien, Pierre-Julien Viailly, Yves Allenbach, Olivier Benveniste, Nadège Cordel, Olivier Boyer, and the OncoMyositis Study Group ................................................................... 1360

Brief Report: Myositis Autoantigen Expression Correlates With Muscle Regeneration but Not Autoantibody Specifi city

Iago Pinal-Fernandez, David R. Amici, Cassie A. Parks, Assia Derfoul, Maria Casal-Dominguez, Katherine Pak, Richard Yeker, Paul Plotz, Jose C. Milisenda, Josep M. Grau-Junyent, Albert Selva-O'Callaghan, Julie J. Paik, Jemima Albayda, Andrea M. Corse, Thomas E. Lloyd, Lisa Christopher-Stine, and Andrew L. Mammen ...................................... 1371

Expression of ConcernExpression of Concern Regarding “Eff ects of Hypoxia on the Expression and Activity of Cyclooxygenase 2 in Fibroblast-Like Synoviocytes: Interactions With Monocyte-Derived Soluble Mediators” (Arthritis Rheum 2004;50:2441–9) ..........................................................................................................................................................................................1376

Pediatric Rheumatology Galectin-9 and CXCL10 as Biomarkers for Disease Activity in Juvenile Dermatomyositis: A Longitudinal Cohort Study and Multicohort Validatio n

Judith Wienke, Felicitas Bellutti Enders, Johan Lim, Jorre S. Mertens, Luuk L. van den Hoogen, Camiel A. Wijngaarde, Joo Guan Yeo, Alain Meyer, Henny G. Otten, Ruth D. E. Fritsch-Stork, Sylvia S. M. Kamphuis, Esther P. A. H. Hoppenreijs, Wineke Armbrust, J. Merlijn van den Berg, Petra C. E. Hissink Muller, Janneke Tekstra, Jessica E. Hoogendijk, Claire T. Deakin, Wilco de Jager, Joël A. G. van Roon, W. Ludo van der Pol, Kiran Nistala, Clarissa Pilkington, Marianne de Visser, Thaschawee Arkachaisri, Timothy R. D. J. Radstake, Anneke J. van der Kooi, Stefan Nierkens, Lucy R. Wedderburn, Annet van Royen-Kerkhof, and Femke van Wijk .............................................................................................. 1377

Letter Excess Deaths Upon Cessation of Xanthine Oxidase Inhibitor Treatment—Data From the Cardiovascular Safety of Febuxostat and Allopurinol in Patients With Gout and Cardiovascular Morbidities Trial: Comment on the Article by Choi et al

Michael R. Bubb ....................................................................................................................................................................................... 1391

Clinical Images Dual-Energy Computed Tomography for the Noninvasive Diagnosis of Coexisting Gout and Calcium Pyrophosphate Deposition Disease

Rami Hajri, Steven D. Hajdu, Thomas Hügle, Pascal Zuff erey, Laurent Guiral, and Fabio Becce ................................................... 1392

Cover image: The fi gure on the cover (from Sammel et al, page 1319) is a fusion axial head PET/CT slice from a patient with acute biopsy-positive giant cell arteritis. Intense FDG tracer uptake is depicted by the orange/yellow color. This is seen in the bilateral superfi cial temporal arteries (anterior to the auditory canals) and maxillary arteries (adjacent to the pterygomaxillary fi ssures) and indicates active arteritis. Uptake in the cerebellum is a normal fi nding on PET/CT.

ART_v71_i8_Toc.indd 2ART_v71_i8_Toc.indd 2 23-07-2019 11:30:3423-07-2019 11:30:34

In this IssueHighlights from this issue of A&R | By Lara C. Pullen, PhD

Frailty Index for Patients With Systemic Lupus ErythematosusIn this issue, Legge et al (p. 1297) report on their derivation and evaluation of the properties of a frailty index (FI) constructed using data from the Systemic

Lupus International Collaborating Clinics (SLICC) inception

cohort. They used a well-characterized, international cohort of patients, who were enrolled within 15 months of systemic lupus erythematosus (SLE) diagnosis, to create a novel, internally validated health measure for SLE.

The baseline data set included 1,683 patients with SLE, 89% of whom were female. The patients had a mean ± SD age of 35.7 ± 13.4 years and a mean ± SD disease duration of 18.8 ± 15.7 months. At baseline, the mean ± SD SLICC-FI score was 0.17 ± 0.08, and the baseline SLICC-FI values exhibited the expected measure-ment properties, such as their associations with sex and age, that had been consis-tently demonstrated by other frailty indices

in non-lupus populations. The baseline SLICC-FI values were also only weakly correlated with baseline SLICC/American College of Rheumatology Damage Index (SDI) scores. After adjusting for age, sex, steroid use, ethnicity/region, and base-line SDI scores, higher baseline SLICC-FI values were associated with increased

mortality risk. The investigators concluded that the SLICC-FI score is potentially valu-able for quantifying vulnerability among patients with SLE and, therefore, adds to existing prognostic scores, especially with predicting future risk of mortality. Frailty also may prove useful in assessing clinical outcome from experimental interventions.

Positron emission tomography (PET) can identify the presence and intensity of infl ammation in arthritic joints. However, it is still unclear to what extent PET-assessed joint counts might be useful in clarifying positive fi ndings of synovitis on clinical joint counts or the extent

of disease activity. While clinical joint count assessment is important for detecting synovitis, its reliability is a subject of controversy. In this

issue, Lee et al (p. 1232) report on their study in which they assess the correlation of PET-derived parameters in 68 joints with disease activity and compare the reliability of joint counts between PET with computed tomography (CT) and clinical assessment in patients with rheumatoid arthritis (RA). Their data indicate that PET-CT could serve as a sensitive and reliable method for evaluating disease activity in RA patients.

The investigators found that the number of PET-positive joints was signifi cantly correlated with the swollen joint count (SJC),

the tender joint count (TJC), and the Disease Activity Score in 28 joints using the erythrocyte sedimentation rate (DAS28-ESR). Moreover, the intraobserver and interobserver reliability of PET for the affected joint counts was excellent. Specifi cally, interobserver reliability between nuclear medicine physicians and rheumatologists was good for the SJC and TJC in both 28 joints and 68 joints.

The researchers performed multivariate analyses that included the ESR and patient global assessment of disease activity (PtGA), in addition to PET-derived parameters, and derived the PET/DAS as follows: (0.063 × number of PET-positive joints in 28 joints) + (0.011 × ESR) + (0.030 × PtGA). In the validation group they found a signifi cant correlation between the PET/DAS and the DAS28-ESR. The results suggest that the PET/DAS might be a useful research tool, particularly in clinical trials.

Development and Validation of a PET-CT Tool for Rheumatoid Arthritis

p. 1297

p. 1232

Figure 1. Observed distribution of the SLICC-FI scores at baseline (n = 1,682) (SLICC-FI scores could not be calculated in 1 patient due to missing data) and at last follow-up visit (n = 1,507) among SLE patients in the SLICC SLE inception cohort.

15

Derivation and Validation of a Major Toxicity Risk Score for NSAID UseNonsteroidal antiinflammatory drugs (NSAIDs) are commonly prescribed, even though they may cause major toxicity. A more precise understanding

of the risk associ-ated with this class of drugs would make it

possible for rheumatologists to more accu-rately calculate the risk/benefi t ratio for a given patient. In this issue, Solomon et al (p. 1225) report on their derivation and validation of a risk score for major toxicity among a large cohort of well-phenotyped

NSAID users enrolled in a randomized clinical trial.

The researchers found multiple signif-icant variables in their derivation cohort: age, male sex, history of cardiovascular disease, hypertension, diabetes mellitus, tobacco use, statin use, elevated serum creatinine level, hematocrit level, and type of arthritis. They assigned patients to 3 risk groups (low, intermediate, and high) and found that the risk score could accurately categorize the 1-year risk of major toxicity among NSAID users with a C-index of 0.73

Both osteoarthritis (OA) and rheumatoid arthritis (RA) have primary symptoms of pain and functional disability. Although historically RA has been regarded as a more severe form of arthritis than OA, one previous study suggested a similar or greater disease burden in OA compared

to RA, which may have been explained by better RA treatments. In this issue, Chua et al (p. 1276) report their analysis of disease burden in OA

using Multidimensional Health Assessment Questionnaire (MDHAQ)/Routine Assessment of Patient Index Data (RAPID3) scores at the initial visit and a 6-month follow-up visit to one rheumatology site. The investigators used RA as a benchmark for high disease burden and compared the OA scores to the RA scores.

The retrospective study included patients at the Rush University Medical Center in Chicago, Illinois, at which all patients complete the MDHAQ in paper form at all visits in routine care. Responses are saved as PDF fi les in the electronic health record. The researchers examined responses for new OA and RA patients seen between May 2011 and February 2017, and calculated 0–10 MDHAQ scores for physical function, pain, and global assessment compiled into composite 0–30 RAPID3 scores, at the initial and the 6-month follow-up visits. They classifi ed patients as self- or physician referred, and RA patients were classifi ed as disease-modifying antirheumatic drug (DMARD) naive or as having had prior DMARD treatment.

Compared to RA patients, OA patients had higher age, body mass index (BMI), and disease duration, and all analyses were adjusted for these variables. The investigators found no signifi cant difference in RAPID3 scores between OA versus RA patients, whether DMARD naive or having prior DMARD treatment or

whether self- or physician-referred. After 6 months, however, while all patients were improved, RAPID3 scores were improved signifi cantly more in RA patients than in OA patients following adjustment for age, BMI, disease duration, education level, and ethnicity. Thus, while mean MDHAQ/RAPID3 scores were similar in OA and RA patients at the initial visit, 6 months later mean scores were higher in OA patients, although at an individual level, some patients with RA had higher disease than OA patients and vice versa. These fi ndings likely refl ect superior RA treatments. The results also suggest that the same MDHAQ/RAPID3 allows comparison of disease burdens in different diseases.

Disease Burden of OA Similar to RA at Initial Visit to a Rheumatology Site

in the validation cohort and 0.71 in the total cohort. When the investigators applied the model to the total population with complete data (n = 23,735), they found that 4.6% had a predicted 1-year risk of major toxicity of <1%, 68.6% had a predicted risk of 1–4%, and 26.9% had a predicted risk of >4%. Although it is still important to perform external validation, the split sampling and large cohort in this study would suggest that the risk score is likely useful for strati-fying patients and identifying those who can safely use NSAIDs.

p. 1225

p. 1276

Figure 1. MDHAQ/RAPID3 scores at the initial visit and a 6-month follow-up visit in patients with OA, patients with RA and prior DMARD treatment, and patients with RA and no prior DMARD treatment. Each box represents the interquartile range (IQR), lines inside the boxes show the median, and lines outside the boxes show the 10th and 90th percentiles. *** = P < 0.001.

1

Clinical Connections

SUMMARY Myositis is a heterogeneous family of diseases that causes muscle weakness. Muscle biopsy samples from each type of myositis, despite their clinical differences, reveal ongoing muscle damage as well as muscle cell regeneration. Interestingly, each type of myositis is associated with a different myositis-specific autoantibody (MSA). Prior studies have shown that some of the myositis autoantigens are expressed at low levels in normal muscle and at high levels in regenerating muscle. Thus, it has been proposed that the selective increased expression of a single myositis autoantigen in a given type of myositis triggers development of MSAs against that autoantigen. In this study by Pinal-Fernandez et al, gene expression profiling of muscle tissue samples obtained from control subjects and patients showed that all myositis autoantigens are expressed at high levels in regenerating muscle from all types of myositis. Furthermore, all myositis autoantigens were expressed at high levels in regenerating mouse muscle cells following muscle injury and in immature human muscle cells grown in culture dishes. Therefore, myositis autoantigen overexpression is a normal part of muscle regeneration, and restricted autoantigen overexpression alone does not account for why myositis patients typically produce only a single MSA.

Myositis Autoantigen Expression Correlates With Muscle Regeneration but Not Autoantibody Specificity Pinal-Fernandez et al, Arthritis Rheumatol 2019;71:1371–1376

CORRESPONDENCE Iago Pinal-Fernandez, MD, PhD: iago.pinalfernandez@nih.govAndrew L. Mammen, MD, PhD: andrew.mammen@nih.gov

KEY POINTS

• Myositis autoantigen expression correlates directly with the expression of markers of muscle regeneration and inversely with the expression of genes encoding mature muscle proteins in human myositis muscle biopsy tissue and regenerating mouse muscle.

• Myositis autoantigens are highly expressed during muscle differentiation in cultured human muscle cells.

• The expression of a given autoantigen in myositis muscle was not associated with autoantibodies recognizing that autoantigen.

17

Clinical Connections

SUMMARY The mechanisms of osteoclastogenesis and the role of circulating osteoclast precursors in rheumatoid arthritis (RA) are unclear. Peroxisome proliferator–activated receptor γ coactivator 1β (PGC-1β) is a transcription factor implicated in the regulation of osteoclastogenesis in mouse models. In their study, Ma et al report an increased nuclear accumulation of PGC-1β in circulating CD14+ monocytes from RA patients, which correlates with the degree of joint destruction. These cells are precursors of osteoclasts and have the capability of osteoclastogenesis. PGC-1β acted as an upstream regulator of the transcription activator NFATc1, and the binding of PGC-1β to the NFATc1 DNA promoter led to transcriptional activation. The authors conclude that activation of the PGC-1β/NFATc1 pathway in circulating osteoclast precursors may contribute to local bone destruction in RA.

Activation of the Peroxisome Proliferator–Activated Receptor γ Coactivator 1β/NFATc1 Pathway in Circulating Osteoclast Precursors Associated With Bone Destruction in Rheumatoid ArthritisMa et al, Arthritis Rheumatol 2019;71:1252–1264

CORRESPONDENCE Lan Shao, PhD: shaolan@mail.sysu.edu.cnLie Dai, MD, PhD: dailie@mail.sysu.edu.cn

KEY POINTS

• Elevated nuclear PGC-1β in peripheral monocytes from RA patients correlates with joint destruction.

• PGC-1β promotes osteoclastogenesis of circulating osteoclast precursors.

• PGC-1β binds to the NFATc1 DNA promoter, leading to transcription al activation.

• The PGC-1β pathway could be a therapeutic target for diseases with osteoclast activation.

18

1207

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1207–1210DOI 10.1002/art.40875 © 2019, American College of Rheumatology

E D I T O R I A L

Evolving Use of Molecular Imaging in Research and in PracticeAhmed Tawakol,1 Sebastian Unizony,1 Michael T. Osborne,1 Elena Massarotti,2 and Jon T. Giles3

Combined molecular and structural imaging using positron emission tomography (PET) has long been clinically employed in the evaluation of cardiovascular and oncologic disorders. More recently, PET imaging has been accepted as an assessment tool in rheumatology, where it has proven value in the evaluation of vasculitis. Emerging research continues to expand the uses for molecular imaging in rheumatic diseases. Such imaging holds the potential to improve the identification of individuals at the highest risk for disease complications and those most likely to benefit from more aggressive treatments.

Among imaging approaches that target inflammation in humans, 18F- fluorodeoxyglucose (18F- FDG) PET with computed tomography (CT) has been the most extensively studied. 18F- FDG accumulates within tissues in proportion to its glycolytic rate (1). Inflammatory cells have particularly high rates of glycolysis, espe-cially after proinflammatory activation (1), and thus accumulate relatively larger amounts of 18F- FDG. The relationship between 18F- FDG uptake and tissue inflammation has been best evaluated in the context of atherosclerosis, which is itself a chronic inflamma-tory condition. In human studies, atherosclerotic 18F- FDG uptake measured using PET- CT imaging has repeatedly been shown to correlate closely with macrophage density on histopathologic evaluation (1). Thus, 18F- FDG–PET- CT provides a noninvasive index of tissue inflammation.

The ability of 18F- FDG–PET- CT to characterize tissue inflam-mation has been leveraged in a wide range of inflammatory cardiovascular conditions (Figure 1). In the evaluation of athero-sclerotic disease, arterial 18F- FDG uptake has been shown to independently predict incident cardiovascular disease (CVD) events beyond traditional risk factors (1). Since therapies that improve clinical CVD outcomes also tend to attenuate the PET imaging signal, 18F- FDG–PET- CT has been repeatedly used to test novel therapies that target atherosclerosis (1). In the evaluation of cardiac sarcoidosis (a condition characterized by macrophage

infiltration of the myocardium), 18F- FDG–PET- CT can be used to detect cardiac involvement with high sensitivity (2). Further, height-ened myocardial 18F- FDG uptake has been shown to predict a higher risk for cardiac events in patients with cardiac sarcoidosis and can be used to evaluate the efficacy of therapies targeting this disease process (3). Additionally, 18F- FDG–PET- CT has been proven to be clinically useful in detecting infections of intracardiac prosthetic materials, such as prosthetic valves, cardiac implant-able electronic devices, and left ventricular assist devices (4).

18F- FDG–PET- CT has also been used with increasing fre-quency in rheumatic conditions and has been demonstrated to be clinically useful in the evaluation of large vessel vasculitis. 18F- FDG–PET- CT provides both high sensitivity and high specificity in the detection of arterial wall inflammation and, along with biopsy and other imaging modalities, has become more frequently used in the evaluation of giant cell arteritis (GCA) and Takayasu arteritis (5). Additionally, 18F- FDG–PET- CT may have a role in the longitu-dinal assessment of disease activity and may be useful in the pre-diction of long- term arterial complications (e.g., aortic aneurysm) (6). In recognition of the utility of PET- CT, a European League Against Rheumatism working group has recommended early 18F- FDG–PET- CT imaging in individuals with suspected GCA (7).

While 18F- FDG–PET- CT imaging has been less extensively studied in rheumatoid arthritis (RA), emerging research sug-gests that it may have a potential role in the evaluation of the dis-ease. Inflamed synovium in RA is characterized by considerable immune cell infiltration, proliferation of resident synoviocytes, and neovascularization (8). These synovial immune cells have height-ened glycolytic metabolism as a result of their immune activation (9). Perhaps unsurprisingly, relatively high synovial 18F- FDG uptake has been frequently observed among patients with RA who have undergone 18F- FDG–PET- CT for oncologic indications. Further, multiple studies have demonstrated that in individuals with RA, synovial 18F- FDG uptake is associated with severity of joint inflam-

1Ahmed Tawakol, MD, Sebastian Unizony, MD, Michael T. Osborne, MD: Massachusetts General Hospital and Harvard Medical School, Boston; 2Elena Massarotti, MD: Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts; 3Jon T. Giles, MD, MPH: Columbia University, New York, New York.

Dr. Tawakol has received consulting fees, speaking fees, and/or honoraria from Actelion (less than $10,000) and research support for Massachusetts

General Hospital from Genentech and Actelion. No other disclosures relevant to this article were reported.

Address correspondence to Ahmed Tawakol, MD, Cardiology Division, Massachusetts General Hospital and Harvard Medical School, 55 Fruit Street, Yawkey 5050, Boston, MA 02114. E-mail: atawakol@mgh.harvard.edu.

Submitted for publication February 7, 2019; accepted in revised form February 28, 2019.

TAWAKOL ET AL 1208       |

mation and predicts drug therapy response and subsequent chronic joint damage (10,11).

While there has been growing interest in advancing the use of molecular imaging in RA, significant limitations have curtailed its clinical application. Importantly, previous studies have used a broad range of 18F- FDG end points to characterize joint activ-ity and there are limited data to compare the various 18F- FDG–PET- CT synovial measures with clinical measures of disease activity, thus limiting standardization of the technique. Accordingly, key questions remain regarding the use of 18F- FDG–PET- CT syno-vial imaging, including how best to measure synovial activity with PET- CT, and how much those measures add to currently utilized clinical parameters.

In this issue of Arthritis & Rheumatology, Lee et al (12) provide important insights that help address these questions. Lee and colleagues evaluated the relationship between 18F- FDG–PET- CT–derived measures of joint activity and clinical assessment of joint disease activity in 91 individuals with RA. The study consisted of 2 groups: a development cohort (n = 69) and a validation cohort (n = 22). The number of PET- positive joints (of 28- or 68- joint counts) significantly correlated with swollen and tender joint counts as well as with the Disease Activity Score in 28 joints using the eryth-rocyte sedimentation rate (DAS28- ESR) (13). Using multivariable analyses that included the ESR, the patient’s global assessment of disease activity (PtGA), and PET- CT–derived parameters, Lee et al

developed a PET- based disease activity score (PET/DAS). When applied to the validation cohort using Pearson’s correlation coef-ficient (0.843), the PET/DAS highly correlated with DAS28- ESR scores (P < 0.001). Based on these observations, the authors pro-posed that PET- CT could serve as a sensitive and reliable method to evaluate joint inflammation. This study adds to the growing number of studies that have evaluated 18F- FDG–PET- CT imaging of joints. Importantly, it provides data that 1) identify the quantita-tive 18F- FDG–PET- CT measures that best correlate with concur-rent clinically determined disease activity and 2) demonstrate that the PET- CT joint measures are highly reproducible.

The limitations of the standard clinical joint counts used to assess disease activity are well recognized. Joint counts are lim-ited by an inherent lack of objectivity related to both operator factors (e.g., training, experience, perception of the patient’s pain level) and patient factors (e.g., body habitus, subjectivity of pain). Exclusion of the least reliable joints from the 28- joint count (i.e., the metatarsophalangeal [MTP] joints of the feet) has been demonstrated as a simplification that does not sacrifice utility across groups of RA patients (14). Yet, such simplifications may not apply equally across patients. For example, consider the dilemma illustrated when comparing a patient with 10 tender and swollen MTP joints (who receives a score of 0 using the tradi-tional 28 tender and swollen joint count) with a patient with scant nontender synovitis in the small joints of the hands (who receives

Figure 1. Clinical uses of 18F- fluorodeoxyglucose (18F- FDG) positron emission tomography (PET) with computed tomography (CT) inflammation imaging. In the clinical entities depicted (suspected prosthetic valve endocarditis, known or suspected cardiac sarcoidosis, suspected device infection, and aortitis/arterial inflammation), 18F- FDG–PET- CT has demonstrated utility and is routinely employed. Arrows indicate increased FDG uptake consistent with higher inflammation. A PET- CT device is also depicted. Reproduced, with permission, from refs. 20–22.

EDITORIAL |      1209

a score of 20). In this example, the higher joint count does not necessarily translate to greater disease activity. To complicate things further, there remains a concern that current joint counts are not sufficiently sensitive, as subclinical joint destruction con-tinues despite apparent clinical remission (as assessed with typical clinical parameters) (15,16). Hence, more sensitive and reproducible methods are needed to quantify disease activity in RA. Indeed, the 18F- FDG–PET- CT–derived method described in the report by Lee and colleagues provides such a measure.

In the future, the assessment of synovial inflammation may be further enhanced by combining PET functional measurements with measurements derived from structural imaging modalities (e.g., volumetric indices derived using simultaneously acquired CT or magnetic resonance imaging data). Such multimodality imaging may further enhance measurement of disease activity by providing data on both the burden and the intensity of synovitis. However, additional study, including assessments of cost- efficacy, is needed before these imaging techniques can be used in the routine clinical care of patients with RA.

On the other hand, 18F- FDG–PET- CT imaging of joint inflam-mation can currently be leveraged in clinical trials on RA, where surrogate end points of disease activity have the potential to improve the identification of effective therapies. Akin to a model that has been highly successful in oncology, results of phase II PET- CT imaging end point trials can be used to inform the selec-tion of treatments that should advance to phase III clinical end point trials. However, before relying on such a strategy, more data are needed to evaluate how well changes in the surrogate PET- CT imaging measures predict clinical efficacy, as has been done in oncologic and atherosclerotic 18F- FDG–PET- CT studies (1).

Additionally, 18F- FDG–PET- CT can be used to gain impor-tant pathobiologic insights into RA, including assessment of the biologic cross- talk that exists between organ systems impacted by RA. For example, previous 18F- FDG–PET- CT imaging stud-ies have shown an association between synovial inflammation and arterial inflammation (17). This may represent an important observation since patients with RA have a substantially increased risk of death attributed to underlying CVD (18,19). A particularly vexing question remains as to whether reducing clinically evident synovial disease activity (by treating to alleviate joint symptoms) is sufficient to attenuate atherosclerotic inflammation and its accompanying risks of CVD in RA. To this end, the Treatments Against RA and Effects on FDG- PET/CT (TARGET) trial (Clini-calTrials.gov identifier NCT02374021) is an ongoing National Institues of Health–sponsored, randomized, controlled, multi-center trial using 18F- FDG–PET- CT to compare the effects of 2 treatment regimens on arterial inflammation among RA patients with inadequate methotrexate (MTX) treatment response (the addition of a tumor necrosis factor inhibitor to background MTX versus the addition of sulfasalazine and hydroxychloroquine to background MTX [triple therapy]). Additionally, the TARGET trial provides a unique opportunity to prospectively evaluate several

synovial PET- CT parameters (both at baseline as well as change over time) to assess which best predict clinical response; the trial should also provide needed data to further assess the value of 18F- FDG–PET- CT imaging in RA.

Multimodal molecular imaging of inflammation has become an important tool in the assessment of several inflammatory conditions. In RA, inflammation imaging with 18F- FDG–PET- CT is being increasingly used to study novel treatment approaches and to develop pathobiologic insights. More studies, such as the one reported by Lee et al, are needed in order to better under-stand the potential role of this imaging modality in the assess-ment of individuals with RA.

AUTHOR CONTRIBUTIONSAll authors drafted the article, revised it critically for important intellec-

tual content, and approved the final version to be published.

REFERENCES 1. Joseph P, Tawakol A. Imaging atherosclerosis with positron emission

tomography. Eur Heart J 2016;37:2974–80.

2. Youssef G, Leung E, Mylonas I, Nery P, Williams K, Wisenberg G, et al. The use of 18F- FDG PET in the diagnosis of cardiac sarcoidosis: a systematic review and metaanalysis including the Ontario experience. J Nucl Med 2012;53:241–8.

3. Blankstein R, Osborne M, Naya M, Waller A, Kim CK, Murthy VL, et al. Cardiac positron emission tomography enhances prognostic assessments of patients with suspected cardiac sarcoidosis. J Am Coll Cardiol 2014;63:329–36.

4. Gomes A, Glaudemans A, Touw DJ, van Melle JP, Willems TP, Maass AH, et al. Diagnostic value of imaging in infective endocarditis: a systematic review. Lancet Infect Dis 2017;17:e1–14.

5. Soussan M, Nicolas P, Schramm C, Katsahian S, Pop G, Fain O, et al. Management of large- vessel vasculitis with FDG- PET: a systematic literature review and meta- analysis. Medicine (Baltimore) 2015;94:e622.

6. Danve A, O’Dell J. The role of 18F fluorodeoxyglucose positron emission tomography scanning in the diagnosis and management of systemic vasculitis. Int J Rheum Dis 2015;18:714–24.

7. Dejaco C, Ramiro S, Duftner C, Besson FL, Bley TA, Blockmans D, et al. EULAR recommendations for the use of imaging in large vessel vasculitis in clinical practice. Ann Rheum Dis 2018;77:636–43.

8. Harris ED. Rheumatoid arthritis: pathophysiology and implications for therapy. N Engl J Med 1990;322:1277–89.

9. Falconer J, Murphy AN, Young SP, Clark AR, Tiziani S, Guma M, et al. Synovial cell metabolism and chronic inflammation in rheumatoid arthritis. Arthritis Rheumatol 2018;70:984–99.

10. Suto T, Okamura K, Yonemoto Y, Okura C, Tsushima Y, Takagishi K. Prediction of large joint destruction in patients with rheumatoid arthritis using 18F- FDG PET/CT and Disease Activity Score. Medicine (Baltimore) 2016;95:e2841.

11. Roivainen A, Hautaniemi S, Möttönen T, Nuutila P, Oikonen V, Parkkola R, et al. Correlation of 18F- FDG PET/CT assessments with disease activity and markers of inflammation in patients with early rheumatoid arthritis following the initiation of combination therapy with triple oral antirheumatic drugs. Eur J Nucl Med Mol Imaging 2013;40:403–10.

12. Lee SJ, Jeong JH, Lee CH, Ahn BC, Eun JS, Kim NR, et al. Develop- ment and validation of an 18f- fluorodeoxyglucose–positron emission

TAWAKOL ET AL 1210       |

tomography with computed tomography–based tool for the evaluation of joint counts and disease activity in patients with rheumatoid arthritis. Arthritis Rheumatol 2019;70:1232–40.

13. Prevoo ML, van ‘t Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity scores that include twenty- eight–joint counts: development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995;38:44–8.

14. Prevoo ML, van Riel PL, van ‘t Hof MA, van Rijswijk MH, van Leeuwen MA, Kuper HH, et al. Validity and reliability of joint indices: a longitudinal study in patients with recent onset rheumatoid arthritis. Br J Rheumatol 1993;32:589–94.

15. Brown AK, Conaghan PG, Karim Z, Quinn MA, Ikeda K, Peterfy CG, et al. An explanation for the apparent dissociation between clinical remission and continued structural deterioration in rheumatoid arthritis. Arthritis Rheum 2008;58:2958–67.

16. McQueen FM, Stewart N, Crabbe J, Robinson E, Yeoman S, Tan PL, et al. Magnetic resonance imaging of the wrist in early rheumatoid arthritis reveals progression of erosions despite clinical improvement. Ann Rheum Dis 1999;58:156–63.

17. Emami H, Vijayakumar J, Subramanian S, Vucic E, Singh P, MacNabb MH, et al. Arterial 18F- FDG uptake in rheumatoid arthritis

correlates with synovial activity. JACC Cardiovasc Imaging 2014;7: 959–60.

18. Aviña-Zubieta JA, Choi HK, Sadatsafavi M, Etminan M, Esdaile JM, Lacaille D. Risk of cardiovascular mortality in patients with rheumatoid arthritis: a meta- analysis of observational studies. Arthritis Rheum 2008;59:1690–7.

19. Gabriel SE, Crowson CS, Kremers HM, Doran MF, Turesson C, O’Fallon WM, et al. Survival in rheumatoid arthritis: a population- based analysis of trends over 40 years. Arthritis Rheum 2003;48:54–8.

20. White JA, Rajchl M, Butler J, Thompson RT, Prato FS, Wisenberg G. Active cardiac sarcoidosis: first clinical experience of simultaneous positron emission tomography–magnetic resonance imaging for the diagnosis of cardiac disease. Circulation 2013;127:e639–41.

21. Saby L, Laas O, Habib G, Cammilleri S, Mancini J, Tessonnier L, et al. Positron emission tomography/computed tomography for diagnosis of prosthetic valve endocarditis: increased valvular 18F- fluorodeoxyglucose uptake as a novel major criterion. J Am Coll Cardiol 2013;61:2374–82.

22. Sarrazin JF, Philippon F, Tessier M, Guimond J, Molin F, Champagne J, et al. Usefulness of fluorine- 18 positron emission tomography/computed tomography for identification of cardiovascular implantable electronic device infections. J Am Coll Cardiol 2012;59:1616–25.

1211

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1211–1214DOI 10.1002/art.40871 © 2019, American College of Rheumatology

E D I T O R I A L

Picturing Giant Cell Arteritis: Projecting Into the FutureSara K. Tedeschi1 and Ayaz Aghayev2

Imaging now plays an important role in the evaluation of giant cell arteritis (GCA). A growing body of literature supports the use of ultrasound, computed tomography (CT), magnetic reso-nance imaging (MRI), and positron emission tomography (PET) to assist with diagnosis. The European League Against Rheuma-tism recently recommended ultrasound or high- resolution MRI as diagnostic tests for GCA involving the superficial cranial arteries; ultrasound, MRI, CT, and PET are all recommended for evaluating large arteries, with the caveat that ultrasound has limited value for detecting aortitis (1). For patients with GCA, imaging is also of great interest for monitoring disease activity and structural conse-quences, such as large artery aneurysm or stenosis. Imaging is particularly promising for flare assessment in GCA patients who develop recurrent clinical symptoms during treatment with bio-logic agents that suppress markers of inflammation, i.e., interleu-kin- 6 inhibitors.

In this issue of Arthritis & Rheumatology, Sammel et al pres-ent the findings from a double- blind study of time- of- flight (TOF) PET/CT of the head, neck, and chest for GCA diagnosis (2). A major limitation of conventional PET/CT has been the lack of spa-tial resolution sufficient for visualization of the small branches of the external carotid artery, such as the temporal artery. Although TOF PET/CT has been utilized for cancer imaging in the last de-cade, this is the first study to report using this technology in GCA. Improved PET/CT technology allows TOF reconstruction, which improves image quality, provides greater signal- to- noise ratio, and allows more accurate quantification of the counts while also reducing radiation exposure and scan time. Consequently, TOF PET/CT can be used to assess small lesions/areas in the whole body, including the small branches of the external carotid artery in the head and neck.

To place this study into context, we considered a hypothet-ical, ideal imaging test for GCA. It would be highly sensitive and specific for the initial diagnosis of GCA, and treatment with ste-roids would not alter the result. The imaging test would visualize all potentially affected arterial beds in the entire body, focusing on the head, neck, and thorax. Results would be rapid, and the modality would be widely available. The protocol for the technologist to

prepare the patient and to perform the study would be easy and reproducible. It would definitively differentiate active vasculitis from nonvasculitic conditions such as atherosclerosis and infectious arteritis. When GCA is not the diagnosis, it could reveal an alter-native explanation for the clinical presentation. Among patients with GCA, the test could be followed up over time to assess treatment response and/or recurrence of arteritis. It would distin-guish between active and treated vasculitis and monitor long- term sequelae such as aneurysm.

Existing imaging modalities, as summarized in Table  1, embody different combinations of these ideal features. Direct comparisons of their accuracy for identifying GCA must be made cautiously, as different reference standards have been used in various studies. Color Doppler ultrasound is highly sensitive and specific but cannot assess intrathoracic or intraabdominal large vessels, which are frequently involved in GCA (3). CT angiography (CTA) and MR angiography (MRA) have excellent resolution, but the wall thickening and enhancement revealed on these modal-ities cannot reliably differentiate active from inactive vasculitis (4,5). Conventional PET/CT is the most widely studied imaging modality for assessing disease activity in the large intrathoracic, intraabdominal, and proximal aortic branches. PET uses fluorine- 18- fluorodeoxyglucose (FDG), a radiotracer that accumulates in monocytes and other inflammatory cells and does not accumu-late in normal vessels. For patients with GCA, PET/CT holds more promise for disease monitoring compared to CTA and MRA, as changes in FDG avidity more reliably reflect changes in disease

activity (5–7).A major concern about the use of PET/CT for diagnosing and

monitoring GCA is the lack of a standardized scoring system for vascular uptake. PET/CT is generally considered positive for large vessel vasculitis when FDG uptake in the arterial wall exceeds uptake in the liver; however, this is a subjective determination (8). The ratio of the maximum standardized uptake value (SUVmax) in specific vessels (e.g., aorta/liver SUVmax ratio, aorta/superior vena cava [blood pool] SUVmax ratio) provides a semiquantitative mea-sure of FDG avidity. However, a threshold for the SUVmax ratio in vasculitis has not been validated (9).

1Sara K. Tedeschi, MD, MPH: Harvard Medical School, Brigham and Women’s Hospital, Boston, Massachusetts; 2Ayaz Aghayev, MD: Brigham and Women’s Hospital, Boston, Massachusetts.

No potential conflicts of interest relevant to this article were reported.

Address correspondence to Sara K. Tedeschi, MD, MPH, 60 Fenwood Road, Boston, MA 02115. E-mail: stedeschi1@bwh.harvard.edu.

Submitted for publication February 13, 2019; accepted in revised form February 26, 2019.

TEDESCHI AND AGHAYEV 1212       |

Another concern is that immunosuppressive treatment (e.g., glucocorticoids) compromises the diagnostic accuracy of PET/CT; recent data suggest that PET/CT sensitivity for large vessel GCA is preserved when performed within 3 days of the start of steroid treatment (10). FDG uptake in the vessel wall is abnormal, but it is not pathognomonic for vasculitis; atherosclerosis can also cause FDG uptake and can be difficult to differentiate from vascu-litis, particularly in the extremities. Additionally, mild- to- moderate FDG uptake on follow- up PET/CT can be attributable to vascular remodeling or persistent active vasculitis. Finally, the yield of PET/CT is highest if patients follow a low- carbohydrate, high- fat diet prior to the test, which may interfere with timely completion of the study.

In the present study by Sammel et al (2), the investigators performed TOF PET/CT to tackle image quality issues, and eval-uated the small branches of the external carotid arteries in the head and neck along with the aorta and its major branches. The investigators enrolled 64 patients within 72 hours of the initiation of steroid treatment for suspected GCA. Nearly all of the patients underwent clinically indicated temporal artery biopsy, which was

one of the reference standards. The treating clinician was blinded with regard to the PET/CT results, and therefore sensitivity and specificity could be determined using the treating clinician’s diag-nosis at 6 months as another reference standard, while avoiding circular reasoning in which the test result would inform the refer-ence standard. This novel application of TOF PET/CT was highly sensitive (92%) with good specificity (85%), using temporal artery biopsy as the reference. Sensitivity was lower (71%) and specificity higher (91%) when clinical diagnosis at 6 months was used as the reference. Radiographic aortitis was present in 42% of patients with a positive finding of GCA on temporal artery biopsy, similar to that in previous reports. Incidental findings were identified in 20% of subjects, some of which pointed to another explanation for the clinical presentation. Among the 6 patients who also had vascu-lar ultrasound performed, PET/CT and ultrasound were globally concordant in all 6 patients.

Because TOF PET/CT was able to visualize both the super-ficial cranial and intrathoracic arteries, investigators were able to observe that some patients had predominantly cranial involvement and others had predominantly large vessel involvement. Among

Table 1. Overview of existing imaging technologies for the diagnosis of giant cell arteritis

Imaging modalitySensitivity, %

(95% CI)Specificity, %

(95% CI) Advantages of technique Disadvantages of technique

Color Doppler ultrasound

77 (62–87)† 96 (85–99)† Widely available; good spatial resolution; no ionizing radiation; inexpensive; classic finding for acute arteritis (halo sign)

Intrathoracic and intraabdominal vessels are not captured; high interobserver variability; longer period of image acquisition

CTA 73 (45–92)‡ 78 (40–97)‡ Excellent spatial resolu-tion; rapid image acqui-sition; lumen patency assessment

Radiation exposure; does not visualize small branches of external carotid (e.g., temporal artery); iodine contrast agent issues (allergy and nephrotox-icity); difficulty differentiating active vs. inactive disease

MRI/MRA 73 (57–85)† 88 (81–92)† Excellent soft tissue resolu-tion/characterization; no ionizing radiation; lumen patency assessment; can visualize temporal arteries

Longer period of image acquisi-tion (claustrophobia); gado-linium contrast agent issues (allergy and nephrogenic sys-temic fibrosis in patients with severe renal disease); expen-sive; difficulty differentiating active vs. inactive disease

Conventional PET/CT 67–77§ 66–100§ Whole body/entire vascular bed visualization; greatest potential for long- term monitoring

Radiation exposure; not widely available; conventional PET/CT does not visualize small branches of external carotid (e.g., temporal artery); lack of standardization of scoring; preparation required (fasting); expensive

MRA = magnetic resonance angiography; PET/CT = positron emission tomography/computed tomography.† Data are the pooled sensitivity and specificity of ultrasound halo sign and high-resolution magnetic resonance imaging (MRI) of the head, respectively, using clinical diagnosis as the reference standard (12). ‡ Data on computed tomography angiography (CTA) were from 1 study, using clinical diagnosis as the reference standard (13). § Data represent the range from 2 studies, of which 1 used clinical diagnosis as the reference standard and 1 used temporal artery biopsy as the reference standard (13,14).

EDITORIAL |      1213

the 7 patients who had negative biopsy findings and positive PET/CT findings for GCA, 3 presented with cranial signs/symptoms and were diagnosed as having GCA by the treating physician. The distribution of abnormalities on PET/CT was not described, and it would be of interest to know if TOF PET/CT detected vasculitis in the small branches of the external carotid artery, and/or if these 3 patients had large vessel arteritis.

Several aspects of the study design warrant consideration when interpreting the results. Radiologists interpreted scans based on the overall impression without using semiquantitative measurements, which limits enthusiasm about its generalizabil-ity and reproducibility at other medical centers. Each image was interpreted by 2 readers, and the interreader reliability was only moderate (κ = 0.65). The authors state that the final consensus interpretation was more accurate than that achieved by either reader; however, the reference standard for the individual read-ers was not clearly reported. Having 2 nuclear medicine physi-cians jointly review a PET/CT is unlikely to be practical in a clinical setting. One patient with classic GCA symptoms and a temporal artery biopsy finding positive for GCA had a negative TOF PET/CT scan, again raising questions about the methods used for scan interpretation.

Comparing the performance of TOF PET/CT against other imaging modalities is challenging, due to the use of different reference standards, as the authors discuss. Not only does the reference standard matter, the patient population also mat-ters; the pretest probability of GCA will affect the positive and negative predictive values of the test. For 8 patients (12.5%), diagnostic consensus was not reached, and a panel of 4 rheu-matologists decided the diagnosis. Because the prevalence of GCA in the study sample affects the positive predictive value of TOF PET/CT, the panel’s decision about classification as GCA or not impacted the test performance. Nonetheless, we applaud the study team for grappling with these difficult cases that reflect the types of cases seen in routine rheumatology practice.

As PET/CT grows increasingly more common in the eval-uation of GCA, a scoring system based on semiquantitative measurements must be defined and validated. Semiquantita-tive measurements may prove a succinct way to interpret the presence or absence of vasculitis, facilitating both diagnosis and long- term monitoring. Specific vessel involvement and the percentage of vessel length involved might be weighted in such a system. A posttreatment radiotracer uptake thresh-old value is also needed to allow for longitudinal assessment of GCA. Along with establishing a scoring system, the termi-nology used to describe vasculitis in radiology reports should be reconsidered. Radiology reports that reference “acute” or “chronic” vasculitis have likely borrowed these terms from the pathology literature. In the absence of studies correlating biopsy results and imaging, we advocate reconsidering or explicitly avoiding the term “chronic vasculitis” in radiology

reports. Rheumatologists reading these reports are confronted with the question as to whether to initiate or resume treat-ment in an asymptomatic patient with “chronic vasculitis” (e.g., wall thickening but no fat stranding or FDG uptake to suggest active vasculitis).

Future studies comparing TOF PET/CT, high- resolution scalp MRI, and ultrasound for GCA diagnosis will inform clin-ical care. These modalities will allow us to better understand whether long- term outcomes differ between patients with cranial- predominant GCA and those with large artery–predom-inant GCA. Improvement in the PET scanning technologies will likely spur the development of new tracers and increase utili-zation of existing non- FDG tracers, such as 11C- PK11195. The tracer PK11195 binds to proteins that are highly expressed in activated monocytes and macrophages, and thus holds great promise for differentiating active from inactive vasculitis (11). Developing and implementing systems for standardized reading and reporting of PET/CT are important next steps to advance GCA research and improve the clinical care of patients with GCA.

ACKNOWLEDGMENT

We thank Dr. William P. Docken for his helpful comments and suggestions.

AUTHOR CONTRIBUTIONS

Both authors were involved in drafting the article, revising it critically for important intellectual content, and approving the final version to be published.

REFERENCES 1. Dejaco C, Ramiro S, Duftner C, Besson FL, Bley TA, Blockmans

D, et al. EULAR recommendations for the use of imaging in large vessel vasculitis in clinical practice. Ann Rheum Dis 2018;77:636–43.

2. Sammel AM, Hsiao E, Schembri G, Nguyen K, Brewer J, Schrieber L, et al. Diagnostic accuracy of positron emission tomography/computed tomography scans of the head, neck, and chest for giant cell arteritis: a prospective, double- blind, cross- sectional study. Arthritis Rheumatol 2019;71:1319–28.

3. Grayson PC, Maksimowicz-McKinnon K, Clark TM, Tomasson G, Cuthbertson D, Carette S, et al. Distribution of arterial lesions in Takayasu’s arteritis and giant cell arteritis. Ann Rheum Dis 2012;71:1329–34.

4. Olthof SC, Krumm P, Henes J, Nikolaou K, la Fougère C, Pfannenberg C, et al. Imaging giant cell arteritis and aortitis in contrast enhanced 18F- FDG PET/CT: which imaging score correlates best with laboratory inflammation markers? Eur J Radiol 2018;99:94–102.

5. Quinn KA, Ahlman MA, Malayeri AA, Marko J, Civelek AC, Rosenblum JS, et al. Comparison of magnetic resonance angiography and 18F- fluorodeoxyglucose positron emission tomography in large- vessel vasculitis. Ann Rheum Dis 2018;77:1165–71.

6. Prieto-González S, García-Martínez A, Tavera-Bahillo I, Hernández-Rodríguez J, Gutiérrez-Chacoff J, Alba MA, et al. Effect of

TEDESCHI AND AGHAYEV 1214       |

glucocorticoid treatment on computed tomography angiography detected large- vessel inflammation in giant- cell arteritis: a prospective, longitudinal study. Medicine (Baltimore) 2015;94:e486.

7. Grayson PC, Alehashemi S, Bagheri AA, Civelek AC, Cupps TR, Kaplan MJ, et al. 18F- fluorodeoxyglucose–positron emission tomography as an imaging biomarker in a prospective, longitudinal cohort of patients with large vessel vasculitis. Arthritis Rheumatol 2018;70:439–49.

8. Besson FL, Parienti JJ, Bienvenu B, Prior JO, Costo S, Bouvard G, et al. Diagnostic performance of 18F- fluorodeoxyglucose positron emission tomography in giant cell arteritis: a systematic review and meta- analysis. Eur J Nucl Med Mol Imaging 2011;38:1764–72.

9. Stellingwerff MD, Brouwer E, Lensen KJ, Rutgers A, Arends S, van der Geest KS, et al. Different scoring methods of FDG PET/CT in giant cell arteritis: need for standardization. Medicine (Baltimore) 2015;94:e1542.

10. Nielsen BD, Gormsen LC, Hansen IT, Keller KK, Therkildsen P, Hauge EM. Three days of high- dose glucocorticoid treatment attenuates large- vessel 18F- FDG uptake in large- vessel giant cell

arteritis but with a limited impact on diagnostic accuracy. Eur J Nucl Med Mol Imaging 2018;45:1119–28.

11. Pugliese F, Gaemperli O, Kinderlerer AR, Lamare F, Shalhoub J, Davies AH, et al. Imaging of vascular inflammation with [11C]- PK11195 and positron emission tomography/computed tomography angiography. J Am Coll Cardiol 2010;56:653–61.

12. Duftner C, Dejaco C, Sepriano A, Falzon L, Schmidt WA, Ramiro S. Imaging in diagnosis, outcome prediction and monitoring of large vessel vasculitis: a systematic literature review and meta- analysis informing the EULAR recommendations. RMD Open 2018;4:e000612.

13. Lariviere D, Benali K, Coustet B, Pasi N, Hyafil F, Klein I, et al. Positron emission tomography and computed tomography angiography for the diagnosis of giant cell arteritis: a real- life prospective study. Medicine (Baltimore) 2016;95:e4146.

14. Blockmans D, Stroobants S, Maes A, Mortelmans L. Positron emission tomography in giant cell arteritis and polymyalgia rheumatica: evidence for inflammation of the aortic arch. Am J Med 2000;108:246–9.

1215

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1215–1216DOI 10.1002/art.40916 © 2019, American College of Rheumatology

I N M E M O R I A M

Shaun Ruddy, MD, 1935–2019Shaun Ruddy passed away on April 3, 2019 at the age of 84.

He grew up in Connecticut and received his undergraduate degree from Yale University and his medical degree (cum laude) from Yale University School of Medicine. He pursued his internal medicine training at the Peter Bent Brigham Hospital in Boston. Shaun served as a medical epidemiologist and commissioned officer in the Hepatitis Surveillance Unit with the Centers for Disease Con-trol. He was a rheumatology research fellow at the Robert Brigham Hospital in Boston, in the laboratory of Dr. K. Frank Austen. Upon completion of his fellowship he joined the faculty of Harvard Med-ical School, where he was an Associate Professor of Medicine. His major research work was in the area of complement, and he published several seminal papers on the role of complement in rheumatoid arthritis and other rheumatic diseases. In 1974 he was recruited to the Medical College of Virginia (Virginia Common-wealth University) to serve as Chairman of the Division of Rheu-matology, Immunology and Allergy (1974–1993). He also served as acting Chairman of the Department of Internal Medicine. He was the Elam C. Toone Professor of Internal Medicine, Microbiol-ogy and Immunology for more than 40 years. During his tenure as Chairman he was able, with his clinical colleagues, to secure fund-ing for more than 5 endowed professorships. He was a scholarly academician and a skilled teacher, mentor, and clinician.

Beginning in 1979, with the first edition, Shaun was co- editor of the Textbook of Rheumatology, along with Bill Kelley, Ted Harris, and Clem Sledge. His in- depth knowledge across a broad range of topics was especially impressive to the other co- editors.

He served as a member of the Advisory Council of the National Institute of Arthritis and Musculoskeletal Diseases, NIH and on the Board of Directors of the American Board of Internal Medicine, chairing the Rheumatology Subspecialty Section. He was an active member of the American College of Rheumatology, serving on multiple committees, chairing the Council on Research, and serving in various officer roles. In 1994 he was named Presi-dent of the ACR.

It was during his tenure as President that he encouraged the ACR to study the question of whether silicone breast implants were associated with the development of rheumatic diseases. At a time when many members of the ACR did not believe the College should get involved in this controversial area, Shaun thought this was a critical topic of study for the ACR. His view was that the ACR had a responsibility to evaluate the evidence and speak out on contentious issues. After a very careful review of the literature the ACR published a position statement that

there was no association of silicone implants with the develop-ment of defined systemic rheumatic diseases.

Shaun’s presidential address in 1995 was entitled “The American College of Rheumatology as a ‘Professional’ Society: An Oxymoron?” He spoke about the importance of profession-alism, altruism, accountability, excellence, duty, honor and integ-rity, and respect for others. For those of us who were fortunate to know Shaun those words reflected not just his professional life, but his personal code of conduct. He felt strongly that these attributes should guide the ACR and that without them we would become a trade union rather than remain a professional society.

He was an outstanding clinician, educator, researcher, and mentor—a true academician. He received multiple awards, includ-ing the ACR Bunim Medal, the ACR Gold Medal, and Master of the ACR.

Shaun was most happy on the island of Nantucket sailing, fishing, and boating. His love of the water was passed on to his children and grandchildren. Some of my (MEW) fondest memories of Shaun are of sailing around Nantucket listening to his maritime stories. Likewise, one of my (WNK) most memorable weeks with

IN MEMORIAM 1216       |

Shaun was our sail with Millie and Lois for a week in the Caribbean among the Leeward Islands of St. Bart’s and St. Martin. Captain Ruddy was fearless.

He was a lover of dogs, especially goldens, and was a voracious reader of books and the New York Times. He was a lifelong lover of music and an accomplished jazz clarinetist and saxophonist. He was proud of his Irish heritage and was a loyal member of the Irish- American Rheumatology Group that met annually at the national meetings.

He was married to his wife Millicent for more than 60 years. Millie was an active participant in all of Shaun’s rheumatology activities and has been a great friend to many rheumatologists

and their spouses. Shaun is also survived by his 2 daughters Candace and Christi and 3 grandchildren.

Shaun will be best remembered as a wonderful mentor, advisor, colleague, and friend. He will be greatly missed.

Michael E. Weinblatt, MDBrigham and Women’s Hospital Harvard Medical School Boston, MAWilliam N. Kelley, MDRaymond and Ruth Perelman School of Medicine University of Pennsylvania Philadelphia, PA

1217

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1217–1224DOI 10.1002/art.40858 © 2019, American College of Rheumatology

Investigating Asthma, Allergic Disease, Passive Smoke Exposure, and Risk of Rheumatoid ArthritisVanessa L. Kronzer,1 Cynthia S. Crowson,1 Jeffrey A. Sparks,2 Robert Vassallo,1 and John M. Davis III1

Objective. Rheumatoid arthritis (RA) is postulated to originate at mucosal surfaces, particularly the airway muco-sa. To investigate this hypothesis, we determined the association between RA and asthma, passive smoke exposure, and age at start of smoking.

Methods. For this case–control study, we identified 1,023 cases of RA (175 incident) within a single- center biobank population, using a rules- based algorithm that combined self- report with 2 diagnostic codes. Exposures were self- reported on biobank questionnaires. Logistic regression models were used to calculate the association of exposures with RA, adjusting for potential confounders. Odds ratios (ORs) with 95% confidence intervals (95% CIs) were calculated.

Results. After adjustment for allergies, urban environment, and passive smoke exposure, asthma was found to be associated with RA in the full cohort (OR 1.28 [95% CI 1.04–1.58; P = 0.02]) but not the incident RA cohort (OR 1.17 [95% CI 0.66–2.06; P = 0.60]). History of allergic disease was associated with RA in both the full cohort (OR 1.30 [95% CI 1.12–1.51; P < 0.001]) and the incident RA cohort (OR 1.61 [95% CI 1.11–2.33; P = 0.01]), especially food allergy, which was significantly associated with RA in the full cohort (OR 1.38 [95% CI 1.08–1.75; P = 0.01]) and showed a trend toward significance in the incident RA cohort (OR 1.83 [95% CI 0.97–3.45; P = 0.06]). Passive smoke exposure at home or work was not associated with RA. Finally, age at start of smoking was not associated with increased odds of developing RA in either the full cohort (OR 1.03 [95% CI 1.00–1.06; P = 0.03]) or the incident RA cohort (OR 1.00 [95% CI 0.92–1.08; P = 0.98]).

Conclusion. Asthma and allergies may be associated with increased risk of RA. Passive smoke exposure and early age at start of smoking do not appear to influence risk of RA.

INTRODUCTION

Rheumatoid arthritis (RA) is one of the most common autoim-mune diseases, affecting nearly 1 in 100 individuals (1). Seroposi-tive RA in particular is hypothesized to originate from inflammation in the respiratory tract, resulting in autoantibody formation that later leads to disease (2,3). However, several questions about this hypothesis of disease generation remain, which, if answered, may help elucidate disease pathogenesis.

First, several studies have shown an association between asthma and RA, which may be explained by a shared immuno-logic mechanism (4–8). However, a major limitation of these pre-vious studies is the lack of adjustment for allergic disease. More-over, none included adjustment for secondhand cigarette smoke or urban pollution, which are known contributors to RA (9–12)

and asthma (13,14). It is unclear whether asthma is associated with RA after adjustment for these important confounders.

Second, the association between personal smoking and RA has been well- established (15,16), yet only 3 studies have investi-gated the association between passive smoke exposure and RA (10,17,18). The results of these studies conflict, with 2 suggesting an association between passive smoking and RA and the other showing no association. None of these prior reports characterized workplace smoke exposure alone or contained information about both the duration and intensity to allow a dose- response analysis. Earlier age of smoking may also be important, but has not yet been studied in patients with RA (19).

Our goal in this study was to clarify these gaps in knowl-edge related to the oral- respiratory factors that may mediate RA pathogenesis. Specifically, we aimed to determine the association

Supported by a Rheumatology Research Foundation Resident Research Preceptorship.

1Vanessa L. Kronzer, MD, MSCI, Cynthia S. Crowson, PhD, Robert Vassallo, MD, John M. Davis III, MD, MS: Mayo Clinic, Rochester, Minnesota; 2Jeffrey A. Sparks, MD, MMSc: Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts.

No potential conflicts of interest relevant to this article were reported.Address correspondence to Vanessa L. Kronzer, MD, MSCI, Mayo Clinic,

200 First Street SW, Rochester, MN 55905. E-mail: kronzer.vanessa@mayo.edu.

Submitted for publication September 11, 2018; accepted in revised form February 7, 2019.

KRONZER ET AL 1218       |

of RA with asthma after controlling for allergic disease, urban environment, passive smoke (exposure at home and work), and age at start of smoking. We hypothesized that the association with asthma would be attenuated after adjustment for allergy and envi-ronmental pollutants, that passive smoke at higher doses would be associated with RA, and that earlier age at start of smoking would be associated with an increased risk of RA.

PATIENTS AND METHODS

Study design. This case–control study was performed according to the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) guidelines for observational studies (20). The study was approved by the Mayo Clinic and Olmsted County Institutional Review Boards (approval nos. 17- 010806 and 060- OMC- 17), and was in compliance with the Declaration of Helsinki.

Questionnaire data for this study came from the Mayo Clinic Biobank repository (21). Active recruitment took place from April 2009 to December 2015 in both Minnesota and Florida. Eligibility criteria included age of 18 years or older, ability to communicate in English, capacity to consent, and residence in the US. Approx-imately 29% of those invited agreed to participate and completed a baseline questionnaire, yielding the 55,898 current participants. Of those, 77% completed the follow- up questionnaire sent ~4 years later. To ensure the validity and quality of data, all ques-tionnaires were visually examined for errors and omissions. Those with >10 errors were returned to participants for correction. A computer program was also used to find logical errors and incor-rect skip patterns, flagging records for manual verification. Of the 55,898 biobank participants, ~87% were recruited from primary care locations, including internal medicine, preventive medicine, family medicine, and obstetrics/gynecology. Most of the remaining participants were recruited from orthopedics.

Study population. In this study, we defined cases of RA using a rules- based algorithm that combined self- reported RA on either the baseline or follow- up questionnaire with at least 2 RA diagnostic codes (714.0 or 714.9) 30 days apart. We excluded 38 individuals who reported RA on the baseline questionnaire but not on the follow- up questionnaire, as those individuals are likely to have misrepresented RA status. Using a separate cohort of 732 individuals screened for inclusion in the RA cohort of the Roches-ter Epidemiology Project (1), which used the 1987 American Col-lege of Rheumatology classification criteria for RA (22), we found the positive predictive value (PPV) of this definition to be 88%. This accuracy was superior to that obtained using diagnostic codes alone, which is a widely accepted method.

We also included 78 participants who did not self- report RA but had 3 or more diagnostic codes for RA, with the first code occurring at least 90 days after the most recent survey (presumed incident RA). The incident RA group also included 97 individuals

who self- reported RA on the follow- up questionnaire but not the baseline questionnaire, for a total of 175 incident cases.

To determine the accuracy of our RA definition in the full cohort, we performed manual chart verification using the 2010 ACR criteria in a subset of 100 randomly selected RA cases (23). One of the authors (VLK) performed the main chart review, and another author (JMD) adjudicated any cases with uncertainty. Of the 100 patients selected, 87 were found to have RA or seronegative inflammatory arthritis consistent with RA, for a PPV of 87% (Supplementary Table 1, available on Arthritis & Rheumatology web site at http://onlineli-brary.wiley.com/doi/10.1002/art.40858/abstract).

We defined the index date as the date of RA diagnosis. The date of RA diagnosis came from the earliest diagnostic code for RA, provided this date was in the age range of RA diagnosis reported by the patient (≤19, 20–49, 50–64, 65–79, or 80+ years), or earlier. If the first diagnostic code occurred after the patient- reported date of RA diagnosis, then we used the patient- reported date of RA diagnosis.

Controls for this study included biobank participants without self- reported RA and no diagnostic codes for RA to ensure none had RA (negative predictive value >99%). We used 3:1 match-ing for the remaining biobank participants, with matching criteria consisting of recruitment year (within 5 years), recruitment location (Minnesota or Florida), distance from recruitment location (within 500 miles), age at baseline survey (within 5 years), and sex. The index date for each control was defined as the date of RA diagno-sis for the corresponding case.

Measures. The prespecified primary exposures of interest included self- reported asthma, presence of passive smoke expo-sure at home and work, and age at start of smoking regularly. For passive smoke exposure, we also studied the age the smoke exposure started, years of exposure before index date, packs per day, and pack- years before index date. Home and workplace smoke exposure in packs per day was only available in the second version of the baseline questionnaire. Self- report is the most accu-rate way to study smoking history, and studies suggest that self- reported asthma has a high PPV as well (24,25). We performed a manual verification of self- reported asthma on a random subset of 50 study participants. We used previously published criteria to identify “definite” asthma as a physician diagnosis of asthma during a clinical visit in the chart and/or the presence of each of the following 3 conditions: 1) history of cough with wheezing or dyspnea, 2) variability in symptoms, and 3) ≥2 features supportive of the diagnosis, including sleep disturbance, nonsmoker, nasal polyps, blood eosinophilia, elevated IgE level, history of wheezing on exposure to antigen, pulmonary function test results indicative of asthma, and/or favorable clinical response to bronchodilator (26). We defined “probable” asthma as having the first 2 condi-tions but not the third. Using these criteria, we found 82% of these 50 participants with self- reported asthma to have definite asthma and 8% to have probable asthma (Supplementary Table 2, http://onlinelibrary.wiley.com/doi/10.1002/art.40858/abstract).

ORAL-RESPIRATORY EXPOSURES AND RA |      1219

History of allergy was another exposure of interest in this study. The questionnaire included the query, “Do you have any allergies?” to which participants responded yes or no. If yes,

participants were prompted to mark all allergies that apply: “food allergies such as shellfish or nuts,” “grasses, pollen, or dust,” “pets,” “insect stings or bites,” and/or “other.”

Table 1. Characteristics of the RA cases and controls at index date of diagnosis*

RA cases (n = 1,023)

Incident RA subset (n = 175)

Controls (n = 3,061)

Age, mean ± SD years 50 ± 16 63 ± 14 50 ± 16Female sex, no. (%) 741 (72) 119 (68) 2215 (72)Body mass index, mean ± SD kg/m2† 30 ± 7 30 ± 6 29 ± 6White, non- Hispanic race, no. (%) 1,000 (98) 172 (98) 2,982 (97)Education, no. (%)

Less than high school 21 (2) 1 (0.6) 61 (2)High school degree 181 (18) 24 (14) 496 (16)Technical school 111 (11) 20 (11) 272 (9)Some college 281 (28) 46 (26) 722 (24)Bachelor’s degree 213 (21) 37 (21) 732 (24)Graduate degree 201 (20) 43 (25) 736 (24)Other 15 (2) 4 (2) 42 (1)

Personal smokingNever, no (%) 532 (52) 91 (52) 1,796 (59)Past, no. (%) 431 (42) 77 (44) 1,105 (36)Current, no. (%)† 60 (6) 7 (4) 160 (5)Duration of smoking, years 22 (13–31) 26 (12–23) 18 (10–28)Packs per day 0.8 (0.5–1.5) 0.8 (0.5–0.8) 0.8 (0.5–0.8)Pack- years 15 (8–29) 19 (10–35) 13 (6–24)Age started smoking regularly, years 18 (16–19) 18 (16–20) 18 (16–20)

Asthma, no. (%)† 169 (17) 21 (13) 383 (13)Urban environment, no. (%) 741 (73) 124 (72) 2260 (74)Any allergy, no. (%)† 643 (64) 112 (65) 1739 (58)

Food 106 (11) 17 (10) 241 (8)Grass, pollen, or dust 323 (32) 56 (33) 932 (31)Pets 125 (12) 19 (11) 360 (12)Insects 76 (8) 11 (6) 228 (8)Other 380 (38) 67 (39) 957 (32)

Home smoke exposureYes, no. (%) 665 (65) 104 (60) 1890 (62)Age home smoke exposure started, years 0 (0–15) 0 (0–15) 0 (0–15)Duration, years 18 (12–21) 20 (16–29) 18 (10–20)Packs per day 0.8 (0.5–1.5) 0.8 (0.5–1.5) 0.8 (0.5–1.5)Pack- years 14 (8–27) 17 (9–29) 12 (8–26)

Workplace smoke exposureYes, no. (%) 394 (39) 62 (35) 1105 (36)Age workplace smoke exposure started, years 25 (17–25) 25 (17–25) 25 (17–25)Duration, years 10 (5–20) 10 (5–20) 10 (5–19)Packs per day 0.8 (0.5–1.5) 0.8 (0.5–1.5) 0.8 (0.5–1.5)Pack- years 8 (4–15) 10 (4–20) 8 (3–15)

Dental visit in previous year, no. (%)† 863 (85) 154 (88) 2646 (87)

* Except where indicated otherwise, values are the median (interquartile range). RA = rheumatoid arthritis. † Values at time of baseline questionnaire rather than index date.

KRONZER ET AL 1220       |

Secondary exposures of interest included history of personal smoking (yes/no), current smoking (yes/no), smoking duration before index date (years), packs per day, and pack- years before index date. Data for all exposures and confounders of interest came from the baseline and follow- up questionnaires. Age, sex, body mass index (BMI), and race were confirmed by the Mayo Clinic medical records, leading to complete data for these fields (27). Obesity was defined as BMI ≥30 kg/m2.

Statistical analysis. Chi- square tests were used to com-pare proportions, and Wilcoxon’s rank sum tests were used to compare continuous variables. To assess the relationship between each exposure and case/control status, unconditional logistic regression models were used. Sensitivity analyses were performed among the 175 cases of incident RA to allow for interpretation of results with clear timing of exposures before disease and without any possibility of recall bias. An additional sensitivity analysis was performed in the subset of 189 participants with known serologic status, of whom 111 were positive for rheumatoid factor (RF) and/or antibodies to cyclic citrullinated peptide (CCP).

Of the 4,084 study participants, 322 (7.8%) had missing data for at least 1 of the key questionnaire items, including history of asthma, allergy, and personal, home, or workplace smoking expo-sure. Because of the low frequency of missing data and even lower missingness among each of the individual models, the impact of missing data on the main study results was deemed low. Partic-ipants with missing data were excluded from the model pertain-ing to that exposure. All analyses were prespecified in a protocol unless described as “post hoc.” We performed analyses using SAS version 9.4 (SAS Institute), with a 2- sided alpha level of 0.05 as the threshold for significance and 95% confidence intervals (95% CIs).

RESULTS

This study included 1,023 cases of RA from the 55,898 biobank participants. The mean ± SD age at index date of RA diagnosis for both cases and controls was 50 ± 16 years, while the mean ± SD age at the time of the baseline survey was 62 ± 13 years. In unadjusted analyses, the RA group included a higher proportion of individuals with a history of personal smoking and allergies, along with increased home smoke exposure pack- years compared to controls (Table 1). Characteristics of the incident RA subgroup were similar to those of the full RA cohort, except for slightly longer dura-tion of smoking before RA and lower frequency of workplace

smoke exposure and asthma (Table 1).In the full study cohort in the unadjusted analysis, asthma

was significantly associated with RA (odds ratio [OR] 1.38 [95% CI 1.14–1.68; P < 0.001]). This difference remained statisti-cally significant even after controlling for home and workplace smoke exposure, urban environment, and presence of allergies (OR 1.28 [95% CI 1.04–1.58; P = 0.02]). However, it was not

statistically significant in the incident RA subset (OR 1.17 [95%

CI 0.66–2.06; P = 0.60]) (Table 2).History of any allergy type was also found to be associated

with RA in the full study cohort (OR 1.30 [95% CI 1.12–1.51; P < 0.001]). Further analyses examined the association of RA among different allergy types. Food allergies in particular were most highly associated with RA (OR 1.38 [95% CI 1.08–1.75; P = 0.01]) (Figure 1). The association with allergy was also sig-nificant in the incident RA subset (OR 1.61 [95% CI 1.11–2.33; P = 0.01]). In particular, a history of food allergy showed a trend toward an increase in the odds of developing RA in the incident RA subset (OR 1.83 [95% CI 0.97–3.45; P = 0.06]).

There was no evidence of an association between home or workplace smoke exposure and RA in either the full cohort or the incident RA subset, except for years of home smoke exposure (Table 3). When home and workplace smoke expo-sure were combined in a post hoc analysis, participants with the highest levels of cumulative passive smoke exposure had

Table 2. Association between asthma and RA*

OR (95% CI)

All RA cases (n = 1,023)

Incident RA subset

(n = 175)

Unadjusted 1.38 (1.14–1.68) 1.28 (0.75–2.20)Standard

adjustment†1.37 (1.12–1.67)‡ 1.31 (0.74–2.28)

Full adjustment§ 1.28 (1.04–1.58)‡ 1.17 (0.66–2.06)

* RA = rheumatoid arthritis; OR = odds ratio; 95% CI = 95% confi-dence interval. † Adjusted for age, sex, body mass index, race, education, and per-sonal smoking (never/past/current). ‡ P < 0.05. § Adjusted for allergic disease, urban environment, and home/workplace smoke exposure, in addition to the factors listed above.

Figure  1. Adjusted association between allergy types and rheumatoid arthritis (RA). Each data point represents a separate model adjusted for age, sex, body mass index, race, education, and personal smoking (never/past/current). Referent is patients without the particular allergy. 95% CI = 95% confidence interval.

ORAL-RESPIRATORY EXPOSURES AND RA |      1221

increased odds of developing RA in the full cohort (OR 1.37 [95% CI 1.02–1.84]), but not in the incident RA subset (OR 1.34 [95% CI 0.68–2.67]) (Table 3). To determine whether the effect of passive smoke exposure may be higher among nonsmokers, we performed an additional analysis examining the interaction between history of personal smoking (yes or no) and pack- years of home, workplace, and combined home/workplace smoke exposure. There was no evidence that passive smoking

had a different effect among nonsmokers compared to smok-ers (P for interaction = 0.29, 0.65, and 0.67, respectively).

Personal smoking, and its duration, intensity, and pack- years, were all significant predictors of RA in the full cohort but not in the incident RA subset (Table 4). Age at start of smoking was not associated with an increase in the odds of developing RA in either the full cohort (OR 1.03 [95% CI 1.00–1.06]) or the

incident RA cohort (OR 1.00 [95% CI 0.92–1.08]).

Table 3. Adjusted association between passive smoke exposure and RA*

Passive smoke type

Multivariable OR (95% CI)

All RA cases (n = 1,023)

Incident RA subset (n = 175)

Home smoke exposureYes 1.06 (0.91–1.23) 0.81 (0.56–1.18)Age home exposure started† 1.00 (0.99–1.01) 0.99 (0.96–1.01)Duration, years‡ 1.09 (0.99–1.21) 1.37 (1.08–1.72)§Packs per day 1.12 (0.94–1.01) 1.20 (0.74–1.94)Pack- years‡ 1.03 (0.97–1.09) 1.14 (0.99–1.32)

None (referent) 1.00 1.001–9 0.97 (0.78–1.21) 0.63 (0.35–1.17)10–19 1.13 (0.88–1.45) 0.83 (0.43–1.62)20–29 1.27 (0.94–1.71) 1.41 (0.68–2.92)30–39 1.13 (0.71–1.78) 1.07 (0.35–3.34)40+ 1.14 (0.80–1.63) 1.09 (0.47–2.51)

Workplace smoke exposureYes 1.01 (0.86–1.17) 1.04 (0.71–1.52)Age workplace exposure started† 1.01 (0.99–1.03) 1.03 (0.97–1.07)Duration, years‡ 1.01 (0.85–1.19) 1.05 (0.66–1.67)Packs per day 1.15 (0.97–1.36) 1.20 (0.74–1.93)Pack- years‡ 1.04 (1.00–1.09) 1.12 (0.83–1.49)

None (referent) 1.00 1.001–9 1.01 (0.81–1.26) 0.94 (0.51–1.74)10–19 0.88 (0.61–1.25) 0.92 (0.38–2.21)20–29 0.84 (0.46–1.52) 1.12 (0.21–6.09)30–39 1.80 (1.01–3.19)§ 2.36 (0.56–9.94)40+ 1.17 (0.64–2.13) 3.55 (0.64–19.8)

Combined home/workplace smoke exposurePack- years‡ 1.04 (1.00–1.09) 1.15 (1.02–1.29)

None (referent) 1.00 1.001–9 0.93 (0.74–1.17) 0.65 (0.35–1.21)10–19 1.08 (0.84–1.39) 0.63 (0.31–1.26)20–29 1.07 (0.79–1.45) 0.96 (0.46–2.01)30–39 1.13 (0.76–1.67) 1.07 (0.31–3.76)40+ 1.37 (1.02–1.84)§ 1.34 (0.68–2.67)

* Each variable was assessed in a separate model with adjustment for age, sex, body mass index, race, education, and personal smoking (never/past/current). RA = rheumatoid arthritis; 95% CI = 95% confidence interval. † Adjusted for pack- years of passive smoke exposure, in addition to the factors listed above. ‡ Odds ratios (ORs) reported per 10 units. § P < 0.05.

KRONZER ET AL 1222       |

Finally, a sensitivity analysis was performed among the subset of patients known to have elevated RF and/or CCP antibodies. Although the group positive for antibodies was small, associations were similar to the main results of the study (see Supplementary Table 3, http://onlinelibrary.wiley.com/doi/10.1002/art.40858/abstract). However, there was a sugges-tion of a stronger association with grass allergy and home smoke exposure (yes or no). Characteristics of patients with missing data are shown in Supplementary Table 4 (http://onlinelibrary.wiley.com/doi/10.1002/art.40858/abstract).

DISCUSSION

This study addresses several gaps related to the oral- respiratory hypothesis of RA pathogenesis. Namely, the results support the concept of a marginal association between asthma and RA, identify a connection between allergy and RA, and pro-vide evidence against the notion that passive smoke or earlier age of smoking predisposes individuals to RA.

First, in the full cohort, asthma was associated with RA even after adjustment for allergy and pollutants, which is an associa-tion not previously shown. The association between asthma and RA is potentially biologically plausible. Both are immunologic disorders profoundly influenced by environmental factors that induce oxidative stress such as cigarette smoke, wood smoke, and environmental pollution (9–14). Moreover, there are many potential mechanisms that may mediate both disorders, includ-ing Th17 inflammation, infectious triggers, premature immune senescence, or other inflammation mediators, such as tumor necrosis factor and leukotrienes (28). However, the strength of association was attenuated after adjustment for newly consid-ered confounders, and the relationship was not statistically sig-nificant in the incident RA cohort. Thus, the overall findings from this study reflect the uncertainty of existing literature, with some studies supporting an association between asthma and RA (4–8,29–31) and others refuting such an association (24,32–36).

On the other hand, the presence of a positive association between allergy and RA was initially unexpected, since current immunologic theory segregates RA into the Th1 pathway and allergy into the Th2 pathway (37). A review of older literature shows that the association between allergy and RA was such a strong clinical observation that rheumatologists thought allergy might be involved in the pathogenesis of RA (38,39). Subsequently, sev-eral small studies showed no association between atopy and RA (32–34,40,41). More recently, however, several larger cohort stud-ies have repeatedly demonstrated a positive association between allergic disease and RA (5,29,42–44), and a study in The Nether-lands showed that parental autoimmunity increased the odds of offspring allergic disease (45). Biologic support for this association includes higher levels of IgE antibodies (46), interleukin- 4 produc-tion (47), and mast cells (48) in patients with RA. It is possible that as with RA and asthma, a broader problem with immune dysreg-ulation underlies both types of diseases, and having one predis-poses to the other.

Passive smoke exposure at home or the workplace, includ-ing the age the exposure began, duration, intensity, and pack- years, was not associated with RA. A prior study also showed no evidence of association when combining home and work-place exposure (17). However, because information about the intensity of smoke exposure was not collected in that study, the dose- response effect was not evaluated. When stratifying pas-sive smoke exposure by pack- years, we did find evidence of an association between RA and the highest levels of workplace smoke exposure and combined home and workplace smoke exposure. Thus, high levels of passive smoke exposure may place individuals at higher risk for RA, even among smokers. This finding may explain why an association of passive smok-ing with RA was suggested in 2 other studies, especially among individuals with a longer duration of exposure extending back to childhood (10,18).

Personal smoking was also associated with RA compared to nonsmoking in the full cohort. This is consistent with find-ings in previous studies (10,16), which provides reassurance about the validity of our study. However, personal smoking was not associated with RA in our incident RA cohort. There are several potential explanations for this. One is that the inci-dent RA cohort reflects a different subtype of RA, compared to older cases with disease etiology rooted in factors other than smoking. Alternatively, it is possible that there were more sub-jects who were misclassified in the incident RA group than the full RA cohort, biasing results in that group to the null hypoth-esis. Thus, it is important to consider both the full RA cohort and the incident RA cohort when interpreting study findings. A novel finding related to personal smoking history was that ear-lier age of smoking was not associated with increased odds of developing RA. Therefore, earlier age at start of smoking may not be more dangerous compared to start of smoking later in life, with respect to RA risk.

Table 4. Adjusted association between personal smoking and RA*

Multivariable OR (95% CI)

All RA cases (n = 1,023)

Incident RA subset (n = 175)

Ever smoked vs. never

1.26 (1.09–1.46)† 1.26 (0.88–1.81)

Current smoker vs. past/never

1.07 (0.78–1.46) 0.68 (0.29–1.59)

Duration, years‡ 1.18 (1.06–1.31)† 0.97 (0.77–1.22)Packs per day 1.20 (1.03–1.40)† 1.06 (0.71–1.58)Pack- years‡ 1.07 (1.01–1.13)† 0.96 (0.84–1.10)

* Each variable was assessed in a separate model with adjustment for age, sex, body mass index, race, education. RA = rheumatoid arthritis; OR = odds ratio; 95% CI = 95% confidence interval. † P < 0.05. ‡ Odds ratios (ORs) reported per 10 units.

ORAL-RESPIRATORY EXPOSURES AND RA |      1223

One strength of this study is its large sample size. Another is its detailed questionnaire data. These detailed survey questions provided impressive granularity for the exposures of interest such as personal and passive smoking, and also potential confound-ers including allergies, urban environment, and educational back-ground. Future studies can leverage not only these questionnaire data, but also the associated lifestyle, clinical, genetic, and sero-logic data contained within this rich and novel dataset (21).

There are also several important limitations to consider. First, using a convenience sample from clinics rather than a population- based study creates potential selection bias and limits general-izability, even when the population comes mainly from primary care as in this study. Second, recall bias is possible when ask-ing patients about past exposures. However, participants were asked about all comorbidities and exposures both related and unrelated to RA, and the similarity of the results from the inci-dent RA group (which did not have RA at the time of the sur-vey) provides some reassurance. Third, the definition of RA relied partially on self- report, creating potential misclassification bias. Nevertheless, such bias would have reduced any observed asso-ciations, and the similarity with incident RA cases was reassuring. Furthermore, the high PPV found with the 2 verification methods provides further reassurance about the rules- based definition of RA used in this study. Finally, for many study participants, data on RA autoantibody status were not available, limiting the number of calculations that could be performed in the sensitivity analysis. Notably, Hedstrom et al showed no difference in the association between passive smoke and RA by antibody status (17).

In conclusion, our results show that asthma and allergies may be associated with increased risk of developing RA, but passive smoke exposure and age at start of smoking are not. Future studies investigating the relationship between RA and early- life atopy are needed.

ACKNOWLEDGMENT

The authors would like to thank Mayo Clinic Center for Indi-vidualized Medicine.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti-cally for important intellectual content, and all authors approved the final version to be published. Dr. Kronzer had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Kronzer, Crowson, Sparks, Vassallo, Davis.Acquisition of data. Kronzer, Crowson, Davis.Analysis and interpretation of data. Kronzer, Crowson, Sparks, Vassallo, Davis.

REFERENCES 1. Myasoedova E, Crowson CS, Kremers HM, Therneau TM, Gabriel SE.

Is the incidence of rheumatoid arthritis rising? Results from Olmsted County, Minnesota, 1955–2007. Arthritis Rheum 2010;62:1576–82.

2. Janssen KM, de Smit MJ, Brouwer E, de Kok FA, Kraan J, Altenburg  J, et al. Rheumatoid arthritis- associated autoantibodies in  non- rheumatoid arthritis patients with mucosal inflammation: a case- control study. Arthritis Res Ther 2015;17:174.

3. Quirke AM, Perry E, Cartwright A, Kelly C, De Soyza A, Eggleton P, et al. Bronchiectasis is a model for chronic bacterial infection inducing autoimmunity in rheumatoid arthritis. Arthritis Rheumatol 2015;67:2335–42.

4. Sheen YH, Rolfes MC, Wi CI, Crowson CS, Pendegraft RS, King KS, et  al. Association of asthma with rheumatoid arthritis: a population- based case- control study. J Allergy Clin Immunol Pract 2018;6:219–26.

5. Lai NS, Tsai TY, Koo M, Lu MC. Association of rheumatoid arthritis with allergic diseases: a nationwide population- based cohort study. Allergy Asthma Proc 2015;36:99–103.

6. Hemminki K, Li X, Sundquist J, Sundquist K. Subsequent autoimmune or related disease in asthma patients: clustering of diseases or medical care? Ann Epidemiol 2010;20:217–22.

7. De Roos AJ, Cooper GS, Alavanja MC, Sandler DP. Personal and family medical history correlates of rheumatoid arthritis. Ann Epidemiol 2008;18:433–9.

8. Hassan WU, Keaney NP, Holland CD, Kelly CA. Bronchial reactivity and airflow obstruction in rheumatoid arthritis. Ann Rheum Dis 1994;53:511–4.

9. Hart JE, Laden F, Puett RC, Costenbader KH, Karlson EW. Exposure to traffic pollution and increased risk of rheumatoid arthritis. Environ Health Perspect 2009;117:1065–9.

10. Costenbader KH, Feskanich D, Mandl LA, Karlson EW. Smoking intensity, duration, and cessation, and the risk of rheumatoid arthritis in women. Am J Med 2006;119:503.

11. Chang KH, Hsu CC, Muo CH, Hsu CY, Liu HC, Kao CH, et al. Air pollution exposure increases the risk of rheumatoid arthritis: a longitudinal and nationwide study. Environ Int 2016;94:495–9.

12. De Roos AJ, Koehoorn M, Tamburic L, Davies HW, Brauer M. Proximity to traffic, ambient air pollution, and community noise in relation to incident rheumatoid arthritis. Environ Health Perspect 2014;122:1075–80.

13. Burke H, Leonardi-Bee J, Hashim A, Pine-Abata H, Chen Y, Cook DG, et al. Prenatal and passive smoke exposure and incidence of asthma and wheeze: systematic review and meta- analysis. Pediatrics 2012;129:735–44.

14. Guan WJ, Zheng XY, Chung KF, Zhong NS. Impact of air pollution on the burden of chronic respiratory diseases in China: time for urgent action. Lancet 2016;388:1939–51.

15. Chang K, Yang SM, Kim SH, Han KH, Park SJ, Shin JI. Smoking and rheumatoid arthritis. Int J Mol Sci 2014;15:22279–95.

16. Di Giuseppe D, Discacciati A, Orsini N, Wolk A. Cigarette smoking and risk of rheumatoid arthritis: a dose- response meta- analysis. Arthritis Res Ther 2014;16:R61.

17. Hedstrom AK, Klareskog L, Alfredsson L. Exposure to passive smoking and rheumatoid arthritis risk: results from the Swedish EIRA study. Ann Rheum Dis 2018;77:970–2.

18. Seror R, Henry J, Gusto G, Aubin H, Boutron-Ruault M, Mariette X. Passive smoking in childhood increases risk of developing rheumatoid arthritis. Rheumatology (Oxford) 2018. E-pub ahead of print.

19. Sparks JA, Chang SC, Deane KD, Gan RW, Demoruelle MK, Feser ML, et al. Associations of smoking and age with inflammatory joint signs among unaffected first- degree relatives of rheumatoid arthritis patients: results from studies of the etiology of rheumatoid arthritis. Arthritis Rheumatol 2016;68:1828–38.

20. Von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Int J Surg 2014;12:1495–9.

KRONZER ET AL 1224       |

21. Olson JE, Ryu E, Johnson KJ, Koenig BA, Maschke KJ, Morrisette  JA, et al. The Mayo Clinic Biobank: a building block for individualized medicine. Mayo Clin Proc 2013;88: 952–62.

22. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper  NS, et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24.

23. Aletaha D, Neogi T, Silman AJ, Funovits J, Felson DT, Bingham CO III, et al. 2010 rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum 2010;62: 2569–81.

24. Sparks JA, Lin TC, Camargo CA Jr, Barbhaiya M, Tedeschi SK, Costenbader KH, et al. Rheumatoid arthritis and risk of chronic obstructive pulmonary disease or asthma among women: a marginal structural model analysis in the Nurses’ Health Study. Semin Arthritis Rheum 2018;47:639–48.

25. Oksanen T, Kivimaki M, Pentti J, Virtanen M, Klaukka T, Vahtera J. Self- report as an indicator of incident disease. Ann Epidemiol 2010;20:547–54.

26. Bang DW, Wi CI, Kim EN, Hagan J, Roger V, Manemann S, et  al. Asthma status and risk of incident myocardial infarction: a population- based case- control study. J Allergy Clin Immunol Pract 2016;4:917–23.

27. Chute CG, Beck SA, Fisk TB, Mohr DN. The Enterprise Data Trust at Mayo Clinic: a semantically integrated warehouse of biomedical data. J Am Med Inform Assoc 2010;17:131–5.

28. Kamradt T. Can infections prevent or cure allergy and autoimmunity? Discov Med 2005;5:283–7.

29. Jeong HE, Jung SM, Cho SI. Association between rheumatoid arthritis and respiratory allergic diseases in Korean adults: a propensity score matched case- control study. Int J Rheumatol 2018;2018: 3798124.

30. Kero J, Gissler M, Hemminki E, Isolauri E. Could TH1 and TH2 diseases coexist? Evaluation of asthma incidence in children with coeliac disease, type 1 diabetes, or rheumatoid arthritis: a register study. J Allergy Clin Immunol 2001;108:781–3.

31. Provenzano G, Donato G, Brai G, Rinaldi F. Prevalence of allergic respiratory diseases in patients with RA. Ann Rheum Dis 2002;61:281.

32. Hartung AD, Bohnert A, Hackstein H, Ohly A, Schmidt KL, Bein G. Th2- mediated atopic disease protection in Th1- mediated rheumatoid arthritis. Clin Exp Rheumatol 2003;21:481–4.

33. Rudwaleit M, Andermann B, Alten R, Sorensen H, Listing J, Zink A, et al. Atopic disorders in ankylosing spondylitis and rheumatoid arthritis. Ann Rheum Dis 2002;61:968–74.

34. Olsson AR, Wingren G, Skogh T, Svernell O, Ernerudh J. Allergic manifestations in patients with rheumatoid arthritis. APMIS 2003;111:940–4.

35. Sheikh A, Smeeth L, Hubbard R. There is no evidence of an inverse relationship between TH2- mediated atopy and TH1- mediated autoimmune disorders: lack of support for the hygiene hypothesis. J Allergy Clin Immunol 2003;111:131–5.

36. Bergstrom U, Jacobsson LT, Nilsson JA, Berglund G, Turesson C. Pulmonary dysfunction, smoking, socioeconomic status and the risk of developing rheumatoid arthritis. Rheumatology (Oxford) 2011;50:2005–13.

37. Singh VK, Mehrotra S, Agarwal SS. The paradigm of Th1 and Th2 cytokines: its relevance to autoimmunity and allergy. Immunol Res 1999;20:147–61.

38. Iakovleva AA. The role of allergy in the pathogenesis of rheumatoid arthritis. Sov Med 1964;27:29–36.

39. Ragan C. Role of hypersensitivity in the pathogenesis of rheumatoid arthritis. Ann Rheum Dis 1959;18:1–7.

40. O’Driscoll BR, Milburn HJ, Kemeny DM, Cochrane GM, Panayi GS. Atopy and rheumatoid arthritis. Clin Allergy 1985;15:547–53.

41. Hilliquin P, Allanore Y, Coste J, Renoux M, Kahan A, Menkes CJ. Reduced incidence and prevalence of atopy in rheumatoid arthritis: results of a case- control study. Rheumatology (Oxford) 2000;39:1020–6.

42. Hou YC, Hu HY, Liu IL, Chang YT, Wu CY. The risk of autoimmune connective tissue diseases in patients with atopy: a nationwide population- based cohort study. Allergy Asthma Proc 2017;38:383–9.

43. Karsh J, Chen Y, Lin M, Dales R. The association between allergy and rheumatoid arthritis in the Canadian population. Eur J Epidemiol 2005;20:783–7.

44. Simpson CR, Anderson WJ, Helms PJ, Taylor MW, Watson L, Prescott GJ, et al. Coincidence of immune- mediated diseases driven by Th1 and Th2 subsets suggests a common aetiology: a population- based study using computerized general practice data. Clin Exp Allergy 2002;32:37–42.

45. Maas T, Nieuwhof C, Passos VL, Robertson C, Boonen A, Landewe  RB, et al. Transgenerational occurrence of allergic disease and autoimmunity: general practice- based epidemiological research. Prim Care Respir J 2014;23:14–21.

46. Burastero SE, Lo Pinto G, Goletti D, Cutolo M, Burlando L, Falagiani  P. Rheumatoid arthritis with monoclonal IgE rheumatoid factor. J Rheumatol 1993;20:489–94.

47. Frieri M. Neuroimmunology and inflammation: implications for therapy of allergic and autoimmune diseases. Ann Allergy Asthma Immunol 2003;90 Suppl 3:34–40.

48. Xu Y, Chen G. Mast cell and autoimmune diseases. Mediators Inflamm 2015;2015:246126.

1225

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1225–1231DOI 10.1002/art.40870 © 2019, American College of Rheumatology

Derivation and Validation of a Major Toxicity Risk Score Among Nonsteroidal Antiinflammatory Drug Users Based on Data From a Randomized Controlled TrialDaniel H. Solomon,1 Ming Shao,2 Kathy Wolski,2 Steven Nissen,2 M. Elaine Husni,2 and Nina Paynter1

Objective. While nonsteroidal antiinflammatory drugs (NSAIDs) are commonly used in rheumatology, they can cause major toxicity. Improving the risk/benefit ratio requires a more precise understanding of risk. This study was undertaken to derive and validate a risk score for major toxicity among NSAID users enrolled in a randomized con-trolled trial.

Methods. Patients enrolled in a randomized controlled trial who had known cardiovascular disease or risk factors as well as osteoarthritis or rheumatoid arthritis were divided into derivation and validation cohorts. Patients were randomized to receive celecoxib, naproxen, or ibuprofen at typical dosages. The risk score was designed to predict the 1- year occurrence of major toxicity among NSAID users, including major adverse cardiovascular events, acute kidney injury, significant gastrointestinal events, and mortality. Variables significantly associated with major toxicity were candidates for inclusion in the final regression model. After derived models were found to have a similar model fit in the validation set, the cohorts were combined, allowing calculation of a risk score.

Results. In the derivation cohort, significant variables included age, male sex, history of cardiovascular disease, hypertension, diabetes mellitus, tobacco use, statin use, elevated serum creatinine level, hematocrit level, and type of arthritis. The C- index was 0.73 in the validation cohort and 0.71 in the total cohort; the model was well calibrated. Of the total population with complete data (n = 23,735), 1,080 participants (4.6%) had a predicted 1- year risk of major toxicity of <1%, 16,273 (68.6%) had a predicted risk of 1–4%, and 6,382 (26.9%) had a predicted risk of >4%.

Conclusion. The risk score accurately categorizes the 1- year risk of major toxicity among NSAID users and may be useful in identifying patients who can safely use these agents.

INTRODUCTION

Nonsteroidal antiinflammatory drugs (NSAIDs) are the most common medications used regularly in the US, with estimates of 10 million people taking them daily (1). They are relatively safe treatment options when used intermittently by most people, prompting their widespread availability as over- the- counter prepa-rations. However, among patients with comorbid conditions who use them at moderate dosages for long periods of time, they pose a risk of significant toxicity. Estimates from the late 1990s sug-gest that ~17,000 people died annually from NSAID toxicities (2). Mortality from NSAIDs may have decreased since the introduction of proton- pump inhibitors (3). Well- recognized toxicities include

gastrointestinal (GI) complications, acute kidney injury, cardiovas-cular (CV) events, and death (4).

Balancing the analgesic and antiinflammatory benefits of NSAIDs with the potential for risk requires an appreciation of how clinical factors impact the relative and absolute risk of tox-icity in a given patient. While genomics may lead us to a pre-cision medicine–based approach to the use of these agents, basing recommendations on clinical risk factor data that is easily obtained at the individual level may help to improve their safety. Risk scores can facilitate application of risk factor epidemiology in the clinic (5). Good examples of this include the Framingham Risk Score (6), the Systematic Coronary Risk Evaluation (7), the American College of Cardiology/American

ClinicalTrials.gov identifier: NCT00346216.The PRECISION trial was supported by Pfizer.1Daniel H. Solomon, MD, MPH, Nina Paynter, PhD: Brigham and Women’s

Hospital, Boston, Massachusetts; 2Ming Shao, PhD, Kathy Wolski, MPH, Steven Nissen, MD, M. Elaine Husni, MD, MPH: Cleveland Clinic Foundation, Cleveland, Ohio.

Dr. Shao and Ms Wolski have received research support from Pfizer. Dr. Nissen has received consulting fees and/or honoraria from Pfizer (less than

$10,000) and research support from Pfizer paid to his institution. No other disclosures relevant to this article were reported.

Address correspondence to Daniel H. Solomon, MD, MPH, Brigham and Women’s Hospital, Division of Rheumatology, 60 Fenwood Road, Boston, MA 02115. E-mail: dsolomon@bwh.harvard.edu.

Submitted for publication January 11, 2019; accepted in revised form February 21, 2019.

SOLOMON ET AL 1226       |

Heart Association guidelines (8), and the fracture risk assess-ment tool (9).

Risk score development and validation requires a large and accurately phenotyped cohort with thorough follow- up. We have applied recommended methods to derive and internally validate a risk score for major NSAID toxicity using data from the Prospec-tive Randomized Evaluation of Celecoxib Integrated Safety versus Ibuprofen or Naproxen (PRECISION) trial (4,5).

PATIENTS AND METHODS

Study design and participants. The current study used information derived from the PRECISION trial, a large randomized controlled trial that compared the safety of celecoxib, naproxen, and ibuprofen (4). No placebo group was included. The trial was performed in 14 countries between 2006 and 2016. Local human subject committees approved the protocol, and all par-ticipants provided written informed consent that included use of data in secondary analyses. These secondary analyses were not prespecified in the statistical analysis plan. The trial sponsor had no authorship in the current manuscript but was given the oppor-tunity to provide feedback.

The study population included patients with a clinical diag-nosis of osteoarthritis (OA) or rheumatoid arthritis (RA) who were ≥18 years of age and required regular daily treatment with an NSAID. All participants were required to have a history of a CV event, occlusive coronary or noncoronary arterial disease, diabe-tes mellitus, or ≥2 CV risk factors for women and ≥3 for men. Risk factors included age ≥65 years for women and >55 years for men, hypertension, dyslipidemia, microalbuminuria, urine protein:creati-nine ratio >2, ankle brachial index <0.9, left ventricular hypertro-phy, current cigarette smoking, waist:hip ratio ≥0.90, and family history of premature CV disease (see Supplementary Table 1, on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40870/ abstract). Exclusion criteria included a CV event within the prior 3 months, New York Heart Associa-tion class III or IV heart failure, and other characteristics listed in Supplementary Table 1. The instructions for completing the case report forms prompted study staff to use medication information and/or ask the patient each question.

All patients in the trial cohort were considered for inclusion in the analyses. Fewer than 1% of patients (n = 215) had some missing data at baseline and were excluded. The full cohort was divided into derivation and validation cohorts. As recommended (5), the split was not random; rather, patients enrolled in the first 4 years of the trial were included in the derivation cohort and those enrolled in the last 5 years were included in the validation cohort.

Outcome measures. The primary outcome measure for the current analyses was a composite of major toxicity among NSAID users. All toxicities were centrally adjudicated in a blinded manner and included major adverse CV events (MACEs), clinically

significant GI events (CSGIEs), acute kidney injury, and death. MACEs included events defined by Antiplatelet Trialists Collabora-tion, in addition to revascularization or hospitalization for transient ischemic attack or unstable angina. CSGIEs included gastrodu-odenal hemorrhage, gastric outlet obstruction, perforation of the gastroduodenum or small/large bowel, hemorrhage of the small/large bowel, acute GI hemorrhage of unknown origin, or symp-tomatic gastric or duodenal ulcer. Acute kidney injury was defined as development of renal insufficiency or renal failure, including any of the following: serum creatinine level of ≥2.0 mg/dl and increase of ≥0.7 mg/dl from baseline, hospitalization for acute renal fail-ure with either a doubling of the baseline serum creatinine level or hyperkalemia with ≥50% elevation in serum creatinine level, or initiation of dialysis. The composite primary outcome measure included all causes of death.

Examination for the occurrence of these outcomes took place during the first 1- year study period, while patients were receiving the study NSAID and for 30 days after termination of the NSAID, to capture only on- drug outcomes. Patients were censored from the analysis at the first instance of any of the following: death, after 1 year of receiving the study NSAID, or 30 days after discontinuing the study NSAID.

Potential risk factors. The potential predictors of major toxicity among NSAID users included variables measured at base-line that had been considered as potential risk factors in prior stud-ies (2,10–12). We also focused on easily measured variables that could be considered in typical clinical practice. However, because we assumed that the variables would be weighted in the final risk score, requiring computer calculation, variables were not simpli-fied into categories but left as continuous whenever possible.

Demographic variables included age and sex. Clinical var-iables included type of arthritis (OA or RA), known CV disease (i.e., prior event, occlusive coronary/noncoronary arterial disease), known diabetes mellitus, known hypertension, known hyperlipi-demia, or a prior GI bleed. Medication use of interest included low- dose aspirin for CV prevention, statins/lipid- lowering drugs, or glucocorticoids (any dosage). We also considered a patient’s functional status as measured with the Health Assessment Ques-tionnaire (13), current tobacco use, and several laboratory meas-ures (including serum creatinine and hematocrit levels).

Statistical analysis. After dividing the trial cohort into the derivation group (enrollment 2006–2010) and the validation group (enrollment 2011–2016), we examined baseline charac-teristics. Major toxicity rates among patients receiving NSAIDs were estimated across both groups. We then used data from the derivation cohort to fit a multivariable model using Cox pro-portional hazards regression, considering only the first toxicity event and a time period of days since randomization (PROC PHREG in SAS). Censoring events included a toxicity event, loss to follow- up, or 1 year, whichever occurred first. Age- and sex-

NSAID TOXICITY RISK SCORE |      1227

adjusted models were examined first. Each potential variable was tested in regression models. Variables with hazard ratios (HRs) of >1.2 or P values of <0.001 were considered poten-tial risk factors and tested as a group in multivariable models, including age and sex. Interquartile ranges (IQRs) and 95% con-fidence intervals (95% CIs) were calculated. Hematocrit was tested using linear splines with a knot set at a hematocrit level of 43%. To construct a relatively parsimonious model, variables were removed if the HRs were no longer >1.2 in the multivariable model. Model discrimination was assessed using Harrell’s C- index, and model calibration was examined by visually graphing observed and predicted risk curves with the calibration intercept and slope. The derivation cohort model was then applied to the validation cohort using the macros developed by Cook et al (14), and performance statistics were again assessed. All analyses were conducted using SAS (version 9.4) and the open- source software R.

Based on good performance in the validation cohort, the cohorts were combined for increased power and the risk model was reestimated using the same risk factors. The apparent per-formance of this final model in the full cohort was also assessed. Using the final risk model, a formula was developed (14) to pre-dict risk probabilities in future patients. To address clinical decision

making, 3 categories of 1- year risk of major toxicity were defined: low risk (<1%), intermediate risk (1–4%), and high risk (>4%). Sur-vival curves were plotted for the 3 categories and risk of outcomes were compared using a log rank test (PROC LIFETEST in SAS). Finally, we assessed model calibration among several subgroups, by treatment arm and dosage. The 3 treatment arms included celecoxib, naproxen, and ibuprofen. The dosage analysis exam-ined patients within each treatment group separately according to dosage, i.e., for each treatment, those who continued to receive the starting dosage (celecoxib 100 mg twice daily, naproxen 375 mg twice daily, or ibuprofen 600 mg 3 times daily) were assessed as a subgroup, and those in whom the dosage was up- titrated to improve analgesia (celecoxib 200 mg twice daily, naproxen 500 mg twice daily, or ibuprofen 800 mg 3 times daily) were assessed as another subgroup. Local drug labeling allowed the up- titration of celecoxib to 200 mg twice daily only for patients with RA.

RESULTS

The populations included in the cohorts to derive and validate the risk score were similar (Table 1). The median age in the com-bined cohort was 63 years (IQR 57–70), 64% were women, and the median body mass index was 31.4 kg/m2 (IQR 27.5–36.3).

Table 1. Patient characteristics*

Full cohort (n = 23,950) Derivation cohort (n = 15,194) Validation cohort (n = 8,756)

Age, median (IQR) years 63.0 (57.0–70.0) 62.0 (57.0–69.0) 65.0 (58.0–70.0)Male sex 8,591 (35.9) 4,668 (30.7) 3,923 (44.8)BMI, median (IQR) kg/m2 31.4 (27.5–36.3) 31.3 (27.4–36.2) 31.6 (27.6–36.5)Tobacco use 4,975 (20.8) 2,701 (17.8) 2,274 (26.0)Type of arthritis

OA 21,525 (89.9) 13,413 (88.3) 8,112 (92.6)RA 2,425 (10.1) 1,781 (11.7) 644 (7.4)

History of diabetes mellitus 8,445 (35.5) 5,360 (35.7) 3,085 (35.2)History of hypertension 18,644 (78.4) 11,617 (77.4) 7,027 (80.3)History of hyperlipidemia 14,971 (63.0) 10,160 (67.7) 4,811 (55.0)Prior CV event 1,201 (5.1) 653 (4.3) 548 (6.3)Prior GI bleed 0 (0) 0 (0) 0 (0)Serum creatinine, median (IQR)

mg/dl0.87 (0.74–1.02) 0.87 (0.74–1.02) 0.87 (0.73–1.01)

Hematocrit, median (IQR) % 41.0 (39.0–44.0) 41.0 (39.0–44.0) 42.0 (39.0–45.0)Hematocrit level <43% 14,658 (61.2) 9,681 (63.7) 4,977 (56.8)

Use of aspirin 11,017 (46.0) 6,740 (44.4) 4,277 (48.8)Use of statins 12,913 (53.9) 7,951 (52.3) 4,962 (56.7)Use of glucocorticoids 3,089 (12.9) 1,914 (12.6) 1,175 (13.4)Use of DMARDs 1,748 (7.3) 1,314 (8.6) 434 (5.0)Functional status, median

(IQR)†1.1 (0.6–1.5) 1.1 (0.6–1.5) 1.1 (0.6–1.5)

* Except where indicated otherwise, values are number (%) of patients. IQR = interquartile range; BMI = body mass index; OA = osteoarthritis; RA = rheumatoid arthritis; CV = cardiovascular; GI = gastrointestinal; DMARDs = disease- modifying antirheumatic drugs. † Measured with the Health Assessment Questionnaire, range 0 (no limitation) to 3 (unable to perform).

SOLOMON ET AL 1228       |

Ninety percent of patients were diagnosed as having OA and 10% as having RA. Twenty- two percent had a known prior CV event, 36% had diabetes mellitus, 78% had hypertension, and 63% had hyperlipidemia. Patients were followed up for outcome measures 1 year after randomization, with 59% having ≥12 months of fol-low- up while receiving an NSAID. Of those with <12 months of follow- up while receiving an NSAID, the median follow- up was 3.7 months. Table 2 shows the incidence rates for the components and the composite of major toxicity among NSAID users in both cohorts. The rate of development of the primary outcome meas-

ure in the total cohort was 3.38% (95% CI 3.12–3.65).To derive the risk score, we treated all variables from Table 1

as potential predictors of major toxicity among NSAID users. Var-iables were individually tested in age- and sex- adjusted models, and those with HRs of >1.2 or P values of <0.001 were included in a combined model. The final model was further pruned to include only variables with 95% CIs that excluded 1.00 (Table 3). The multivariable model fit statistics for the derivation cohort included a Harrell’s C- index of 0.73 and near complete agree-ment in models examining observed risk in relation to predicted risk (Supplementary Figure 1A, http://onlin elibr ary.wiley.com/doi/10.1002/art.40870/ abstract). The model fit statistics in the validation cohort were similar to those in the derivation cohort, with a Harrell’s C- index of 0.68 and a calibration slope of 0.774 (Supplementary Figure 1B).

After deriving the multivariable risk model and assessing its validity, we used the total cohort to calculate a risk score: (0.0325 × age) + (0.2666 × sex [male = 1]) + (0.8352 × known CV disease [yes = 1]) + (0.2252 × known hypertension [yes = 1]) + (0.3434 × known diabetes [yes = 1]) + (0.3653 × current

cigarette use [yes = 1]) + (0.1849 × statin/lipid-lowering drug use [yes = 1]) + (1.0964 × baseline serum creatinine [per 1 mg/dl increase]) + (0.5403 × known RA [yes = 1]) – (0.0742 ×

Table 2. One- year outcome rates for major toxicity among NSAID users*

Full cohort (n = 23,950)

Derivation cohort (n = 15,194)

Validation cohort (n = 8,756)

EventsPerson-

yearsRate

(95% CI) EventsPerson-

yearsRate

(95% CI) EventsPerson-

yearsRate

(95% CI)

Primary outcome measure

617 18,261 3.38 (3.12–3.65)

390 11,527 3.38 (3.06–3.72)

227 6,734 3.37 (2.94–3.81)

MACE 374 18,296 2.04 (1.83–2.25)

234 11,550 2.03 (1.76–2.29)

140 6,746 2.08 (1.74–2.41)

CSGIE 133 18,339 0.73 (0.60–0.85)

89 11,581 0.77 (0.60–0.93)

44 6,758 0.65 (0.45–0.84)

Acute kidney injury†

89 18,348 0.49 (0.38–0.59)

57 11,586 0.49 (0.37–0.62)

32 6,762 0.47 (0.31–0.63)

Death‡ 90 18,354 0.49 (0.39–0.59)

52 11,591 0.45 (0.33–0.58)

38 6,763 0.56 (0.39–0.74)

* Rates are per 100 person- years. The primary outcome measure is a composite of major toxicities among nonsteroidal antiinflammatory drug (NSAID) users, which includes 4 components: major adverse cardiovascular event (MACE), clinically significant gastrointestinal event (CSGIE), acute kidney injury, and death. 95% CI = 95% confidence interval. † Acute kidney injury was defined as development of renal insufficiency or renal failure, including any of the following: serum creatinine level of ≥2.0 mg/dl and increase of ≥0.7 mg/dl from baseline, hospitalization for acute renal failure with a doubling of the baseline serum creatinine level or hyperkalemia with ≥50% elevation in serum creatinine, or initiation of dialysis. ‡ Includes all causes.

Table 3. Multivariable HRs from final multivariable- adjusted models predicting primary outcome*

Derivation cohort, HR (95% CI)

Full cohort, HR (95% CI)

Age, years 1.03 (1.02–1.04) 1.03 (1.02–1.04)Male sex 1.31 (1.02–1.68) 1.31 (1.07–1.59)Tobacco use 1.53 (1.17–1.99) 1.44 (1.17–1.77)History of diabetes

mellitus1.49 (1.21–1.82) 1.41 (1.20–1.66)

History of hypertension

1.33 (1.01–1.76) 1.25 (1.01–1.56)

Prior CV event 2.75 (2.22–3.39) 2.31 (1.95–2.72)Serum creatinine,

mg/dl2.54 (1.61–4.01) 2.99 (2.09–4.28)

Hematocrit level <43%†

0.92 (0.89–0.96) 0.93 (0.90–0.96)

Hematocrit level ≥43%†

1.06 (1.003–1.12) 1.04 (1.001–1.09)

Use of statins 1.33 (1.06–1.65) 1.20 (1.01–1.43)RA 1.70 (1.29–2.23) 1.72 (1.37–2.15)

* All variables from Table 1 were tested in age- and sex- adjusted models. Those with hazard ratios (HRs) of >1.2 or P values of <0.001 were tested in adjusted models. We then removed variables with 95% confidence intervals (95% CIs) that included 1.00. CV = cardio-vascular; RA = rheumatoid arthritis. † Hematocrit was tested using linear splines with a knot set at a hematocrit level of 43%.

NSAID TOXICITY RISK SCORE |      1229

[the lesser of hematocrit level or 43]) + (0.0433 × [the greater of 0 or hematocrit level – 43]).

We then created the 3 risk groups—low risk (<1%), interme-diate risk (1–4%), and high risk (>4%)—based on the predicted 1- year risk probabilities. The 1- year risk probability of major toxicity among NSAID users can be predicted with the following formula:

Table 4 illustrates the number of patients in the trial cohort in each risk category and the 1- year risk of the primary outcome measure, as well as each major toxicity. In the total population with complete data (n = 23,735), 1,080 patients (4.6%) had a predicted 1- year risk of <1%, 16,273 (68.6%) had a predicted risk of 1–4%, and 6,382 (26.9%) had a predicted risk of >4%. The Kaplan- Meier survival curves for the 3 categories of risk diverged early during follow- up (Figure 1), and the log rank tests

showed significant differences (P < 0.001).

In subgroup analyses of specific NSAID exposure, we exam-ined the predicted and observed risk of major toxicity among NSAID users and found strong calibration between them in the following subgroups: calibration slope was 1.0831 for celecoxib, 0.9625 for ibuprofen, and 0.9753 for naproxen (see Supple-mentary Figure 2, http://onlin elibr ary.wiley.com/doi/10.1002/art.40870/ abstract). Additionally, a separate subgroup analysis examined the performance of the risk score among patients receiving typical starting dosages versus those receiving higher dosages of the 3 NSAIDs. The risk score had strong calibration in these subgroups: calibration slope was 1.0615 for typical dos-ages and 0.9077 for higher dosages (Supplementary Figure 3, http://onlin elibr ary.wiley.com/doi/10.1002/art.40870/ abstract).

DISCUSSION

We used a large cohort of well- phenotyped patients to derive and validate a risk score for major toxicity among NSAID users. All patients had OA or RA and were enrolled in a randomized controlled trial comparing the safety of celecoxib, naproxen, and ibuprofen; they were offered concomitant esomeprazole, a proton- pump inhibitor. Adjusted models tested a broad range of potential risk factors, and 11 easily assessed variables were found to be significant in multivariable models. Calibration and discrimi-nation statistics were moderately strong in both the derivation and the validation cohorts, and a risk score accurately predicted major toxicity among NSAID users at 1 year. The risk score defined 3 groups of patients: 4.6% of patients were at low risk (<1%), 68.6% at intermediate risk (1–4%), and 26.9% at high risk (>4%).

External validation will be important, but the split- sampling and large cohort in this study suggest that the risk score is likely to be useful in stratifying patients. Testing the risk score in pop-ulations that are not receiving proton- pump inhibitors, those with a different distribution of risk factors (i.e., fewer CV risk factors), and patients with different underlying causes of chronic pain will help determine the external validity and generalizability of the risk score. There were some slight differences between the derivation

100%×

(

1−0.96855erisk score−1.05501

)

Table 4. Risk of major toxicity among NSAID users and component outcomes by risk category*

No. (%)

Major NSAID toxicity, rate

(95% CI)MACE, rate

(95% CI)CSGIE, rate

(95% CI)

Acute kidney injury, rate (95% CI)†

Death, rate (95% CI)‡

Low risk (<1%) 1,080 (4.6)

0.40 (0.05–0.76)

0.24 (0.0–0.52)

0.16 (0.0–0.38)

0 0.16 (0.0–0.38)

Intermediate risk (1–4%)

16,273 (68.6)

1.69 (1.49–1.89)

0.99 (0.84–1.14)

0.48 (0.38–0.59)

0.15 (0.09–0.21)

0.24 (0.16–0.31)

High risk (>4%)

6,382 (26.9)

5.56 (4.98–6.14)

3.48 (3.01–3.95)

0.84 (0.61–1.08)

1.06 (0.80–1.33)

0.83 (0.60–1.06)

* Rates are per 100 patients. NSAID = nonsteroidal antiinflammatory drug; 95% CI = 95% confidence interval; MACE = major adverse cardio-vascular event; CSGIE = clinically significant gastrointestinal event. † Acute kidney injury was defined as development of renal insufficiency or renal failure, including any of the following: serum creatinine level of ≥2.0 mg/dl and increase of ≥0.7 mg/dl from baseline, hospitalization for acute renal failure with a doubling of the baseline serum creatinine level or hyperkalemia with ≥50% elevation in serum creatinine, or initiation of dialysis. ‡ Includes all causes.

Figure  1. Kaplan- Meier survival curve illustrating the event- free survival of subjects in each of the 3 risk score categories for major toxicity among nonsteroidal antiinflammatory drug (NSAID) users. P < 0.001 in all log rank tests.

SOLOMON ET AL 1230       |

and validation cohorts due to a midtrial protocol amendment that attempted to increase the CV event rates (Table 1). In the current analyses, all patients were receiving an NSAID, so the incremental risk associated with the use of these drugs cannot be estimated compared to nonuse.

Further analysis of the percentage of patients found to be in each of the 3 risk groups (low, intermediate, and high) can provide insight into the safety of NSAID use in adults with chronic painful conditions. In the study cohort, a relatively small group (4.6%) had a low risk (<1%) of major toxicity within 1 year, and a fairly large group (26.9%) had a high risk (>4%). We did not choose the 3 risk thresholds using formal decision analysis but rather through clinician input about what level of risk would impact decision- making. A 1- year risk of >4% translates into a number needed to harm of ≤25; this level of risk was considered to be too high to suggest taking an NSAID. Conversely, a risk of <1% translates into a number needed to harm of ≥100; this level of risk was felt to be acceptable for almost all patients. A risk between 1% and 4% was considered to be an intermediate risk, for which some providers and patients might choose that the patient take an NSAID and others not. This middle group contained the larg-est portion of patients in our study cohort. Future work should attempt to further break down this intermediate group, as it is likely that many patients with a 1- year risk of <2% would be will-ing to use an NSAID if it provided enough analgesia. We give examples of patients in all 3 risk categories in Supplementary Table 2 and provide a web calculator in Supplementary Figure 4 (http://onlin elibr ary.wiley.com/doi/10.1002/art.40870/ abstract).

There are several notable strengths of these analyses. First, the study data came from a prospective randomized trial with well- defined baseline characteristics, standardized adjudication of outcomes, <1% missing data, and >90% follow- up for outcomes. Additionally, using data from a randomized trial removes any con-founding by indication and allowed us to develop the risk score using an equal distribution of 3 commonly prescribed NSAIDs. Second, we followed methodologic standards for risk score anal-yses (5). Third, the risk score performed well in the total group as well as in the subgroups divided by NSAID type. This suggests that the choice of NSAID does not substantially modify risks across the risk factors assessed. Fourth, the trial population was very large and treated with typical dosages of NSAIDs. This improves the generalizability of the risk score. Finally, the risk score variables are all easily obtained, improving the applicability of the score.

Limitations of the current analysis include its post hoc nature and the lack of an external validation cohort. The PRECISION trial population included people ages 18–100 years. Since there are few people at either extreme, estimates are unstable for the very young or the very old in this range. Another potential con-cern is the relatively poor long- term adherence to NSAID treat-ment observed in the PRECISION trial, which was the rationale for limiting the current analyses to 1 year. At 1 year, 59% of patients were still taking the study NSAID. We acknowledge that the risks

cannot be directly estimated at 1 year for all patients. In the trial, as in typical practice, many patients who planned to regularly take NSAIDs did not maintain the treatment; this was the rationale for censoring patients before 1 year if they were not adherent with the study NSAID. Given the variation in follow- up time in our study, we assessed the ability of our model to predict risk of events at 6 months as well as at 1 year and found good calibration. Further-more, the dosages of the 3 NSAIDs in the PRECISION trial have been criticized as nonequivalent. While these are valid criticisms of the trial, we included all 3 drugs at all dosages in the current set of analyses. We found that the risk score performed well across the 3 drugs and among patients receiving typical dosages and higher dosages. It is somewhat surprising that NSAID dosage was not an important modifier of the risk score performance. We did not test it as a potential predictor in the main analyses. While NSAID dosage has been found to be a risk factor for adverse events, after adjustment for all other variables in the risk score, it did not make a substantial difference.

In conclusion, if the risk score is found to be valid in external populations that may be more heterogeneous in characteristics, the risk of major NSAID toxicity can be predicted using simple clinical factors that should allow patients and providers to make a personalized decision regarding initiation of NSAIDs. Since NSAIDs represent one of the most common drug groups regularly taken in the US, safe use of these agents could provide impor-tant public health benefits. Moreover, data on the factors included in the risk score can be gleaned from most electronic medical records, making the dissemination of such a risk score relatively straightforward.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Solomon had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Solomon, Nissen, Paynter.Acquisition of data. Nissen, Husni.Analysis and interpretation of data. Solomon, Shao, Wolski, Paynter.

ROLE OF THE STUDY SPONSOR

Pfizer funded the PRECISION trial and supported the effort of several coauthors but had no role in the current manuscript. They were given the opportunity to review and comment on the final manuscript before submission. Publication of this article was not contingent upon approval by Pfizer.

REFERENCES 1. IMS Health. National prescription audit: oral NSAID market. 2013.

2. Wolfe MM, Lichtenstein DR, Singh G. Gastrointestinal toxicity of nonsteroidal antiinflammatory drugs. N Engl J Med 1999;340:1888–99.

3. Straube S, Tramèr MR, Moore RA, Derry S, McQuay HJ. Mortality with upper gastrointestinal bleeding and perforation: effects of time and NSAID use. BMC Gastroenterol 2009;9:41.

NSAID TOXICITY RISK SCORE |      1231

4. Nissen SE, Yeomans ND, Solomon DH, Lüscher TF, Libby P, Husni  ME, et al. Cardiovascular safety of celecoxib, naproxen, or ibuprofen for arthritis. N Engl J Med 2016;375:2519–29.

5. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med 2015;162:W1–73.

6. Lloyd-Jones DM, Wilson PW, Larson MG, Beiser A, Leip EP, D’Agostino RB, et al. Framingham risk score and prediction of lifetime risk for coronary heart disease. Am J Cardiol 2004;94:20–4.

7. Graham I, Atar D, Borch-Johnsen K, Boysen G, Burell G, Cifkova R, et al. European guidelines on cardiovascular disease prevention in clinical practice: full text. Fourth Joint Task Force of the European Society of Cardiology and other societies on cardiovascular disease prevention in clinical practice (constituted by representatives of nine societies and by invited experts). Eur J Cardiovasc Prev Rehabil 2007;14 Suppl 2:S1–113.

8. Stone NJ, Robinson JG, Lichtenstein AH, Bairey Merz CN, Blum CB, Eckel RH, et al. 2013 ACC/AHA guideline on the treatment of blood cholesterol to reduce atherosclerotic cardiovascular risk in adults: a report of the American College of Cardiology/American Heart

Association Task Force on Practice Guidelines. Circulation 2014;129 Suppl 2:S1–45.

9. Kanis JA, Johnell O, Oden A, Johansson H, McCloskey E. FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int 2008;19:385–97.

10. Fries JF. The epidemiology of NSAID gastropathy: the ARAMIS experience. J Clin Rheumatol 1998;4 Suppl:S11–6.

11. Fries JF, Williams CA, Bloch DA, Michel BA. Nonsteroidal anti- inflammatory drug- associated gastropathy: incidence and risk factor models. Am J Med 1991;91:213–22.

12. Gabriel SE, Jaakkimainen L, Bombardier C. Risk for serious gastrointestinal complications related to use of nonsteroidal anti- inflammatory drugs: a meta- analysis. Ann Intern Med 1991;115:787–96.

13. Wolfe F. Which HAQ is best? A comparison of the HAQ, MHAQ and RA- HAQ, a difficult 8 item HAQ (DHAQ), and a rescored 20 item HAQ (HAQ20): analyses in 2,491 rheumatoid arthritis patients following leflunomide initiation. J Rheumatol 2001;28:982–9.

14. Cook N, Paynter N, Demler O. Risk prediction modeling: division of preventive medicine. 2018. URL: http://ncook.bwh.harva rd.edu/sas-macros.html.

1232

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1232–1240DOI 10.1002/art.40860 © 2019, American College of Rheumatology

Development and Validation of an 18F- Fluorodeoxyglucose–Positron Emission Tomography With Computed Tomography–Based Tool for the Evaluation of Joint Counts and Disease Activity in Patients With Rheumatoid ArthritisSang Jin Lee,1 Ju Hye Jeong,2 Chang-Hee Lee,2 Byeong-Cheol Ahn,1 Jung Su Eun,2 Na Ri Kim,2 Jong Whan Kang,2 Eon Jeong Nam,1 and Young Mo Kang1

Objective. Clinical joint count assessment is important for detecting synovitis but its reliability is a subject of controversy. This study was undertaken to assess the correlation of positron emission tomography (PET)–derived parameters in 68 joints with disease activity and to compare the reliability of joint counts between PET with computed tomography (CT) and clinical assessment in patients with rheumatoid arthritis (RA).

Methods. We enrolled 91 patients with active RA (69 in a development group and 22 in a validation group) who underwent concurrent 18F- fluorodeoxyglucose (18F- FDG)–PET- CT and clinical disease activity evaluation. PET- derived parameters were compared with disease activity assessed using clinical joint count parameters. A Disease Activity Score (DAS) using counts of PET- positive joints was developed, and then validation studies were performed in an independent group.

Results. The number of PET- positive joints (of 28 and 68 joints) was significantly correlated with the swollen joint count (SJC) and tender joint count (TJC) and the DAS in 28 joints using the erythrocyte sedimentation rate (DAS28- ESR). Intraobserver and interobserver reliability of PET for the affected joint counts were excellent. Interobserver reliability between nuclear medicine physicians and rheumatologists was good for the SJC and TJC in both 28 joints and 68 joints. After multivariate analyses, including ESR and patient’s global assessment of disease activity (PtGA) in addition to PET- derived parameters, the PET/DAS was derived as (0.063 × number of PET- positive joints in 28 joints) + (0.011 × ESR) + (0.030 × PtGA). A significant correlation between the PET/DAS and the DAS28- ESR was confirmed in the validation group (P < 0.001).

Conclusion. PET- CT could serve as a sensitive and reliable method in the evaluation of disease activity in RA patients, and may be applicable as a research tool, particularly in clinical trials.

INTRODUCTION

Rheumatoid arthritis (RA) is a chronic inflammatory disor-der characterized by synovial infiltration of metabolically active immune cells (1). Ultrasound (US) and magnetic resonance imag-ing (MRI) can reveal synovitis (2–4) but cannot be used to assess the activation of immune cells. Recently, 18F- fluorodeoxyglucose (18F- FDG) positron emission tomography (PET) with computed tomography (CT) has been proposed as a method of evaluating

disease activity in patients with RA. Because of the increased glycolytic activity of immune cells, the degree of 18F- FDG accu-mulation is associated with regional inflammation. Moreover, assessments of joint inflammation made using PET- CT have correlated with those made using US and MRI (5,6).

The semiquantitative measurement of FDG uptake in selected joints has been reported to correlate with disease activity in RA patients (5,7,8), as have cumulative standardized uptake values (SUVs) in large joints (9). Further, in another study, the differences

Supported by the Republic of Korea Basic Research Program (National Research Foundation grant 2017R1A2B2008288) and the Ministry of Health and Welfare of the Republic of Korea (grant HI18C2112 of the Korea Health Technology R&D Project).

1Sang Jin Lee, MD, Byeong-Cheol Ahn, MD, PhD, Eon Jeong Nam, MD, PhD, Young Mo Kang, MD, PhD: Kyungpook National University Hospital and Kyungpook National University, Daegu, Republic of Korea; 2Ju Hye Jeong, MD, Chang-Hee Lee, MD, Jung Su Eun, MD, Na Ri Kim, MD, Jong Whan Kang, MD: Kyungpook National University Hospital, Daegu, Republic of Korea.

Drs. Lee and Jeong contributed equally to this work.No potential conflicts of interest relevant to this article were reported.Address correspondence to Young Mo Kang, MD, PhD, Kyungpook

National University Hospital, 130 Dongdeok-ro, Jung-gu, Daegu 41944, Republic of Korea. E-mail: ymkang@knu.ac.kr.

Submitted for publication August 9, 2018; accepted in revised form February 12, 2019.

18F- FDG– PET-CT FOR THE EVALUATION OF DISEASE ACTIVITY IN RA |      1233

in SUV in the hand joints during follow- up reflected therapeutic response (10). However, total joint counts (number of joints from a total of, e.g., 68 that exhibit positive findings by PET) were not evaluated in previous studies, and to date, it is unclear to what extent positive findings in joints as assessed by PET correlate with positive findings in joints as assessed clinically.

Clinical joint count assessment (i.e., number of swollen and tender joints) is considered important in the detection of synovitis, and this is supported by its inclusion in the Disease Activity Score in 28 joints (DAS28) (11) and its use as an outcome measure in clinical trials, research, and practice. However, the reliability of joint count assessments, considering both intraobserver and interobserver findings, is a subject of controversy (12). Although a standardized training protocol described in a previous report was shown to improve the ability to evaluate joint counts, the reliability of SJCs remained variable (13). Therefore, a more objective evalu-ation tool is needed for the assessment of joint counts.

In the present study, we evaluated whether joints (from among 68 joints) shown by PET- CT to be affected correlated with disease activity parameters. Additionally, we sought to determine and compare the reliability, at the individual joint level, of PET- positive joints in relation to joints found to be positive by clinical assessment, in patients with RA.

PATIENTS AND METHODS

Patients and study design. Ninety- one patients who had joints with active RA and underwent 18F- FDG–PET- CT evaluation at Kyungpook National University Hospital from December 2010 through November 2016 were enrolled in the study. The study consisted of 2 groups: a development group (n = 69), in which a DAS was derived by PET- CT, and a validation group (n = 22), in which this DAS was applied. All patients were diagnosed as hav-ing RA according to the 2010 American College of Rheumatology/European League Against Rheumatism criteria (14). At the time of PET- CT evaluation, clinical disease activity was assessed using the swollen joint count (SJC), the tender joint count (TJC), the erythro-cyte sedimentation rate (ESR), the C- reactive protein (CRP) level, and the patient’s global assessment of disease activity (PtGA). The standard DAS28 was calculated based on results of the SJC (of 28 joints) (SJC28), TJC (of 28 joints) (TJC28), ESR, and CRP. SJC28 and TJC28 were assessed clinically in each patient by 3 rheumatolo-gists (JSE, JWK, NRK). PET- CT images were evaluated by 2 nuclear medicine physicians (JHJ and CHL) who were blinded with regard to patient clinical status. The study protocol was approved by the Institutional Review Board at Kyungpook National University Hospi-tal. The study was conducted in concordance with the principles of the Declaration of Helsinki.

F- FDG–PET- CT acquisition protocol. All patients fasted for at least 6 hours, and blood glucose levels were assessed prior to 18F- FDG administration. The blood glucose concentration was con-

firmed to be <150 mg/dl in each patient. 18F- FDG (~5 MBq/kg body weight) was injected intravenously, and patients rested for 1 hour before PET- CT images were acquired. The PET- CT images were obtained from the skull vertex to the feet and performed with patients in a supine position with hands palmar side down on a scanner table using Reveal HiREZ 6- slice CT (CTI Molecular Imaging). First, a low- dose CT scan without contrast enhancement was obtained for attenuation correction, and all images were reconstructed using a 3.75- mm slice thickness at 2.5- mm increments. A 3- dimensional PET scan with a maximum spatial resolution of 6.5 mm was per-formed. PET images were reconstructed with a 128 × 128 matrix, an ordered- subset expectation maximization algorithm with 4 itera-tions and 8 subsets, a Gaussian filter of 5.0 mm, and a slice thick-ness of 2.5 mm.

Image analysis. Two experienced nuclear medicine phy-sicians (JHJ and CHL) interpreted the PET- CT images using a dedicated workstation and software (Advantage Workstation; GE Healthcare). PET- CT images were reviewed in 3 orthogonal planes (axial, coronal, and sagittal), and rotating maximum intensity projection images were collected. Joints were considered posi-tive for synovitis when increased FDG uptake exceeding normal regional tracer accumulation was present in areas presumed to correspond to joint synovium. Volume of interest (VOI) for a PET- positive joint was placed over a joint synovium on PET images, an iso- contour VOI including all voxels >42% of the maximum was then created, and SUVmax value was calculated automatically (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40860/ abstract). The SUVmax was calculated using the following formula:

FDG uptake in 68 joints was evaluated, except that the distal interphalangeal joints of the hands were excluded because of the partial volume effect and the midtarsal joints were not differenti-ated from the ankle joints. Interpretation of the PET- CT images was repeated 2 months later, either by 1 nuclear medicine phy-sician or 2 independent nuclear medicine physicians. PET28, PET68, SUVlarge, SUV28, and SUV68 were defined as follows: PET28 = number of PET- positive joints of a total of 28 joints; PET68 = number of PET- positive joints of a total of 68 joints; SUV-

large = mean SUVmax of large joints (i. e., shoulder, elbow, hip, knee, and ankle); SUV28 = mean SUVmax of 28 joints; SUV68 = mean SUVmax of 68 joints.

Statistical analysis. Baseline clinical data are presented as the mean ± SD. The significance of differences between continuous variables was calculated using the Mann- Whitney test. Correlations between the PET- derived parameters and disease activity were calculated using Pearson’s correlation

SUVmax=Maximumactivity in the region of interest(MBq/ml)

Injecteddose (MBq)/bodyweight(gm)

Maximum activity in the region of interest(MBq/ml)

Injected dose (MBq)/body weight(gm)SUVmax =

LEE ET AL 1234       |

test, with Bonferroni correction. Cumulative frequencies of individual joint involvement as determined by PET- CT findings and by clinical assessments were compared using intraclass correlation coefficients (ICCs).

Intraobserver differences in joint counts (repeat readings by each individual nuclear medicine physician) and interobserver differences (between nuclear medicine physicians or between a nuclear medicine physician and a rheumatologist) were cal-culated by the Cohen’s kappa coefficient, McNemar’s test, and ICCs. A kappa value of 0–0.20 was considered as poor agree-ment, 0.21–0.40 fair, 0.41–0.60 moderate, 0.61–0.80 good, and 0.81–1.00 excellent (15). ICCs between PET- CT results and the clinical assessment of joint counts in the development group were estimated (with 95% confidence intervals [95% CIs]) using a two- way mixed- effects model and the Bland- Altman approach (16).

For the development of the PET/DAS, univariate and multi-variate analyses using the backward stepwise regression method were utilized to evaluate the association between clinical fac-

tors, including PET- derived parameters, and disease activity in patients with RA. After the PET/DAS was generated, it was cal-culated for each patient in the validation group. Pearson’s corre-lation coefficient was used to compare the correlation of the PET/DAS with the DAS28- ESR. P values less than 0.05 were con-sidered significant. All statistical analyses were performed using SPSS version 19, and graphics were generated with GraphPad Prism version 5.

RESULTS

Baseline characteristics in the development and validation groups. Table 1 shows the baseline characteristics of enrolled patients with RA in the development group (n = 69) and validation group (n = 22). The mean ± SD ages of the devel-opment and validation groups at the time of PET- CT evaluation were 67.16 ± 13.8 years and 64.68 ± 14.4 years, respectively. The mean DAS28- ESR was 6.77 ± 1.1 and 6.41 ± 1.0, respec-tively, with all patients in both groups exhibiting moderate- to- high disease activity. In the development group, 54 patients were naive to disease- modifying antirheumatic drugs (DMARDs), and 15 showed inadequate response to DMARDs. Disease activity was not significantly different irrespective of treatment with DMARDs (see Supplementary Figure 1, available on the Arthritis & Rheu-matology web site at http://onlin elibr ary.wiley.com/doi/10.1002/

art.40860/ abstract).

Table 1. Baseline characteristics of the study subjects with RA*

Development group

(n = 69)

Validation group

(n = 22)

Age at PET- CT, years 67.16 ± 13.8 64.68 ± 14.4Age at diagnosis,

years65.26 ± 15.7 64.64 ± 14.4

Female sex, no. (%) 45 (65.2) 15 (68.2)Seropositive (RF

and/or anti- CCP)45 (65.2) 11 (50)

RF, IU/ml 122.56 ± 234.9 77.37 ± 166.7Anti- CCP, units/ml 176.87 ± 208.1 91.56 ± 184.2

ESR, mm/hour 63.51 ± 31.0 69.96 ± 25.1CRP, mg/dl 7.95 ± 5.8 7.21 ± 5.9DAS28- ESR 6.77 ± 1.1 6.41 ± 1.0DAS28- CRP 6.36 ± 1.2 5.68 ± 1.0PtGA 71.24 ± 18.0 64.23 ± 16.9DMARD- naive 54 (78.3) 19 (86.4)Active joint counts

SJC28 12.20 ± 7.4 9.27 ± 5.6TJC28 13.88 ± 7.1 11.09 ± 6.1SJC68 19.12 ± 12.7 12.94 ± 7.5†TJC68 23.90 ± 14.2 16.00 ± 7.1†

* Except where indicated otherwise, values are the mean ± SD. RA = rheumatoid arthritis; PET = positron emission tomography; CT = computed tomography; RF = rheumatoid factor; anti- CCP = anti–cyclic citrullinated peptide; ESR = erythrocyte sedimentation rate; CRP = C- reactive protein; DAS28 = Disease Activity Score in 28 joints; PtGA = patient’s global assessment of disease activity; DMARD = disease- modifying antirheumatic drug; SJC = swollen joint count (in 28 joints or 68 joints); TJC = tender joint count (in 28 joints or 68 joints). † Measured in 16 study subjects in the validation group.

Table 2. Correlation between PET- derived parameters and disease activity assessed clinically in a development group of 69 patients with RA*

PET28 PET68 SUVlarge SUV28 SUV68

ESRρ 0.137 0.123 0.003 0.080 0.063P 0.261 0.316 0.978 0.513 0.607

DAS28- ESRρ 0.546 0.507 0.431 0.565 0.515P <0.001† <0.001† <0.001† <0.001† <0.001†

CRPρ 0.293 0.202 0.383 0.362 0.260P 0.015 0.096 0.001† 0.002† 0.032

DAS28- CRPρ 0.547 0.498 0.513 0.607 0.541P <0.001† <0.001† <0.001† <0.001† <0.001†

* Pearson’s correlation coefficient was used to compare the cor-relation of the PET- derived parameters with clinically assessed disease activity. Standardized uptake values (SUVs) were obtained from large joints (i. e., shoulder, elbow, hip, knee, and ankle joints). The SUVmax was calculated using the formula maximum activity in the region of interest (MBq/ml)/(injected dose [MBq]/body weight [gm]). SUVlarge = mean SUVmax of large joints; SUV28 = mean SUVmax of 28 joints; SUV68 = mean SUVmax of 68 joints (see Table 1 for other definitions). † Correlation remained significant after Bonferroni correction.

18F- FDG– PET-CT FOR THE EVALUATION OF DISEASE ACTIVITY IN RA |      1235

Correlation between PET- derived parameters and disease activity in the development group. To inves-tigate the correlation between PET- derived parameters and disease activity, the number of PET- positive joints and the mean SUVmax of selected joints were compared with disease activity derived by clinical assessment (ESR, DAS28- ESR, CRP levels, and DAS28- CRP). The PET28 was significantly correlated with the DAS28- ESR (ρ = 0.546, P < 0.001) as was the PET68 (ρ = 0.507, P < 0.001). SUVlarge, SUV28, and

SUV68 were also significantly correlated with the DAS28- ESR. These parameters were all significantly correlated with the DAS28- CRP (Table 2).

At the time of PET- CT evaluation, clinical disease activity was assessed using the swollen joint count (SJC), the tender joint count (TJC), the ESR, the CRP level, and the patient’s global assessment of disease activity (PtGA). The standard DAS28 was calculated based on results of the SJC (of 28 joints) (SJC28), TJC (of 28 joints) (TJC28), ESR, and CRP.

Figure 1. Correlation between positive findings in joints assessed by positron emission tomography (PET) and clinical assessment. The number of PET- positive joints in 28 joints (PET28) was significantly correlated with the tender joint count in 28 joints (TJC28) (A), the TJC in 68 joints (B), the swollen joint count in 28 joints (SJC28) (D), and the SJC in 68 joints (E). The number of PET- positive joints in 68 joints (PET68) was also significantly correlated with the TJC68 (C). When the number of affected joints was compared between the PET28 and PET68, a highly significant correlation was noted (F). Each symbol represents an individual data point; dotted lines represent the 95% confidence interval. The correlation coefficients and P values were calculated by Pearson’s correlation test.

LEE ET AL 1236       |

When PET- derived joint counts were compared with clini-cal joint counts, the PET28 was significantly correlated with the TJC28 (ρ = 0.574, P < 0.001) and the TJC68 (ρ = 0.543, P < 0.001) (Figures 1A and B). Although the PET68 was also sig-

nificantly correlated with the TJC68, the correlation coefficient between the PET28 and the TJC68 was not inferior compared with that between the PET68 and TJC68 (Figures 1B and C). These results were also similar between PET- positive joint counts and the SJC (Figures 1D and E). When the PET28 and PET68 were compared, a highly significant correlation was noted (Fig-ure 1F).

Reliability of joint counts between PET- CT and clin-ical assessment in the development group. To clarify why the correlation between the PET28 and TJC68 was noninferior to the correlation between the PET68 and TJC68, affected indi-vidual joints (expressed in terms of cumulative frequencies of involvement) were compared between the 69 patients (see Sup-plementary Figure 2, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40860/ abstract). The most commonly affected joints, assessed by a nuclear medicine physician and rheumatologists, were included in the 28- joint area. The mean cumulative joint count in the 28- joint area was higher than in the 40 remnant–joint area (i. e., joints included in the 68- joint count but not included in the 28- joint count), and the differences in joint counts between PET- CT findings and clinical assessment were less variable in the 28- joint area than in the 40 remnant–joint area. Further, ICC values of reliability for the cumulative frequencies of affected individual joints between PET- CT findings and clinical assessments of the TJC were higher in the 28- joint area (0.895 [95% CI 0.674–0.966]) compared with the 40 remnant–joint area (0.830 [95% CI 0.492–0.943]). The ICC values of PET- CT findings and clini-cal assessments of the SJC, excluding the shoulder joints, were also consistent with those between PET- CT findings and clinical assessments of the TJC.

Kappa values at the individual joint between PET- CT results and clinical assessments were variable, ranging from –0.087 to 0.593. However, kappa values at the individual joint in 28 joints, excluding the shoulder joints, showed constant fair- to- moderate agreement, while those in the 40 remnant joints demonstrated poor agreement. The kappa values of the total 28 joints were higher than those of the total 40 rem-nant joints (Table 3). Moreover, McNemar’s test showed that PET- CT results and the clinical assessments of the TJC for positive joint counts did not significantly differ for the 28 joints (P < 0.730) but differed for the 40 remnant joints in the 68- joint count (P < 0.001). Values of reliability assessed by ICCs at the patient level were good between the PET28 and SJC28 (0.697 [95% CI 0.511–0.813]) and between the PET28 and TJC28 (0.728 [95% CI 0.561–0.832]) (Table  3). The level of reliability of the PET28 in relation to the TJC28 was further illustrated by the Bland- Altman plots (Figure 2). The mean dif-ference between the 2 values was –0.48 with a range of –2.12 to 1.17, and the majority of plots (64 of 69 [92.8%]) were within the upper and lower limits of 2 SD.

Table 3. Interobserver reliability of joint counts between a nuclear medicine physician and rheumatologists*

Interobserver reliabil-ity using the SJC

Interobserver reliabil-ity using the TJC

κ P κ P

PIP1 0.184 0.015 0.217 0.007PIP2 0.536 <0.001 0.593 <0.001PIP3 0.486 <0.001 0.458 <0.001PIP4 0.415 <0.001 0.378 <0.001PIP5 0.427 <0.001 0.471 <0.001MCP1 0.374 <0.001 0.363 <0.001MCP2 0.448 <0.001 0.507 <0.001MCP3 0.301 <0.001 0.316 <0.001MCP4 0.405 <0.001 0.405 <0.001MCP5 0.394 <0.001 0.432 <0.001Wrist 0.293 <0.001 0.283 <0.001Elbow 0.281 0.001 0.320 <0.001Shoulder 0.081 0.140 –0.087 0.304Knee 0.242 <0.001 0.287 <0.001ACJ –0.043 0.593 0.124 0.144Hip – – 0.109 0.125Ankle 0.257 0.001 0.288 <0.001MTP1 0.285 0.001 0.236 0.005MTP2 0.154 0.042 0.122 0.095MTP3 0.044 0.546 0.102 0.128MTP4 0.275 <0.001 0.243 0.001MTP5 0.069 0.410 0.262 0.002f- PIP1 0.072 0.393 0.030 0.719f- PIP2 – – – –f- PIP3 –0.023 0.742 –0.024 0.707f- PIP4 – – – –f- PIP5 –0.021 0.782 –0.024 0.724TMJ 0.031 0.584 0.042 0.605SCJ 0.002 0.971 0.203 0.01328 joints 0.399 <0.001 0.442 <0.00140 joints† 0.254 <0.001 0.293 <0.001

* Interobserver reliability was calculated using the Cohen kappa test and intraclass correlation coefficients (ICCs). For the SJC28 and the SJC68, the ICC was 0.697 (95% confidence interval [95% CI] 0.511–0.813) and 0.622 (95% CI 0.390–0.766), respectively. For the TJC28 and the TJC68, the ICC was 0.728 (95% CI 0.561–0.832) and 0.652 (95% CI 0.438–0.785), respectively. PIP = proximal interphalangeal joint; MCP = metacarpophalangeal joint; ACJ = acromioclavicular joint; MTP = metatarsophalangeal joint; f- PIP = proximal interpha-langeal joint in the foot; TMJ = temporomandibular joint; SCJ = ster-noclavicular joint (see Table 1 for other definitions). † Joints included in the 68- joint count but not in the 28- joint count.

18F- FDG– PET-CT FOR THE EVALUATION OF DISEASE ACTIVITY IN RA |      1237

When intraobserver reliability for the nuclear medicine phy-sician’s findings was calculated, kappa values at the individual joint level showed excellent agreement in the majority of joints, and ICC values at the patient level also showed excellent reli-ability in both 28- joint counts (ICC 0.977 [95% CI 0.906–0.994])

and 68- joint counts (ICC 0.991 [95% CI 0.963–0.998]). Addition-ally, kappa values of the interobserver findings (2 nuclear medi-cine physicians) ranged from good to excellent in the majority of joints, and ICC values were excellent for both the 28- joint counts (ICC 0.989 [95% CI 0.957–0.997]) and the 68- joint counts (ICC 0.980 [95% CI 0.921–0.995]) (see Supplementary Table 2, avail-able on the Arthritis & Rheumatology web site at http://onlin e libr ary.wiley.com/doi/10.1002/art.40860/ abstract).

Generation of the DAS using PET- CT (PET/DAS). For the development of the DAS using PET- CT, linear regression analyses were performed between PET- derived parameters and the DAS28- ESR. PET- derived parameters (PET28, PET68, SUVlarge, SUV28, and SUV68) were all positively associated with the DAS28- ESR. After multivariate analyses (including ESR levels and PtGA findings in addition to PET- derived parameters), PET28, ESR, and PtGA variates were independently associated with the DAS28- ESR (Table  4). Thus, the PET/DAS was gen-erated based on the regression coefficients of the multivariate analyses using the following formula:

PET/DAS = 0.063 × PET28 + 0.011 × ESR + 0.030 × PtGA

Figure 2. Bland- Altman plot showing interobserver reliability of assessment of the number of PET- positive joints in 28 joints (PET28) in relation to the tender joint count in 28 joints (TJC28). “Mean between PET28 and TJC28” = average of the joint count as determined by the PET28 and the joint count as determined by the TJC28. The mean difference between the 2 values was –0.48 (range –2.12, 1.17), and the majority of plots were within the upper and lower limits of 2 SD (lines with asterisks). UCI = upper confidence interval; LCI = lower confidence interval.

Table  4. Association of the DAS28- ESR with PET- derived parameters determined via univariate and multivariate linear regression*

Variate

Univariate analysis Multivariate analysis

β ± SE P β ± SE P

PET28 0.081 ± 0.015 <0.001 0.063 ± 0.012 <0.001PET68 0.051 ± 0.011 <0.001 – –SUVlarge 0.039 ± 0.010 <0.001 – –SUV28 0.032 ± 0.006 <0.001 – –SUV68 0.017 ± 0.004 <0.001 – –ESR 0.015 ± 0.004 <0.001 0.011 ± 0.003 <0.001PtGA 0.036 ± 0.007 <0.001 0.030 ± 0.005 <0.001

* DAS28- ESR = Disease Activity Score in 28 joints using the eryth-rocyte sedimentation rate; PET = positron emission tomography; PtGA = patient’s global assessment of disease activity (see Table 2 for other definitions). PET/DAS = 0.063 × PET28 + 0.011 × ESR + 0.030 × PtGA

LEE ET AL 1238       |

Validation of the PET/DAS in the independent group. In the validation group, 19 patients were naive to treat-ment with DMARDs, with 3 showing inadequate response. Dis-ease activity was not significantly different compared with the development group (Table 1). PET/DAS scores in the validation group were significantly correlated with DAS28- ESR results (ρ = 0.843, P < 0.001) (see Supplementary Figure 3, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40860/ abstract). PET/DAS results were also significantly correlated with the DAS28- CRP, TJC28, and SJC28 (see Supplementary Table 3, available on the Arthri-tis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40860/ abstract).

DISCUSSION

To date, it is unclear to what extent PET- assessed joint counts might be useful as an additional tool for elucidating positive findings of synovitis on clinical joints counts or for assessing the extent of disease activity. The results of our study showed that the affected joints on PET were significantly correlated with disease activity parameters derived by clini-cal assessment, including the SJCs and TJCs. The reliability of both intraobserver and interobserver evaluations of PET- based joint counts were excellent. Moreover, interobserver reliability was good when comparing positive findings of PET- based joint counts in relation to clinical joint counts. Using the regression coefficients obtained from the evaluation of associ-ations between the PET- derived parameters and the DAS28, we developed a novel disease activity formula, the PET/DAS, which was composed using ESR levels, the PtGA, and the PET28. A highly significant correlation between the PET/DAS and the DAS28- ESR was validated in an independent group. To our knowledge, this is the first study to examine the entire set of peripheral joints using PET, followed by the generation of a novel composite parameter (PET/DAS). Importantly, our results indicate that PET- derived parameters could comple-ment clinical assessment of disease activity in patients with RA, and may serve as research tool, particularly in clinical tri-als, where exposure would presumably have defined radiation limits.

PET is capable of identifying the presence and intensity of inflammation in arthritic joints. Previous studies have demon-strated that the quantitative measurement of FDG uptake revealed by PET may be useful for evaluating disease activity and predicting drug therapy response in patients with RA (5,17–19). Moreover, SUVmax at baseline can be a prognostic factor for large joint destruction during follow- up (20,21). However, these studies did not evaluate total joint counts (68 joints) on PET images and measured SUVs at only selected joints; additionally, these studies did not investigate the reliability between PET- derived parameters and clinical assessments.

In our study, ICCs and kappa values in both intraobserver and interobserver evaluations of joints assessed as positive on PET- CT showed greater improvement compared with pre-viously reported reliability parameters for clinical joint counts (22). Despite the important role of joint counts in the evalua-tion of RA, joint counts are not routinely assessed at specialist clinics in many countries due to time constraints on the clini-cians (23). A systematic literature review regarding the clinical assessment of joint counts showed that the intraobserver reli-ability of ICCs generated from clinical assessments performed by healthcare professionals ranged from 0.49 to 0.98 in TJCs and 0.47 to 0.95 in SJCs (22), while reliability of the kappa value at the joint level ranged from fair to good in SJCs (24), reflecting inconsis tent joint assessment in real- world practice. Moreover, the range of interobserver reliability assessed with ICCs depended on the variation among study samples (25,26). Although US and MRI assessments of synovitis are more sen-sitive than clinical examination in the detection of inflammation, they are time- consuming and have definite limitations in sys-temic joint evaluation (27–30). On the contrary, joint counts on PET evaluation are a reproducible method for assessing syn-ovial inflammation, with excellent reliability between readers. Moreover, it is feasible to computerize the measurement of joint counts using PET- CT without the involvement of experienced joint assessors.

Importantly, interobserver reliability with ICCs of positive find-ings in PET- assessed joints in relation to positive findings in clini-cally assessed joints was good when using both the TJC28 and SJC28 as well as when using the TJC68 and SJC68. These val-ues were not inferior to those of positive findings of joints assessed clinically by different physicians (22). These results imply that it is possible for nuclear medicine physicians who are inexperienced in the evaluation of synovitis in patients with RA to count joints with active synovitis. Although the PET28 represents the extent of inflamed joints, with the inclusion of the 68- joint count, composite parameters based on the PET28, which can better explain dis-ease activity in patients with RA, are needed.

Finally, we developed a novel PET/DAS formula derived from PET assessment alone, without using clinical joint assessment. This formula was validated in an independent patient group. The PET/DAS, which may overcome variability of clinical evaluation by utilizing the observations of joint assessors with diverse back-grounds, may be an excellent tool that can complement the use of the DAS28- ESR in the evaluation of patients with RA. More-over, the PET/DAS can be applied to comparative clinical trials and research, which should markedly reduce bias.

Because this study was performed at a single center, a multicenter validation of PET- derived parameters is needed to determine whether our results can be generalized. Further-more, prospective studies are needed to monitor therapeutic efficacy, which could provide strong support for the appli-cation of the PET/DAS in rheumatology practice. Repeated

18F- FDG– PET-CT FOR THE EVALUATION OF DISEASE ACTIVITY IN RA |      1239

PET- CT examinations result in increased radiation exposure and costs in patients with RA, which should be overcome by the development of an ultrasensitive detector and highly effi-cient PET- CT machine.

In conclusion, 18F- FDG–PET- CT can serve as a sensitive and reliable method to complement the clinical evaluation of disease activity in RA. In the near future, we anticipate that the PET/DAS will be measured using the programmed assessment of VOI to determine disease activity (31). Moreover, the incorporation of deep learning from PET images into computer- aided diagnosis is promising for the development of a novel evaluation tool in the assessment of RA.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final ver-sion to be submitted for publication. Dr. Lee had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. S. J. Lee, Y. M. Kang.Acquisition of data. S. J. Lee, Jeong, C.-H. Lee, Ahn, Eun, Kim, J. W. Kang, Nam, Y. M. Kang.Analysis and interpretation of data. S. J. Lee, Y. M. Kang.

REFERENCES 1. Harris ED Jr. Rheumatoid arthritis: pathophysiology and implications

for therapy. N Engl J Med 1990;322:1277–89.

2. Mathew AJ, Danda D, Conaghan PG. MRI and ultrasound in rheumatoid arthritis. Curr Opin Rheumatol 2016;28:323–9.

3. Schmidt WA, Volker L, Zacher J, Schlafke M, Ruhnke M, Gromnica-Ihle E. Colour Doppler ultrasonography to detect pannus in knee joint synovitis. Clin Exp Rheumatol 2000;18:439–44.

4. Palmer WE, Rosenthal DI, Schoenberg OI, Fischman AJ, Simon LS, Rubin RH, et al. Quantification of inflammation in the wrist with gadolinium- enhanced MR imaging and PET with 2- [F- 18]- fluoro- 2- deoxy- D- glucose. Radiology 1995;196:647–55.

5. Beckers C, Ribbens C, Andre B, Marcelis S, Kaye O, Mathy L, et al. Assessment of disease activity in rheumatoid arthritis with 18F- FDG PET. J Nucl Med 2004;45:956–64.

6. Polisson RP, Schoenberg OI, Fischman A, Rubin R, Simon LS, Rosenthal D, et al. Use of magnetic resonance imaging and positron emission tomography in the assessment of synovial volume and glucose metabolism in patients with rheumatoid arthritis. Arthritis Rheum 1995;38:819–25.

7. Goerres GW, Forster A, Uebelhart D, Seifert B, Treyer V, Michel B, et al. F- 18 FDG whole- body PET for the assessment of disease activity in patients with rheumatoid arthritis. Clin Nucl Med 2006;31: 386–90.

8. Kubota K, Ito K, Morooka M, Mitsumoto T, Kurihara K, Yamashita H, et al. Whole- body FDG- PET/CT on rheumatoid arthritis of large joints. Ann Nucl Med 2009;23:783–91.

9. Okamura K, Yonemoto Y, Arisaka Y, Takeuchi K, Kobayashi T, Oriuchi N, et al. The assessment of biologic treatment in patients with rheumatoid arthritis using FDG- PET/CT. Rheumatology (Oxford) 2012;51:1484–91.

10. Elzinga EH, van der Laken CJ, Comans EF, Boellaard R, Hoekstra OS, Dijkmans BA, et al. 18F- FDG PET as a tool to predict the clinical outcome of infliximab treatment of rheumatoid arthritis: an explorative study. J Nucl Med 2011;52:77–80.

11. Prevoo ML, van ‘t Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity scores that include twenty- eight–joint counts: development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995;38:44–8.

12. Sokka T. How should rheumatoid arthritis disease activity be measured today and in the future in clinical care? Rheum Dis Clin North Am 2010;36:243–57.

13. Cheung PP, Dougados M, Andre V, Balandraud N, Chales G, Chary-Valckenaere I, et al. The learning curve of nurses for the assessment of swollen and tender joints in rheumatoid arthritis. Joint Bone Spine 2014;81:154–9.

14. Aletaha D, Neogi T, Silman AJ, Funovits J, Felson DT, Bingham CO III, et al. 2010 rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum 2010;62:2569–81.

15. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–74.

16. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 1979;86:420–8.

17. Okamura K, Yonemoto Y, Okura C, Higuchi T, Tsushima Y, Takagishi K. Evaluation of tocilizumab therapy in patients with rheumatoid arthritis based on FDG- PET/CT. BMC Musculoskelet Disord 2014;15:393.

18. Roivainen A, Hautaniemi S, Mottonen T, Nuutila P, Oikonen V, Parkkola R, et al. Correlation of 18F- FDG PET/CT assessments with disease activity and markers of inflammation in patients with early rheumatoid arthritis following the initiation of combination therapy with triple oral antirheumatic drugs. Eur J Nucl Med Mol Imaging 2013;40:403–10.

19. Chaudhari AJ, Ferrero A, Godinez F, Yang K, Shelton DK, Hunter JC, et al. High- resolution 18F- FDG PET/CT for assessing disease activity in rheumatoid and psoriatic arthritis: findings of a prospective pilot study. Br J Radiol 2016;89:20160138.

20. Suto T, Okamura K, Yonemoto Y, Okura C, Tsushima Y, Takagishi K. Prediction of large joint destruction in patients with rheumatoid arthritis using 18F- FDG PET/CT and disease activity score. Medicine (Baltimore) 2016;95:e2841.

21. Suto T, Yonemoto Y, Okamura K, Okura C, Kaneko T, Kobayashi T, et al. Predictive factors associated with the progression of large- joint destruction in patients with rheumatoid arthritis after biologic therapy: a post- hoc analysis using FDG- PET/CT and the ARASHI (assessment of rheumatoid arthritis by scoring of large- joint destruction and healing in radiographic imaging) scoring method. Mod Rheumatol 2017;27:820–7.

22. Cheung PP, Gossec L, Mak A, March L. Reliability of joint count assessment in rheumatoid arthritis: a systematic literature review. Semin Arthritis Rheum 2014;43:721–9.

23. Scott DL, Antoni C, Choy EH, Van Riel PC. Joint counts in routine practice. Rheumatology (Oxford) 2003;42:919–23.

24. Marhadour T, Jousse-Joulin S, Chales G, Grange L, Hacquard C, Loeuille D, et al. Reproducibility of joint swelling assessments in long- lasting rheumatoid arthritis: influence on Disease Activity Score- 28 values (SEA- Repro study part I). J Rheumatol 2010;37:932–7.

25. Walsh CA, Mullan RH, Minnock PB, Slattery C, FitzGerald O, Bresnihan B. Consistency in assessing the Disease Activity Score- 28 in routine clinical practice. Ann Rheum Dis 2008;67:135–6.

26. Bellamy N, Anastassiades TP, Buchanan WW, Davis P, Lee P, McCain GA, et al. Rheumatoid arthritis antirheumatic drug trials. I. Effects of standardization procedures on observer dependent outcome measures. J Rheumatol 1991;18:1893–900.

27. Naredo E, Bonilla G, Gamero F, Uson J, Carmona L, Laffon A. Assessment of inflammatory activity in rheumatoid arthritis:

LEE ET AL 1240       |

a comparative study of clinical evaluation with grey scale and power Doppler ultrasonography. Ann Rheum Dis 2005;64: 375–81.

28. Salaffi F, Filippucci E, Carotti M, Naredo E, Meenagh G, Ciapetti  A, et al. Inter- observer agreement of standard joint counts in early  rheumatoid arthritis: a comparison with grey scale ultrasonography: a preliminary study. Rheumatology (Oxford) 2008;47:54–8.

29. Boer AC, Boeters DM, van der Helm-van Mil AH. The use of MRI- detected synovitis to determine the number of involved joints for the

2010 ACR/EULAR classification criteria for rheumatoid arthritis: is it of additional benefit? Ann Rheum Dis 2018;77:1125–9.

30. Fujimori M, Kamishima T, Kato M, Seno Y, Sutherland K, Sugimori H, et al. Composite assessment of power Doppler ultrasonography and MRI in rheumatoid arthritis: a pilot study of predictive value in radiographic progression after one year. Br J Radiol 2018;91:20170748.

31. Ulrich EJ, Sunderland JJ, Smith BJ, Mohiuddin I, Parkhurst J, Plichta KA, et al. Automated model- based quantitative analysis of phantoms with spherical inserts in FDG PET scans. Med Phys 2018;45:258–76.

1241

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1241–1251DOI 10.1002/art.40886 © 2019 The Authors. Arthritis & Rheumatology published by Wiley Periodicals, Inc. on behalf of American College of Rheumatology. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Citrullinated Inhibitor of DNA Binding 1 Is a Novel Autoantigen in Rheumatoid ArthritisRay A. Ohara,1 Gautam Edhayan,1 Stephanie M. Rasmussen,1 Takeo Isozaki,2 Henriette A. Remmer,1 Thomas M. Lanigan,1 Phillip L. Campbell,1 Andrew G. Urquhart,3 Jeffrey N. Lawton,3 Kevin C. Chung,3 David A. Fox,1 and Jeffrey H. Ruth1

Objective. To explore the intrinsic role of inhibitor of DNA binding 1 (ID- 1) in rheumatoid arthritis (RA) fibroblast- like synoviocytes (FLS) and to investigate whether ID- 1 is citrullinated and autoantigenic in RA.

Methods. RA patient serum ID- 1 levels were measured before and after infliximab treatment. RA FLS were trans-fected with a clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR- associated protein 9 con-struct targeting ID- 1 to examine the effects of ID- 1 deletion. RA synovial fluid (SF) and homogenized synovial tissue (ST) were immunoprecipitated for ID- 1 and measured for citrullinated residues using an enzyme- linked immuno-sorbent assay and Western blotting. Liquid chromatography tandem mass spectrometry (LC- MS/MS) was performed on in vitro–citrullinated recombinant human ID- 1 (cit–ID- 1) to localize the sites of citrullination. Normal and RA sera and SF were analyzed by immunodot blotting for anti–citrullinated protein antibodies (ACPAs) to cit–ID- 1.

Results. RA patient serum ID- 1 levels positively correlated with several disease parameters and were reduced after infliximab treatment. RA FLS displayed reduced growth and a robust increase in interleukin- 6 (IL- 6) and IL- 8 production upon deletion of ID- 1. ID- 1 immunodepletion significantly reduced the levels of citrullinated residues in RA SF, and citrullinated ID- 1 was detected in homogenized RA ST (n = 5 samples; P < 0.05). Immunodot blot analyses revealed ACPAs to cit–ID- 1 but not to native ID- 1, in RA peripheral blood (PB) sera (n = 30 samples; P < 0.001) and SF (n = 18 samples; P < 0.05) but not in normal PB sera. Following analyses of LC- MS/MS results for citrullination sites and corresponding reactivity in immunodot assays, we determined the critical arginines in ID- 1 for autoantigenicity: R33, R52, and R121.

Conclusion. Novel roles of ID- 1 in RA include regulation of FLS proliferation and cytokine secretion as well as autoantigenicity following citrullination.

INTRODUCTION

Inhibitor of DNA binding 1 (ID- 1) is a nuclear transcrip-tion factor containing a helix- loop- helix domain that it uti-lizes to regulate cell growth and differentiation via selective binding and sequestering of distinct transcription factors. By this method, ID- 1 controls transcriptional activation of target genes. ID- 1 is also known to be actively transcribed in cells exhibiting hyperproliferative responses and is regarded as a marker of cellular self- renewal. Rheumatoid arthritis (RA) syno-vial fluid (SF) contains abundant amounts of ID- 1, and the pri-

mary source is activated RA fibroblast- like synoviocytes (FLS). Once released, ID- 1 acts as a potent inducer of angiogene-sis and also exhibits endothelial progenitor cell chemotactic activity (1), suggesting that ID- 1 may contribute to angiogen-esis and vasculogenesis by independent mechanisms. ID- 1 is packaged into exosomes, which are released from FLS and potentially delivered to other inflammatory cells within RA synovium (2). Although the concept of a secreted nuclear pro-tein may be unconventional, a similar phenomenon occurs in the inflamed joint with DEK, a nuclear protein that functions as a regulator of transcription involved in chromatin architecture

Supported by the Department of Defense (grant PR120641), the NIH (National Institute of Allergy and Infectious Diseases grant UM1-A1-110498), and the Frederick G. L. Huetwell and William D. Robinson Professorship in Rheumatology.

1Ray A. Ohara, BS, Gautam Edhayan, MS, Stephanie M. Rasmussen, BS, Henriette A. Remmer, PhD, Thomas M. Lanigan, PhD, Phillip L. Campbell, BS, David A. Fox, MD, Jeffrey H. Ruth, PhD: University of Michigan Medical School, Ann Arbor; 2Takeo Isozaki, MD, PhD: Showa University School of Medicine, Tokyo, Japan; 3Andrew G. Urquhart, MD, Jeffrey N. Lawton, MD,

Kevin C. Chung, MD, MS: University of Michigan Health System and A. Alfred Taubman Health Care Center, Ann Arbor.

No potential conflicts of interest relevant to this article were reported.Address correspondence to Jeffrey H. Ruth, PhD, University of Michigan

Medical School, Department of Medicine, Division of Rheumatology, 109 Zina Pitcher Drive, 4023 BSRB, Ann Arbor, MI 48109-2200. E-mail: jhruth@med.umich.edu.

Submitted for publication April 18, 2018; accepted in revised form March 7, 2019.

OHARA ET AL 1242       |

and messenger RNA processing. DEK is secreted by mac-rophages, found in exosomes, can be detected in the SF of juvenile arthritis patients, and is both an autoantigen and a potent neutrophil chemotactic factor (3,4).

In the current study, we investigated 3 potential key roles for ID- 1 in RA: first, as a secreted nuclear protein whose levels corre-late with several disease parameters; second, as a regulator of cell proliferation and inflammatory cytokine production by FLS; and third, as a citrullinated autoantigen. We assessed the roles of ID- 1 in FLS by use of clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR- associated protein 9 (Cas9) targeting of ID- 1. Immunoassays were used to measure citrullinated ID- 1 (cit–ID- 1) in synovial tissue (ST). Mass spectrometry was used to define the specific arginines within ID- 1 that are converted to citrulline as well as the identity of specific citrulline residues that render ID- 1 autoantigenic. The results suggest multidimensional contributions of ID- 1 and cit–ID- 1 to the pathogenesis of RA.

PATIENTS AND METHODS

Patient samples. Data were collected from a cohort of 27 RA patients (2009–2012), who met the 1987 American College of Rheumatology (ACR) classification criteria (5) for RA. Twenty- seven serum samples (median patient age 49 years [range 21–76]) were collected from the patients before the initial treatment with inflix-imab. All RA patients were receiving methotrexate and 14 were receiving additional disease- modifying antirheumatic drugs (sul-fasalazine or bucillamine). Fourteen age- and sex- matched healthy volunteers (median age 42.5 years [range 29–55]) were recruited as controls. All specimens were obtained with written informed consent and collected following approval from the Showa Univer-sity Institutional Review Board (IRB). SF samples were obtained from RA patients during arthrocentesis and stored at −80°C in aliquots after centrifugation to remove SF cells. ST samples were obtained from RA patients undergoing total joint replacement and were snap- frozen (liquid nitrogen) in 10% DMSO in fetal bovine serum and stored at −80°C. These samples were obtained, with IRB approval, from RA patients who met the ACR criteria for RA.

Enzyme- linked immunosorbent assay. Patient sam-ples and normal control samples were analyzed using enzyme- linked immunosorbent assay (ELISA) kits for human ID- 1 (MyBioSource) and rheumatoid factor (RF; Alpha Diagnostic International). For the ID- 1 targeting experiments, cell culture supernatants were mea sured with ELISA kits for human inter-leukin- 13 (IL- 13), epithelial neutrophil–activating peptide 78 (ENA- 78/CXCL5), IL- 6, and IL- 8 (R&D Systems). (For detection of citrulline residues by ELISA [cit-ELISA], see Supplementary Methods, on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract.)

CRISPR/Cas9 targeting of ID- 1. CRISPR U6gRNA- pCMV- Cas9- 2A–red fluorescent protein (RFP) plasmid (Sigma- Aldrich)

containing the guide RNA (gRNA) 5′- GAATCATGAAAGTCG CCAGTGG- 3′ was designed to target the human ID- 1 gene. CRISPR human EMX1- s4 positive control plasmid (no. CRISPR11- 1EA; Sigma- Aldrich) and CRISPR plasmid target-ing an irrelevant gene (e.g., THBS1) were used as controls. All transfections were performed by electroporation using Amaxa Nucleofector Technology (Lonza). Transfected cells were sorted for fluorescent marker (RFP or green fluorescent protein [GFP]) via fluorescence- activated cell sorting (FACS) and analyzed for CRISPR/Cas9 activity by Tracking of Indels by Decomposition (TIDE) using Desktop Genetics web tool (6). To verify ID- 1 deple-tion, transfected cells were lysed in radioimmunoprecipitation assay (RIPA) buffer supplemented with EDTA- free protease inhib-itor cocktail (ThermoFisher Scientific) and analyzed by Western blotting. For assays of cell proliferation and cytokine expressions, sorted cells were plated in 96- well plates (6,000–10,000 cells/well) in complete cell culture medium and imaged hourly for 120 hours using the IncuCyte S3 Live Cell Analysis System (Essen Biosciences). Cell culture supernatants were collected at 24 hours postseeding for analysis by ELISA.

Immunoprecipitation. For ID- 1 pulldown experiments, a Direct Immunoprecipitation kit (ThermoFisher Scientific) was used with polyclonal rabbit anti–ID- 1 antibodies (nos. ab170511 and ab192303; Abcam) or with rabbit IgG isotype control (Ther-moFisher Scientific). RA ST (~0.5 cm3) homogenates were pre-pared in ice- cold RIPA buffer supplemented with EDTA- free protease inhibitor cocktail using an electric homogenizer and were centrifuged and filtered (45 μm) to collect the supernatant. For all immunoprecipitation experiments, manufacturers’ kit protocols were followed. Samples were eluted with low- pH elution buffer supplied in the kit and prepared in Laemmli sample buffer for West-ern blotting; flowthroughs were retained for ELISA. All sera, SF, and ST homogenates were incubated with polyclonal goat anti- human IgM (μ- chain specific)–conjugated Agarose (Sigma- Aldrich) at 4°C overnight for removal of RF prior to all immunoassays.

Citrullination of recombinant human ID- 1 protein. Recombinant human ID- 1 protein (rhID- 1) (OriGene Technolo-gies) was citrullinated in vitro using recombinant human pepti-dylarginine deiminase 4 (rhPAD4) enzyme (Cayman Chemical) or rabbit PAD enzyme (Sigma- Aldrich) as previously described (7). The preparation of noncitrullinated ID- 1 (noncit–ID- 1) was per-formed similarly, with cell culture–grade water (Sigma- Aldrich) in place of PAD enzymes.

Western blotting. Citrullinated and noncitrullinated proteins (100 ng each) and IP- eluted samples from RA ST homogenates (30 μl) were prepared in Laemmli sample buffer and were resolved by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS- PAGE) before transfer to nitrocellulose membranes (NCMs) using a semi- dry transfer cell (Bio- Rad).

ID- 1 CRISPR/Cas9 TARGETING AND CITRULLINATION IN RA |      1243

Blots were washed with 0.05% Tween 20 in Tris buffered saline (TBST) between all subsequent steps. After blocking in 5% non-fat dry milk in TBST, the blots were probed with monoclonal rab-bit anti–ID- 1 (1:1,000 dilution) (no. TA310605, clone EPR7098; OriGene Technologies) and then probed with horseradish per-oxidase–linked anti- rabbit IgG (1:1,000 dilution) (no. 7074; Cell

Signaling Technology). SuperSignal West Dura Extended Dura-tion Substrate (ThermoFisher Scientific) was used for detection prior to image acquisition with an Amersham Imager 600 (GE Healthcare Life Sciences). (For detection of citrulline residues by Western blotting, see Supplementary Methods, available at http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract.)

Figure 1. Effect of infliximab treatment on serum inhibitor of DNA binding 1 (ID- 1) levels in patients with rheumatoid arthritis (RA), and clinical correlations. A, ID- 1 levels were elevated in RA serum (n = 27) compared to those in serum from age- and sex- matched normal controls (n = 14). B, Serum ID- 1 levels in patients were decreased at the 12- week time point following infliximab treatment. The red broken line indicates the detection limit of the ID- 1 enzyme- linked immunosorbent assay (7.81 pg/ml). Bars in A and B represent the mean ± SEM. C, ID- 1 levels and clinical parameters were decreased following treatment with infliximab. Each colored line represents a single patient. D, Baseline serum ID- 1 levels correlated with various clinical and laboratory parameters. E, Change in ID- 1 levels from baseline to posttreatment correlated with change in the Simplified Disease Activity Index (SDAI), showing improved SDAI scores with decreased ID- 1 levels (n = 24). * = P < 0.05; ** = P < 0.01. MMP- 3 = matrix metalloproteinase 3; ESR = erythrocyte sedimentation rate; DAS28 = Disease Activity Score in 28 joints; CRP = C- reactive protein; CDAI = Clinical Disease Activity Index. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40886/abstract.

OHARA ET AL 1244       |

In- gel digestion and liquid chromatography tan-dem mass spectrometry (LC- MS/MS). LC- MS/MS was performed by Proteomics and Peptide Synthesis Core at the University of Michigan (Supplementary Methods, http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract).

Immunodot blotting. Samples were dotted onto NCMs at 10 ng/dot and blocked in 5% goat serum (Sigma- Aldrich) in TBS. Blots were incubated in either samples (RA SF or PB serum, normal PB serum; 1:10,000 dilution), nor-mal human IgG control (1 μg/ml) (no. 1001A; R&D Systems),

or control antibodies, including monoclonal mouse anti–ID- 1 (no. ab66495, clone 2456C1a; Abcam) and polyclonal rab-bit anti- PAD4 (1 μg/mL) (no. ab50247; Abcam), then probed with peroxidase AffiniPure goat anti- human IgG (1 μg/ml, dilu-tion 1:5,000) (no. 109035003; ImmunoResearch). To verify specificity for cit–ID- 1 reactivity of RA specimens, additional control blots were included on the same NCMs using various proteins including noncit–ID- 1, bovine serum albumin (BSA), cit- BSA, ENA- 78, cit–ENA- 78, and rhPAD4. Densitometry analysis was performed using ImageJ (National Institutes of Health).

Figure 2. Transfection of RA fibroblast- like synoviocytes (FLS) with clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR- associated protein 9 (Cas9) plasmid with guide RNA targeting the ID- 1 gene. A, Fluorescence- activated cell sorting of RA FLS that were transfected with ID- 1 CRISPR/Cas9 plasmid or control THBS1 CRISPR/Cas9 plasmid is shown. B, RA FLS transfected with ID- 1 CRISPR/Cas9 plasmid (n = 9) or THBS1 control plasmid (n = 19) were assayed for cell proliferation using the IncuCyte S3 Live- Cell Analysis System. C, Due to imperfect repair by Cas9 nuclease, DNA in the cell pool contained a mixture of indels, yielding a composite sequence trace after the break site (black dashed line). An overview of Tracking of Indels by Decomposition algorithm and output is shown, consisting of visualization of aberrant sequence signal in controls (black) and treated samples (green). Cas9 nuclease cut the genome, and the cell repaired the damage caused by nonhomologous end–joining resulting in aberrant sequences. D, RA FLS transfected with sham control (n = 17 or more per experiment), green fluorescent protein (GFP) control (n = 5 per experiment), or ID- 1 CRISPR/Cas9 plasmid (n = 5 or more per experiment) were sorted and cultured for 24 hours for cytokine expression analysis using enzyme- linked immunosorbent assays. Values are the mean ± SEM. * = P < 0.05. IL- 13 = interleukin- 13 (see Figure 1 for other definitions).

ID- 1 CRISPR/Cas9 TARGETING AND CITRULLINATION IN RA |      1245

Statistical analysis. GraphPad Prism was used for all statistical analyses. Group differences in serum ID- 1 levels were evaluated by unpaired nonparametric Mann- Whitney U test or paired nonparametric Wilcoxon signed rank test. Correlations between serum ID- 1 levels and clinical parameters were as-sessed by nonparametric Spearman’s rank correlation. Results of cell proliferation assay and ELISAs used to detect cytokine expression were analyzed by unpaired parametric t- test. Lev-els of total citrullinated antigens in RA SF depleted of ID- 1 were evaluated by paired nonparametric Wilcoxon signed rank test. Results are expressed as the mean ± SEM. Two- tailed P values less than 0.05 were considered significant.

RESULTS

Correlations between serum ID- 1 levels and dis-ease parameters, and effect of infliximab treatment. Serum ID- 1 levels were measured in a Japanese cohort of 27 RA patients before and after infliximab treatment. We found that serum ID- 1 levels were significantly elevated compared to those in age- and sex- matched normal controls (Figure  1A). Serum ID- 1 levels before and 12 weeks after initiation of infliximab were compared. We observed a significant decrease in ID- 1 levels after infliximab treatment (Figure 1B). Additionally, several clini-cal and laboratory parameters were measured, including matrix metallopeptidase 3 (MMP- 3) level, erythrocyte sedimentation rate (ESR), the Disease Activity Score in 28 joints using ESR

(DAS28- ESR) (8), and the Simple Disease Activity Index (SDAI) (9), with significant improvements observed after infliximab treat-ment (Figure 1C). Next, baseline serum ID- 1 levels were analyzed for correlation with disease parameters including RF, ESR, C- re-active protein (CRP) level, MMP- 3 level, SDAI, Clinical Disease Activity Index (CDAI) (9), DAS28 using CRP level (DAS28- CRP), and DAS28- ESR. We found positive correlations between ID- 1 levels and all of these parameters except RF (Figure 1D). Anal-ysis using the change in SDAI score of responders and nonre-sponders after infliximab treatment showed that the serum ID- 1 level correlated significantly with reduction in disease activity in the responders (24 of 27 patients) (Figure 1E).

In vitro targeting of ID- 1 in RA FLS with the CRISPR/Cas9 system. We have previously shown that ID- 1 is up- regulated in RA synovium and that soluble ID- 1 exhibits inflam-matory and angiogenic properties (1,2). To examine the effects of ID- 1 targeting in RA FLS by CRISPR/Cas9, cell proliferation assays were performed using the IncuCyte S3 Live Cell Analysis System. Results of ID- 1 gene targeting were compared to those from the targeting of an irrelevant gene, THBS1. We achieved a maximum efficiency of 24.8% (Figure  2A), which was sufficient (due to an ample starting number of cells) for downstream experi-ments that used only the sorted RFP- positive cells. We found sig-nificant reductions in FLS growth beginning 13 hours after plating in culture (Figure 2B). To fully verify active and accurate genome editing by Cas9, transfected cells were sorted via FACS, and

Figure 3. Immunodepletion of inhibitor of DNA binding 1 (ID- 1) and detection of citrullinated ID- 1 (cit–ID- 1) in rheumatoid arthritis (RA) synovial fluid (SF) and synovial tissue (ST). A, Immunodepletion of ID- 1 in RA SF (n = 5) using anti–ID- 1 polyclonal antibody (pAb) was verified by enzyme- linked immunosorbent assay. B, Levels of total citrullinated antigens were measured in ID- 1–depleted RA SF. Significant reduction in total citrullinated antigens after ID- 1 depletion suggested that ID- 1 is present in citrullinated forms in RA SF. C, Total citrullinated protein concentration in isotype control versus in anti–ID- 1 antibody–treated RA SF was measured. Bars show the mean ± SEM. D, Detection of citrullinated forms of ID- 1 in RA ST is shown. Immunoprecipitation (IP) was performed on RA ST homogenates using anti–ID- 1 antibody, and blots were probed for ID- 1 or total citrullinated antigens. Cit–ID- 1 was detected in 2 of 3 RA ST homogenates. * = P < 0.05. WB = Western blotting; AMC = antimodified citrulline. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40886/abstract.

OHARA ET AL 1246       |

genomic DNA was isolated and sequenced. We then analyzed for aberrant sequences at the target site caused by indels of nonho-mologous end–joining (NHEJ) using TIDE analysis (Figure 2C). We observed aberrant sequences and corresponding indel frequen-cies with an overall efficiency of 4.4% at the P value threshold of 0.001 with an R2 of 0.99. To verify ID- 1 depletion, Western blot analysis was performed on the sorted cells (Supplementary Figure 1, http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract).

Decreased cellular proliferation and increased IL- 6 and IL- 8 production resulting from deletion of ID- 1 in RA FLS. Sorted cells were cultured for 24 hours following over-night serum starvation, and supernatants were analyzed for cytokine expression (IL- 13, ENA- 78/CXCL5, IL- 6, and IL- 8) by ELISA (Figure 2D). RA FLS transfected with GFP control plasmid served as experimental control. We found significantly increased production of IL- 6 and IL- 8 compared to the controls, with no significant changes in IL- 13 or ENA- 78/CXCL5. RA FLS do not secrete appreciable amounts of IL- 13, which was measured as a control cytokine in these experiments. RA FLS do spontaneously secrete ENA- 78/CXCL5, as well as IL- 6 and IL- 8, but no sub-stantial changes in ENA- 78/CXCL5 levels were seen, indicating that the increases in IL- 6 and IL- 8 are specific effects of CRISPR/Cas9 targeting of the ID- 1 gene.

Detection of cit–ID- 1 in RA SF and ST. In the 2- step strategy to discover cit–ID- 1, we first performed immunoprecip-itation on RA SF using polyclonal antibodies independently to pull out both native and citrullinated forms of ID- 1. Before proceeding to the second step, effective depletion of ID- 1 from RA SF was verified using ELISA (Figure 3A). We then conducted an indirect ELISA on the depleted RA SF to detect total citrullinated anti-gens using antimodified citrulline (AMC) antibody (Supplementary Methods, http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract). Depletion of ID- 1 significantly reduced the level of the total citrullinated antigens in RA SF (Figure 3B). This suggests that the anti–ID- 1 antibody that was used recognized both forms of ID- 1, and that cit–ID- 1 is present in RA SF. The data also showed that total citrullinated protein concentration was significantly higher in isotype controls versus anti–ID- 1 antibody–treated SF samples (Figure 3C). Next, ID- 1 immunoprecipitated from RA ST homoge-nates (using the same anti–ID- 1 antibody) was analyzed by West-ern blotting using an AMC antibody (Supplementary Methods, http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract). We detected a single prominent band (or complex of adjacent bands) when probing either for ID- 1 directly with an anti–ID- 1 anti-body or for cit–ID- 1 indirectly with an AMC antibody. Two of the 3 samples tested contained citrullinated forms of ID- 1 (Figure 3D).

In vitro citrullination of rhID- 1 protein. We incubated rhID- 1 with rhPAD4 at various enzyme:substrate molar ratios in appropriate reaction buffer. To control for the potential confound-

ing effects of conditions and reaction buffer, parallel aliquots of rhID- 1 were subjected to the same conditions except rhPAD4 was replaced with sterile water. To verify whether rhID- 1 was citrul-linated, we performed Western blotting and found that cit–ID- 1 was recognized by the AMC antibody, while noncit–ID- 1 was not recognized, as expected (Figure 4A). As a control, we also probed the samples using an anti–ID- 1 antibody and confirmed that non-cit–ID- 1 was recognized (Figure 4A); cit–ID- 1 was also recognized with this antibody, suggesting that citrullination did not alter its

Figure 4. In vitro citrullination of recombinant human ID- 1 (rhID- 1) by recombinant human anti–peptidylarginine deiminase 4 (rhPAD4). A, Citrullination of rhID- 1 by rhPAD4 was verified by Western blotting. Representative blots of cit–ID- 1 (lane 1) and noncit–ID- 1 (lane 2) were probed with anti–ID- 1 or AMC antibodies. Cit–ID- 1 exhibited multiple forms corresponding to the degree of modification and extended noticeably higher than noncit–ID- 1, likely due to the increased hydrophobicity and the change in charge from the citrullination reaction. B, Citrullination of rhID- 1 by rhPAD4 was verified via enzyme- linked immunosorbent assay (ELISA). Standard indirect ELISA protocol was modified to incorporate an acidic modification step for the AMC antibody as described. Cit–ID- 1 and noncit–ID- 1 can be distinguished by AMC antibody. Cit–ID- 1 produced an OD equivalent to ~300 ng/ml of citrullinated bovine serum albumin (BSA), which was used as a relative standard due to its abundant modifiable arginines. Values are the mean ± SEM. See Figure 3 for other definitions. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40886/abstract.

ID- 1 CRISPR/Cas9 TARGETING AND CITRULLINATION IN RA |      1247

epitope and that both forms can be recognized by a single anti-body. In addition, we performed ELISAs using the AMC antibody on cit–ID- 1 for further confirmation of citrullination. We observed a markedly higher absorbance in the cit–ID- 1 sample as compared to that of the noncit–ID- 1 sample (Figure 4B). Citrullinated BSA (cit- BSA) was used as a positive control for citrullination and as a relative standard for this ELISA.

LC- MS/MS identification of specific citrullines in in vitro–citrullinated rhID- 1. Human ID- 1 contains 10 arginines (Figure 5A), all of which are potential candidates for citrullination by PAD enzymes. We found that cit–ID- 1 extends visibly higher than noncit–ID- 1 on a gel (Figure 5B), likely due to the increase in hydro-phobicity and the change in charge from positive to neutral due to the citrullination reaction. Subsequently, the samples were analyzed by LC- MS/MS to detect the site(s) of citrullination based on the tandem MS fragmentation pattern and the expected neutral loss of isocyanic acid that is diagnostic for citrullination in MS (10). We found multiple citrullinated arginines, varying in number depending on the batch of rhID- 1 and the experimental conditions of in vitro cit-rullination. Arginine R121 was consistently citrullinated by rhPAD4,

as shown in a representative spectrum (Figure  5C). Interestingly, LC-MS/MS analysis identified native citrulline residues in the noncit–ID- 1 samples, which were expected from the production of rhID- 1 in HEK 293T cells. However, with Western blot analysis using an AMC antibody these citrulline residues were not recognized (Fig-ure 4A), suggesting that LC-MS/MS analysis is a more sensitive technique to identify low- level constitutive citrullination.

Measurement of ACPA reactivity with cit–ID- 1 in normal and RA PB sera and RA SF. We found positive ACPA reactivity against cit–ID- 1 but not against noncit–ID- 1, in RA PB sera and SF (Figure 6A). All antigens, including control antigens (noncit–ID- 1, BSA, cit- BSA, ENA- 78, citENA- 78, as well as rhPAD4 [data not shown]), were dotted in triplicate on the same blot and probed together in the same patient sample to ensure citrulline- specific binding by the ACPAs. All control blots showed no significant evidence of antibody binding but did show pos-itivity for cit–ID- 1 (Supplementary Figure 2, http://onlin elibr ary.wiley.com/doi/10.1002/art.40886/ abstract). Normal sera were used as controls, and no reactivity was observed. To control for nonspecific serum antibody binding, human IgG was used at a

Figure 5. Liquid chromatography tandem mass spectrometry analysis (LC- MS/MS) of in vitro citrullinated recombinant human ID- 1 (rhID- 1). A, Sequence of human ID- 1 isoform 1 is shown. Human ID- 1 contains 10 arginines (R, highlighted red). RhID- 1 was citrullinated in vitro by recombinant human anti–peptidylarginine deiminase 4 (rhPAD4) and analyzed via LC- MS/MS. B, Cit–ID- 1 and noncit–ID- 1 were run on sodium dodecyl sulfate–polyacrylamide gel electrophoresis and stained with “blue- silver” Coomassie for in- gel digestion prior to LC- MS/MS. C, The annotated mass spectrum of tryptic peptide confirmed citrullination of R121 in rhID- 1. All detected ions of the peptide are shown in black in the sequence and are annotated in the spectrum. The red r denotes citrulline. Ions highlighted in red in the spectrum show the neutral loss of isocyanic acid (HNCO; ~43 daltons), diagnostic marker ions for citrullination. See Figure 3 for other definitions. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40886/abstract.

OHARA ET AL 1248       |

relatively high dilution (1 μg/ml). Furthermore, all additional con-trol antibodies were negative (data not shown). By densitometry analysis of immunodot blotting, we found that 40% of RA sera recognized cit–ID- 1 versus no control sera. SF from 16.7% of RA patients tested positive for anti–cit–ID- 1 (Figure 6B).

Identification of the critical arginines conferring autoantigenicity to cit–ID- 1. Immunodot blot analysis for ACPA reactivity showed that cit–ID- 1 displayed unique autoan-tigenicity and autoantibody reactivity, depending on the citrul-lination pattern. Specifically, samples that showed reactivity with batches 1 and 2 did not shown any reactivity with batch 3, likely due to the lack of key epitopes. Thus, a batch was desig-nated ACPA positive if it showed reactivity with any of the limited patient samples. We did not observe any ACPA binding to the baseline- modified arginines that were present in rhID- 1 obtained from the vendor. Using a series of LC- MS/MS and immunodot blot analyses, we identified key arginines in rhID- 1 that may be autoantigenic targets for ACPA development in RA. Of the 10 available arginines in rhID- 1, the critical arginines for ACPA reac-tivity were located at positions R33, R52, and R121 (Figure 6C).

To further study the role of these arginines, citrullinated peptides (12 amino acids in length) spanning these regions were tested but were negative for ACPA binding (data not shown). Moreo-ver, denaturing of cit–ID- 1 before immunodot blotting negated its reactivity (Figure 6A). Thus, the critical arginines at R33, R52, and R121 may control conformational epitopes that render cit–ID- 1 antigenic in vivo. These data corroborate our observation that multiple forms of ID- 1 are expressed in vitro and in vivo, but only certain modified forms bind ACPAs (Figure 4A).

DISCUSSION

As a nuclear protein that alters the activity of many transcription factors, ID- 1 appears to affect multiple cellular properties includ-ing proinflammatory cytokine expression. This would place ID- 1 in a strategic position to regulate chronic inflammatory responses directly by inhibition of cytokine production at the transcriptional level. Such regulatory activity could profoundly influence the sever-ity and progression of inflammatory outcomes in chronic diseases including RA. There is mounting evidence that permanently altered FLS function is the result of somatic mutations in key genes that

Figure 6. Detection of anti–citrullinated protein antibodies (ACPAs) to cit–ID- 1 RA samples. A, RA SF, RA sera, and normal (NL) sera were assayed for autoantibodies to cit–ID- 1 by immunodot blotting. ACPA reactivity with cit–ID- 1, but not native ID- 1, was found in RA SF and peripheral blood (PB), but not in normal PB. Human IgG (1 μg/ml) was used as a control and showed no reactivity. Additionally, ACPAs from RA PB did not bind when cit–ID- 1 was boiled (denatured). Control dots showed no significant evidence of ACPAs or antibody binding reactivity in RA SF, RA PB, or normal PB. B, Analysis of immunodot blots via densitometry showed higher reactivity with cit–ID- 1 than noncit–ID- 1 in RA PB (n = 30 samples from 10 patients) and in RA SF (n = 18 samples from 6 patients), but not in normal PB (n = 12 samples from 4 patients). Bars show the mean ± SEM. Dotted line represents the cutoff value used to determine positive reactivity (red dots) in experimental samples. Values were derived from the highest white point in healthy control samples. * = P < 0.05; *** = P < 0.001. C, Analysis of the roles of individual citrulline residues in ACPA reactivity of in vitro citrullinated rhID- 1 is shown. Arginines R33, R52, and R121 were critically necessary for ACPA reactivity. C denotes citrullines; # denotes native citrullines from the production of rhID- 1 in HEK 293 cells. BSA = bovine serum albumin (see Figure 3 for other definitions). Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40886/abstract.

ID- 1 CRISPR/Cas9 TARGETING AND CITRULLINATION IN RA |      1249

regulate the FLS cell cycle, proliferation, and apoptosis (11,12). It has also been suggested that RA synoviocytes have characteristics similar to those of tumor cells, as a number of oncogenes involved in cell cycle regulation—or those that act as transcription factors, such as c- Fos, c-ras, c-raf, c-myc, and c-myb—are expressed at high levels in RA FLS (12). We explored the possibility that another nuclear regulatory protein, namely ID- 1, plays a central role in RA pathogenesis, independent of tumor necrosis factor, both by reg-ulating cytokine secretion and as an inflammatory protein that can undergo posttranslational modifications.

We successfully transfected primary RA FLS with a plasmid containing a CRISPR/Cas9 construct, demonstrated successful ID- 1 gene targeting using a TIDE algorithm and output analy-sis (6), and confirmed these findings by FACS and Western blot analysis. We found substantial increases in IL- 6 (30- fold) and IL- 8 (50- fold) but not in ENA- 78/CXCL5 in supernatants from transfected FLS, compared to sham- transfected control FLS. Using a cell imaging system, we also found that FLS deleted of ID- 1 showed a >20% sustained reduction in proliferation. This may be due in part to elevated production of IL- 6, a cytokine known to inhibit fibroblast proliferation (13). More likely, a perma-nent mutation in an important nuclear regulatory protein that is critical for cell proliferation may have altered the FLS population into a phenotype capable of elevated proinflammatory cytokine secretion. Notably, it has been shown that ID- 1 antisense RNA prevents early- passage fibroblasts from entering the S phase of the cell cycle (14). Firestein et  al proposed a model suggest-ing a duality of FLS populations labeled “passive responders” and “transformed aggressors” (12), which arose partly from the combination of a highly inflamed environment in the RA joint and somatic mutations. Thus, both the data from the current study and previously reported findings suggest that FLS proliferation and heightened FLS secretion of IL- 6 represent distinct stages of the contributions of FLS to the pathogenesis of RA. Moreover, the current findings indicate that ID- 1 can mediate transition of FLS between these two important pathogenic phases.

Analysis of clinical specimens revealed that soluble ID- 1 is present and up- regulated in the serum of RA patients and shows a significant positive correlation with a number of disease parame-ters including ESR, CRP level, MMP- 3 level, SDAI, CDAI, DAS28- CRP, and DAS28- ESR. This indicates that circulating serum ID- 1 levels could be a potential biomarker for RA severity. Additionally, because serum ID- 1 concentration does not correlate with RF titer, it is possible to identify patients with RF- negative RA whose disease will be severe, by measurement of elevated ID- 1. Further-more, after 12 weeks of infliximab, serum levels of ID- 1 showed significant reductions. We found that 24 of 27 patients responded to infliximab (based on the SDAI) and that reduction in the level of ID- 1 significantly correlated with reduction in disease activity in the responders. Because of the elevated circulating concentrations of ID- 1, we surmised that citrullinated forms of ID- 1 could be pres-ent and immunogenic in RA patients, as autoantibodies to citrul-

linated proteins are well- known disease- associated phenomena in RA (15–19).

ACPAs are implicated in RA pathogenesis in synergy with smoking, an environmental risk factor for RA, and the shared epitope major histocompatibility complex allele (20). However, the full range of citrullinated autoantigens in RA is not yet defined. We showed that ID- 1 can be citrullinated in vitro. Moreover, by depleting ID- 1 from RA SFs we reduced the total amount of cit-rullinated proteins detected by as much as 64% (mean 33%) as measured by cit- ELISA. We also found, by immunoblotting, that a subset of RA patients had high- titer autoantibodies to cit–ID- 1. Therefore, cit–ID- 1 in RA SF may account for a significant portion of citrullinated proteins that are the targets of the ACPA response in some RA patients.

Next, we performed immunodot blotting with RA and nor-mal PB sera, as well as RA SF, against various citrullinated and noncitrullinated proteins. We were able to confirm the presence of ACPAs to cit–ID- 1 in both RA sera and SF but not in the PB from healthy individuals, which supports our hypothesis that citrullination of ID- 1 increases its autoantigenicity in RA. Experi-ments using immunodot blotting for ID- 1 and cit–ID- 1 revealed that antibodies to cit–ID- 1 can be detected in RA sera and SF. In SF, the presence of large amounts of cit–ID- 1 could sequester anti–cit–ID- 1 in immune complexes, which could then incorpo-rate RF and therefore be undetectable by the cit- ELISAs. Such mechanisms could account for the relatively low frequency of positivity for anti–cit–ID- 1 in SF samples.

Various citrullinated forms of ID- 1 may function differently as agonists or autoantigens, complicating the analysis of cit–ID- 1 activity. This is because citrullination reactions with the PAD enzyme are notoriously inconsistent, resulting in differences in the number and patterns of arginines that are converted to citrullines. These inconsistencies make it difficult to determine the extent to which citrullination of ID- 1 leads to alterations in its activity. Indeed, we have found that the number and locations of arginines citrullinated in ID- 1 can change by simply altering the source of the PAD enzyme used in the reaction mix. Moreover, many of the mechanistic roles of ACPAs remain unknown. Notably, Schett et al reported that autoantibody to citrullinated vimentin directly induces bone loss, providing a mechanism for osteopenia in early or preclinical RA (21,22). In the case of the chemokine IL- 8/CXCL8, only 1 of 3 arginines was citrullinated, yet this resulted in alteration of function (23). Another example in which citrullina-tion of a single arginine by these methods resulted in functional changes occurred in stromal cell–derived factor 1 CXCL12 (24).

We found that some forms of cit–ID- 1 can be highly reactive with ACPAs formed in RA but that less deiminated forms do not retain autoantigenicity. It appears that the degree of citrullina-tion of ID- 1 may alter the folding pattern and immune properties of ID- 1, leading to autoantigenicity. We have previously inves-tigated the autoantigenicity of citrullinated ENA- 78/CXCL5, which is also highly up- regulated in RA (7). Similar to cit–ID- 1,

OHARA ET AL 1250       |

citrullination of ENA- 78/CXCL5 at arginine R48 enabled binding to ACPA in RA sera and SF in our assays. Furthermore, we previously showed that citrullination of ENA- 78/CXCL5 induced a functional change in the protein from a neutrophil chemoat-tractant to a monocyte chemoattractant (7). Since citrullination of a protein with only 2 arginines, such as ENA- 78/CXCL5, can cause a profound increase in autoantigenicity and change in function, the impact citrullination can have on ID- 1, which con-tains 10 arginines, could be very substantial.

Through detailed LC-MS/MS analysis, we observed conver-sions of ID- 1 in all but 1 arginine at R103, perhaps explained by structural unavailability or PAD preference for certain arginines. As an example, the annotated fragmentation spectrum presented in Figure 5 shows the ions corresponding to the partial sequence of ID- 1 around arginine 121. These ions show the neutral loss of iso-cyanic acid resulting from the fragmentation of the ureido group in citrullines, which is a marker for citrullination, confirming the modi-fication of ID- 1. Through a series of citrullination reactions followed by LC-MS/MS, we identified that the modifications at arginines R33, R52, and R121 in cit–ID- 1 can bind to and perhaps induce ACPAs. However, small peptides spanning these regions were not reactive with RA serum, and boiling cit–ID- 1 before immunodot blot assays negated binding by RA sera. Overall, the data is most consistent with a model that involves recognition by ACPA of con-formational, but not linear, autoantigen epitopes on cit–ID- 1.

We previously reported that ID- 1 is secreted by inflamma-tory FLS and is an angiogenic mediator (1,2). It is possible that free ID- 1 is citrullinated either in FLS and/or extracellularly in the inflamed RA joint to render it autoantigenic in RA tissues. We demonstrated a potential role of ID- 1 in transforming FLS into a pathogenic phenotype and identified cit–ID- 1 as a novel autoantigen candidate in RA. Further investigation is needed to determine whether and how citrullination of ID- 1 may alter its functions. Potentially, ID- 1, cit–ID- 1, and/or ACPAs to cit–ID- 1 may serve as promising therapeutic targets or biomarkers in RA.

ACKNOWLEDGMENTS

Mass spectrometry analyses were conducted in the Mass Spectrometry Core Laboratory at the University of Texas Health Science Center at San Antonio under the direction of Dr. Susan T. Weintraub. The expert technical assistance of Sammy Pardo is gratefully acknowledged. Support for purchase of the Orbitrap mass spectrometer was provided by NIH grant 1S10RR025111- 01 (STW). The authors would also like to thank Dr. Brian R. Hallstrom, MD, for generously providing synovial specimens that were critical for the completion of this study.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti-cally for important intellectual content, and all authors approved the final version to be published. Mr. Ohara had full access to all of the data in

the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Ohara, Edhayan, Rasmussen, Isozaki, Remmer, Lanigan, Campbell, Fox, Ruth.Acquisition of data. Ohara, Edhayan, Rasmussen, Isozaki, Urquhart, Law-ton, Chung, Fox.Analysis and interpretation of data. Ohara, Edhayan, Rasmussen, Isozaki, Remmer, Lanigan, Campbell, Fox, Ruth.

REFERENCES 1. Isozaki T, Amin MA, Arbab AS, Koch AE, Ha CM, Edhayan G, et al.

Inhibitor of DNA binding 1 as a secreted angiogenic transcription factor in rheumatoid arthritis. Arthritis Res Ther 2014;16:R68.

2. Edhayan G, Ohara RA, Stinson WA, Amin MA, Isozaki T, Ha CM, et al. Inflammatory properties of inhibitor of DNA binding 1 secret-ed by synovial fibroblasts in rheumatoid arthritis. Arthritis Res Ther 2016;18:87.

3. Mor-Vaknin N, Punturieri A, Sitwala K, Faulkner N, Legendre M, Khodadoust MS, et al. The DEK nuclear autoantigen is a secreted chemotactic factor. Mol Cell Biol 2006;26:9484–96.

4. Mor-Vaknin N, Kappes F, Dick AE, Legendre M, Damoc C, Teitz-Tennenbaum S, et al. DEK in the synovium of patients with juvenile idiopathic arthritis: characterization of DEK antibodies and posttranslational modification of the DEK autoantigen. Arthritis Rheum 2011;63:556–67.

5. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24.

6. Brinkman EK, Chen T, Amendola M, van Steensel B. Easy quantita-tive assessment of genome editing by sequence trace decomposi-tion. Nucleic Acids Res 2014;42:e168.

7. Yoshida K, Korchynskyi O, Tak PP, Isozaki T, Ruth JH, Campbell PL, et al. Citrullination of epithelial neutrophil–activating peptide 78/CXCL5 results in conversion from a non–monocyte- recruiting chemokine to a monocyte- recruiting chemokine. Arthritis Rheumatol 2014;66:2716–27.

8. Prevoo ML, van ‘t Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity scores that include twenty- eight–joint counts: development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995;38:44–8.

9. Aletaha D, Smolen J. The Simplified Disease Activity Index (SDAI) and the Clinical Disease Activity Index (CDAI): a review of their usefulness and validity in rheumatoid arthritis. Clin Exp Rheumatol 2005;23 Suppl 39:S100–8.

10. Hao G, Wang D, Gu J, Shen Q, Gross SS, Wang Y. Neutral loss of isocyanic acid in peptide CID spectra: a novel diagnostic marker for mass spectrometric identification of protein citrullination. J Am Soc Mass Spectrom 2009;20:723–7.

11. Han Z, Boyle DL, Shi Y, Green DR, Firestein GS. Dominant- negative p53 mutations in rheumatoid arthritis. Arthritis Rheum 1999;42:1088–92.

12. Bartok B, Firestein GS. Fibroblast- like synoviocytes: key effector cells in rheumatoid arthritis. Immunol Rev 2010;233:233–55.

13. Nishimoto N, Ito A, Ono M, Tagoh H, Matsumoto T, Tomita T, et al. IL- 6 inhibits the proliferation of fibroblastic synovial cells from rheu-matoid arthritis patients in the presence of soluble IL- 6 receptor. Int Immunol 2000;12:187–93.

14. Hara E, Yamaguchi T, Nojima H, Ide T, Campisi J, Okayama H, et al. Id- related genes encoding helix- loop- helix proteins are required for G1 progression and are repressed in senescent human fibroblasts. J Biol Chem 1994;269:2139–45.

ID- 1 CRISPR/Cas9 TARGETING AND CITRULLINATION IN RA |      1251

15. Clavel C, Nogueira L, Laurent L, Iobagiu C, Vincent C, Sebbag M, et al. Induction of macrophage secretion of tumor necrosis factor α through Fcγ receptor IIa engagement by rheumatoid arthritis- specific autoantibodies to citrullinated proteins complexed with fibrinogen. Arthritis Rheum 2008;58:678–88.

16. Sokolove J, Zhao X, Chandra PE, Robinson WH. Immune complexes containing citrullinated fibrinogen costimulate macrophages via Toll- like receptor 4 and Fcgγ receptor. Arthritis Rheum 2011;63:53–62.

17. Pratesi F, Dioni I, Tommasi C, Alcaro MC, Paolini I, Barbetti F, et al. Antibodies from patients with rheumatoid arthritis target citrullinated histone 4 contained in neutrophils extracellular traps. Ann Rheum Dis 2014;73:1414–22.

18. Wigerblad G, Bas DB, Fernades-Cerqueira C, Krishnamurthy A, Nandakumar KS, Rogoz K, et al. Autoantibodies to citrullinated pro-teins induce joint pain independent of inflammation via a chemokine- dependent mechanism. Ann Rheum Dis 2016;75:730–8.

19. Habets KL, Trouw LA, Levarht EW, Korporaal SJ, Habets PA, de Groot P, et al. Anti- citrullinated protein antibodies contribute to platelet activation in rheumatoid arthritis. Arthritis Res Ther 2015;17:209.

20. Klareskog L, Stolt P, Lundberg K, Källberg H, Bengtsson C, Grunewald J, et al. A new model for an etiology of rheumatoid ar-thritis: smoking may trigger HLA–DR (shared epitope)–restricted im-mune reactions to autoantigens modified by citrullination. Arthritis Rheum 2006;54:38–46.

21. Harre U, Georgess D, Bang H, Bozec A, Axmann R, Ossipova E, et al. Induction of osteoclastogenesis and bone loss by hu-man autoantibodies against citrullinated vimentin. J Clin Invest 2012;122:1791–802.

22. Engdahl C, Bang H, Dietel K, Lang SC, Harre U, Schett G. Peri-articular bone loss in arthritis is induced by autoantibodies against citrullinated vimentin. J Bone Miner Res 2017;32:1681–91.

23. Proost P, Loos T, Mortier A, Schutyser E, Gouwy M, Noppen S, et al. Citrullination of CXCL8 by peptidylarginine deiminase alters receptor usage, prevents proteolysis, and dampens tissue inflammation. J Exp Med 2008;205:2085–97.

24. Struyf S, Noppen S, Loos T, Mortier A, Gouwy M, Verbeke H, et al. Citrullination of CXCL12 differentially reduces CXCR4 and CXCR7 binding with loss of inflammatory and anti- HIV- 1 activity via CXCR4. J Immunol 2009;182:666–74.

1252

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1252–1264DOI 10.1002/art.40868 © 2019 The Authors. Arthritis & Rheumatology published by Wiley Periodicals, Inc. on behalf of American College of Rheumatology. This is an open access article under the terms of the Creative Commons Attribution-NonCommer-cial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

Activation of the Peroxisome Proliferator–Activated Receptor γ Coactivator 1β/NFATc1 Pathway in Circulating Osteoclast Precursors Associated With Bone Destruction in Rheumatoid ArthritisJian-Da Ma,1 Jun Jing,1 Jun-Wei Wang,1 Ying-Qian Mo,1 Qian-Hua Li,1 Jian-Zi Lin,1 Le-Feng Chen,1 Lan Shao,2 Pierre Miossec,3 and Lie Dai1

Objective. Activation of osteoclastogenesis at the bone site in rheumatoid arthritis (RA) is well established. The mechanisms by which circulating osteoclast precursors contribute are still unclear. Peroxisome proliferator–activated receptor γ coactivator 1β (PGC- 1β) is implicated in transcriptional regulation of osteoclastogenesis in mouse models. This study was undertaken to investigate the contribution of PGC- 1β to circulating osteoclast precursors and its link to bone destruction in RA.

Methods. PGC- 1β expression in RA peripheral blood CD14+ monocytes was increased and showed correlation with joint destruction shown on radiographs. Cells from RA patients or healthy controls were transfected with a lentivirus vec-tor for PGC- 1β gene silencing or overexpression and cultured with macrophage colony- stimulating factor and RANKL. Bone resorption activity, bone- degrading enzymes, and signaling molecules were measured in these mature osteoclasts.

Results. Increased nuclear accumulation of PGC- 1β was observed in RA peripheral blood CD14+ monocytes, and these cells had stronger osteoclastogenesis than in healthy controls. PGC- 1β protein expression was positively correlated with radiographic joint destruction (r = 0.396–0.413; all P < 0.05). PGC- 1β knockdown suppressed (51–82% reduction) the expression of cathepsin K, tartrate- resistant acid phosphatase (TRAP), and matrix metalloprotein-ase 9 (MMP- 9), as well as osteoclast differentiation and bone resorption activity. Conversely, PGC- 1β overexpression increased these markers (by 1.5–1.8- fold) and osteoclastogenesis. VIVIT, an inhibitor of NFATc1 activation, inhibited the effect of overexpressed PGC- 1β by reducing cathepsin K, TRAP, and MMP- 9 expression. Chromatin immuno-precipitation assay and dual- luciferase reporter gene assay showed PGC- 1β bound to NFATc1 promoter, leading to transcriptional activation.

Conclusion. Activation of the PGC- 1β/NFATc1 pathway in circulating osteoclast precursors was associated with bone destruction in RA. This may represent a new treatment target.

INTRODUCTION

Rheumatoid arthritis (RA) is a systemic autoimmune disease characterized by progressive joint destruction leading to functional disability (1). More than 45% of patients with early RA are reported to have bone erosion at an early stage (2). Overproduction and

activation of osteoclasts at the local site are responsible for bone erosion in RA (3). Osteoclast precursors derive from human periph-eral blood monocytes, and their differentiation occurs in vitro in the presence of RANKL and macrophage colony- stimulating factor (M- CSF) (4). The contribution of monocyte- derived osteoclast precursors in vivo is unclear. Previous in vitro studies showed

Supported by the National Natural Science Foundation of China (grants 81471597, 81671612, and 81801606), Guangdong Natural Science Foundation (grants 2017A030313576, 2017A030310236, and 2018A030313541), Guangdong Medical Scientific Research Foundation (grant A2017109), Fundamental Research Funds for the Central Universities (17ykjc12), and the Scientific Program of Traditional Chinese Medicine Bureau of Guangdong Province (grant 20161058).

1Jian-Da Ma, PhD, Jun Jing, MD, Jun-Wei Wang, MD, Ying-Qian Mo, MD, PhD, Qian-Hua Li, MD, Jian-Zi Lin, MD, Le-Feng Chen, MD, Lie Dai, MD, PhD: Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University, Guangzhou, China; 2Lan Shao, PhD: First Affiliated Hospital, Sun Yat-Sen University, Guangzhou,

China; 3Pierre Miossec, MD, PhD: University of Lyon and Hospices Civils de Lyon, Lyon, France.

Drs. Ma and Jing contributed equally to this work.No potential conflicts of interest relevant to this article were reported.Address correspondence to Lan Shao, PhD, First Affiliated Hospital,

Sun Yat-Sen University, The Center for Translational Medicine, Guangzhou 510080, China (e-mail: shaolan@mail.sysu.edu.cn); or to Lie Dai, MD, PhD, Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University, Department of Rheumatology, Guangzhou 510120, China (e-mail: dailie@mail.sysu.edu.cn).

Submitted for publication January 11, 2019; accepted in revised form February 21, 2019.

PGC-­1β/NFATc1­PATHWAY­AND­BONE­DESTRUCTION­IN­RA­ |      1253

that osteoclast differentiation from peripheral blood monocytes was enhanced in patients with RA compared to healthy controls (5,6). The identification of mechanisms leading to an association between circulating activation and actual bone destruction may represent a new target for treatment in diseases, such as RA, that are associated with increased activation of osteoclasts.

Changes in intracellular metabolic pathways in immune cells could alter their function (7–9). Ex vivo–generated macrophages from peripheral blood CD14+ monocytes from RA patients pro-duced higher levels of ATP and mitochondrial activity. Monocytes are the common precursors of macrophages and osteoclasts, and such activation in RA may be linked to bone destruction in this disease (10). Peroxisome proliferator–activated receptor γ coactivator 1α (PGC- 1ɑ) and PGC- 1β are part of a group of mas-ter regulators of mitochondrial biogenesis and respiration (11). PGC- 1β, but not PGC- 1α, is induced during osteoclast differen-tiation. PGC- 1β–deficient mice show osteoclast defects associ-ated with impaired bone resorption, suggesting a role of PGC- 1β in osteoclast differentiation and function (12,13). However, the contribution of PGC- 1β in osteoclast differentiation from osteo-clast precursors remains to be studied in RA.

In the current study, we investigated the role of PGC- 1β in circulating osteoclast precursors and its contribution to oste-oclastogenesis in RA. In addition, the contribution of NFATc1, a critical transcriptional regulator of osteoclastogenesis, was determined (14). We found an increased nuclear accumulation of PGC- 1β in peripheral blood CD14+ monocytes from patients with RA that correlated with the degree of joint destruction. These cells exhibited a strong capacity for osteoclastogenesis, which was increased by overexpression of PGC- 1β and decreased by PGC- 1β knockdown. The inhibition of NFATc1 activation limited the effect of overexpressed PGC- 1β. These results indicated that activation of the PGC- 1β/NFATc1 pathway in circulating osteo-clast precursors was associated with bone destruction in RA.

PATIENTS AND METHODS

Patients and controls. Forty- two RA patients who ful-filled the 1987 American College of Rheumatology (ACR) revised classification criteria for RA (15) or the ACR/European League Against Rheumatism (EULAR) 2010 classification criteria for RA (16) were recruited from the Department of Rheumatology at Sun Yat- Sen Memorial Hospital from April 2016 to September 2018. Sex- matched healthy volunteers (n = 26) and patients with osteo-arthritis (OA) (n = 16) were recruited as controls. The exclusion cri-teria were being age >65 years; having diabetes mellitus, obesity, severe infection, malignancy, or neurologic disease; or having an overlap with other autoimmune diseases such as lupus, myosi-tis, or scleroderma. The demographic characteristics of the RA patients and controls are summarized in Supplementary Table 1 (available on the Arthritis & Rheumatology web site at http://onlin e libr ary.wiley.com/doi/10.1002/art.40868/ abstract). The study was

approved by the Medical Ethics Committee of Sun Yat- Sen Memo-rial Hospital (SYSEC- 2014- LSY- 89). All participants provided writ-ten informed consent before clinical data collection.

Demographic and clinical data were collected at the time of recruitment as described previously (17,18). RA disease activity was assessed using the Disease Activity Score in 28 joints with 4 variables, including C- reactive protein level (19). Radiographs of bilateral hands, wrists, and feet (anteroposterior view) were per-formed on all RA patients and assessed with the modified Sharp/van der Heijde score by 2 experienced observers (J- ZL and L- FC), who were blinded with regard to clinical data (20,21). Reliability and agreement were assessed using an intraclass correlation coefficient (ICC); the mean ICC for interobserver agreement was 0.90. Erosive disease was defined according to the 2013 EULAR definition when a cortical break was detected by radiography (22).

Isolation of peripheral blood CD14+ monocytes. Peripheral blood mononuclear cells (PBMCs) were isolated from RA patients and OA patients, as well as healthy controls, by using Ficoll- Hypaque density- gradient centrifugation (Sigma- Aldrich). Cells were washed twice with cold phosphate buffered saline (PBS; Gibco), and monocytes were isolated from PBMCs using CD14+ magnetic bead separation (BD Biosciences), following the protocol of the manufacturer. Separated monocytes were stained with phycoerythrin (PE)–anti- CD14 antibody (BD Biosciences) for flow cytometric analysis.

Osteoclast differentiation and bone resorption assay. For the osteoclast differentiation assay, 1 × 106 periph-eral blood CD14+ monocytes were cultured in 24- well culture plates in the presence of 100 ng/ml of RANKL and 50 ng/ml of M- CSF (both from PeproTech). Osteoclasts were confirmed by tartrate- resistant acid phosphatase (TRAP) staining and further staining with fluorescein isothiocyanate (FITC)–phalloi-din (both from Sigma- Aldrich) to detect F- actin ring using a Leica DMI4000B inverted wide- field fluorescence microscope. Mature osteoclasts were defined as TRAP- positive giant cells including ≥3 nuclei and bands of F- actin–containing podo-somes.

For bone resorption assay, bovine cortical bone slices were layered at the bottom of 48- well culture plates, and 5 × 105 peripheral blood CD14+ monocytes were seeded onto the slices and cultured in the presence of 100 ng/ml of recombinant human RANKL and 50 ng/ml of M- CSF. Resorption pits on the slices were shown by toluidine blue staining and measured using an ImageJ 1.47 analysis system (NIH).

Immunofluorescence staining. Peripheral blood CD14+ monocytes were plated on 24- well culture plates with coverslips. After incubation for 4 hours, the medium was aspirated and cells were washed twice in PBS, fixed in 4% paraformaldehyde for 15–30 minutes, permeabilized in 0.2% Triton X- 100 for 10 min-

MA ET AL 1254       |

utes at room temperature for the exposure of intracellular antigen, and blocked in PBS containing 3% bovine serum albumin (BSA; Affymetrix) for 30 minutes. Cells were then washed in PBS 3 times for 10 minutes each time and incubated in PBS containing rab-bit anti- human polyclonal antibody against PGC- 1β (Bioss) and mouse anti- human CD14 monoclonal antibody (Abcam) or nor-mal rabbit IgG overnight at 4°C. Alexa Fluor 594–conjugated goat anti- rabbit IgG and Alexa Fluor 488–conjugated goat anti- mouse IgG (1:1,000; concentrations of stock solutions 2 mg/ml) (both from Invitrogen) were added and incubated for 1 hour at 37°C. After washing in PBS, the nucleus was stained with DAPI (Sigma- Aldrich) for 3 minutes and coverslips were mounted with ProLong Gold Antifade Reagent (Invitrogen). Images were analyzed using a Zeiss LSM 710 Confocal Imaging System (23).

Flow cytometric analysis. Peripheral blood CD14+ mono-cytes were stained first with PE- conjugated anti- CD14 monoclo-nal antibody (BD Biosciences) for cell surface antigen. Cells were washed twice with staining wash buffer and centrifuged (1,000 revolutions per minute for 5 minutes) to pellet the cells. They were then resuspended with 100 μl of fixation/permeabilization solution (eBioscience) for 30 minutes at 4°C to expose intracellular anti-gen. Cells were washed twice with 500 μl of wash buffer and sus-pended with 100 μl of permeabilization buffer mixed with 1 μl of rabbit anti- human/mouse PGC- 1β antibody (Bioss) in the dark for 30 minutes at room temperature. Next, they were washed twice and resuspended with 100 μl of permeabilization buffer mixed with 1 μl of FITC- conjugated anti- rabbit secondary antibody (Invitrogen) in the dark for 20 minutes at room temperature. Stained cells were washed with permeabilization buffer and resuspended with 200 μl of phosphate buffered albumin (0.5% BSA and 0.05% NaN3 in PBS) before flow cytometric analysis. In each case, staining was compared to that of the appropriately labeled isotype control anti-body.

PGC- 1β gene silencing or overexpression by lentivi-rus transfection. To obtain cell lines with stable silencing or over-expression of PGC- 1β, peripheral blood CD14+ monocytes were transfected with PGC- 1β sequence–specific short hairpin RNA (shRNA) expression lentivirus or overexpression lentivirus, which were synthesized by Shanghai GeneChem. The target sequences for human PGC- 1β knockdown were GCATAGTCTAGGCAAA-GAAAT, marked as Lv- sh- PGC- 1β, and shRNA targeting CCTAA-GGTTAAGTCGCCCTCG (noncoding in human) was cloned into the same vector, used as control, and marked as Lv- sh- GFP. Human full- length PGC- 1β complementary DNA (cDNA) was cloned into lentiviral vector pLV[Exp] and marked as Lv- PGC- 1β, and empty pLV[Exp] vector expressing green fluorescent protein (GFP) only were used as negative control, and referred to as Lv- GFP. The production and transfection of lentivirus were conducted as described previously (24). Stably transduced cells were selected by addition of puromycin (1 μg/ml) for 48 hours and verified by real-

time quantitative poly merase chain reaction (qPCR) and Western blot analysis.

Real- time qPCR analysis. Total RNA was prepared from cells, using RNAiso reagent (Takara). RNA was reverse tran-scribed into cDNA using a reverse transcript kit (Takara) according to the instructions of the manufacturer. Complementary DNA was amplified by using recombinant Taq DNA polymerase (Takara) and specific oligonucleotide primers of PGC- 1β, tumor necrosis factor receptor–associated factor 6 (TRAF6), and β- actin (Supplemen-tary Table 2, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40868/ abstract). SYBR Green–based qPCR was performed using a Roche Light-Cycler 480 sequence detector system (25).

Western blot analysis. Cytoplasmic or nuclear proteins from human peripheral blood CD14+ monocytes were extracted separately using nuclear protein extraction kits (Pierce). Target proteins from cytoplasm were detected with primary antibod-ies to TRAP, DC- STAMP, cathepsin K (1:1,000; all from Abcam), matrix metalloproteinase 9 (MMP- 9), TRAF6, phospho- p38, p38, phospho- ERK1/2, ERK1/2, phospho- JNK, JNK, and β- tubulin (1:1,000; all from Cell Signaling Technology). Target proteins from nuclei were detected with primary antibodies to NFATc1, c- Fos, c- Jun, and fibrillarin (1:1,000; all from Cell Signaling Technology), as well as PGC- 1β (1:1,000; Abcam). Protein bands were visual-ized using enhanced chemiluminescence (Millipore) plus Western blot detection reagents, followed by exposure to a scanning imager (G:BOX Gel & Blot Imaging Series; Syngene) (26).

VIVIT treatment of human peripheral blood CD14+ monocytes. VIVIT (MCE) was used as an inhibitor of NFATc1 acti-vation (27). VIVIT powder was dissolved in Dulbecco’s modified Eagle’s medium at a concentration of 10 μM. After treatment with 10 μM VIVIT plus 100 ng/ml of recombinant human RANKL and 50 ng/ml of M- CSF for 24 hours, nuclear expression of PGC- 1β and NFATc1 in peripheral blood CD14+ monocytes from healthy controls was detected by Western blotting. After treatment with 10 μM VIVIT plus 100 ng/ml of recombinant human RANKL and 50 ng/ml of M- CSF for 21 days, cytoplasmic proteins of DC- STAMP, cathepsin K, TRAP, and MMP- 9 in peripheral blood CD14+ mono-cytes from healthy controls were detected by Western blotting.

Chromatin immunoprecipitation (ChIP). ChIP was performed using a ChIP assay kit (Cell Signaling Technology) according to the instructions of the manufacturer. Briefly, cells in a 10- cm culture plate were crosslinked with 1% formaldehyde for 10 minutes. Crosslinking was neutralized with 0.2M glycine. Cells were collected and suspended in lysis buffer. Genomic fragments were sonicated to a proper length. Protein–DNA complexes were precipitated with PGC- 1β antibody or IgG (Cell Signaling Technol-ogy) as a negative control, and anti– RNA polymerase II (Cell Sign-

PGC-­1β/NFATc1­PATHWAY­AND­BONE­DESTRUCTION­IN­RA­ |      1255

aling Technology) antibody as a positive control, overnight at 4°C. The complexes were purified using protein A/G magnetic beads, and the crosslinks were reversed at 68°C. The DNA was then purified by applying the sample to a DNA separation column. The purified DNA was amplified by PCR, and the PCR products were analyzed by electrophoresis on a Gel Red–stained 2% agarose gel. The binding capacity of PGC- 1β to the NFATc1 promoter was analyzed by qPCR, and the shear DNA sample served as an input control (28). Primer sequences used in ChIP- qPCR for NFATc1 were as follows: 5ʹ- CCCCCTAGTAAGCCCTTTCCT- 3ʹ (forward) and 5ʹ- GGGAAAGAGTTGAGGGACTTAGAA- 3ʹ (reverse).

Dual- luciferase reporter gene assay. Plasmid pcDNA3.1- PGC- 1β was purchased from GeneChem. The DNA sequences of NFATc1 were custom synthesized by GeneChem and cloned into a firefly luciferase plasmid. Periph-eral blood CD14+ monocytes from healthy controls with 80% confluence in 24- well plates were transfected using Lipo-fectamine 2000 Reagent (Life Technologies) according to the instructions of the manufacturer. Firefly luciferase plasmid of NFATc1 (0.1 μg) and pcDNA3.1- PGC- 1β (0.2 μg, 0.4 μg, and 0.6 μg) were cotransfected with Renilla luciferase vector (Pro-mega) for normalization. Forty- eight hours after transfection,

Figure 1. Expression of peroxisome proliferator–activated receptor γ coactivator 1β (PGC- 1β) in peripheral blood (PB) CD14+ monocytes. CD14+ monocytes were isolated from PB mononuclear cells from healthy controls (HCs), osteoarthritis (OA) patients, and rheumatoid arthritis (RA) patients by the use of CD14+ magnetic beads. A, Transcription of PGC- 1β was measured using quantitative polymerase chain reaction. B, Localization of PGC- 1β in PB CD14+ monocytes was detected by immunofluorescence staining. Original magnification × 1,000. C, PGC- 1β expression in total and nuclear protein from PB CD14+ monocytes was detected by Western blot analysis. Data in A and C are representative of results from independent experiments using samples from 6 healthy controls, 6 OA patients, and 6 RA patients. D, PGC- 1β protein levels in PB CD14+ monocytes were analyzed by flow cytometric analysis with fluorescein isothiocyanate (FITC). Left, Representative histograms showing the mean fluorescence intensity (MFI) in samples from 1 healthy control, 1 OA patient, and 1 RA patient are shown. Right, The MFIs of FITC-conjugated PGC-1β from peripheral CD14+ monocytes from 16 healthy controls, 11 OA patients, and 30 RA patients were compared. Values in A, C, and D are the mean ± SD. E, Left, Nuclear PGC- 1β protein expression in PB CD14+ monocytes was detected by Western blot analysis. Right, PGC- 1β band intensities were normalized to the values for fibrillarin and compared between groups. Symbols represent individual subjects (n = 16 healthy controls, n = 11 OA patients, and n = 30 RA patients) in independent analysis; bars show the mean ± SD. ** = P < 0.01; *** = P < 0.001, by Student’s t- test. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40868/abstract.

MA ET AL 1256       |

Figure 2. Expression of PGC- 1β in PB CD14+ monocytes from RA patients with erosive disease, and their osteoclastogenesis potential. A, Nuclear PGC- 1β protein expression in PB CD14+ monocytes from 17 patients with nonerosive RA and 13 patients with erosive RA was detected by Western blot analysis and normalized to the values for fibrillarin. Symbols represent individual subjects; bars show the mean ± SD. B, Relative PGC- 1β protein expression in 30 RA patients was assessed for correlation with the modified total Sharp score of radiographic progression (left), joint space narrowing subscore (middle), and erosion subscore (right). C, PB CD14+ monocytes from 6 healthy controls, 6 patients with nonerosive RA, and 10 patients with erosive RA were cultured with RANKL and macrophage colony- stimulating factor. Mature osteoclasts were detected by tartrate- resistant acid phosphatase (TRAP) staining (original magnification × 40) on days 7, 14, and 21, and fluorescein isothiocyanate–phalloidin staining (original magnification × 100) on day 21. Bone resorption activity of osteoclasts was detected by toluidine blue staining (original magnification × 400) on day 21. Representative results are shown. D and E, Cell counts of mature osteoclasts on days 7, 14, and 21 (D) and the pit area of bone resorption lacunae on day 21 (E) are shown. Bars show the mean ± SD. * = P < 0.05; ** = P < 0.01; *** = P < 0.001, by Student’s t- test. See Figure 1 for other definitions. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40868/abstract.

PGC-­1β/NFATc1­PATHWAY­AND­BONE­DESTRUCTION­IN­RA­ |      1257

luciferase activity was mea sured using a Dual- Glo Luciferase Assay System (Promega).

Statistical analysis. All data were analyzed using SPSS, version 13.0. For categorical variables, data were expressed as frequencies and percentages. For continuous variables, data were expressed as the mean ± SD or the median and interquartile range. Parametric data were compared by Student’s t- test, while nonparametric data were compared by Mann- Whitney rank sum test. The chi- square test was used for comparison of categorical variables in different groups. The correlation of parametric data was assessed by Pearson’s correlation test. P values less than 0.05 were considered significant.

RESULTS

Elevated PGC- 1β expression in peripheral blood CD14+ monocytes from RA patients. To assess PGC- 1β expression in circulating osteoclast precursors, peripheral blood CD14+ monocytes were isolated from RA patients and sex- matched OA patients and healthy controls. As shown in Figure 1A, the mean ± SD PGC- 1β transcript levels were significantly higher in peripheral blood CD14+ monocytes from RA patients than those from OA patients (2.51 ± 0.32 versus 1.20 ± 0.36; P < 0.001) and healthy controls (2.51 ± 0.32 versus 1.00; P < 0.001). Intense PGC- 1β expression in the nucleus was visualized by dual- color immunostaining in peripheral blood CD14+ monocytes, and the accumulation of PGC- 1β in the nucleus was confirmed by Western blot anal-ysis (Figures 1B and C). The expression of PGC- 1β protein in RA peripheral blood CD14+ monocytes was detected by flow cytometric analysis of intracellular staining. The mean ± SD mean fluorescence intensity of FITC-conjugated PGC- 1β in RA patients was significantly higher than in OA patients (85.32 ± 14.20 versus 11.42 ± 3.10; P < 0.001) and healthy controls (85.32 ± 14.20 versus 1.52 ± 0.24; P < 0.001) (Figure  1D). Western blot analysis confirmed the higher PGC- 1β accumu-lation in peripheral blood CD14+ monocytes in RA patients than in OA patients (0.97 ± 0.68 versus 0.52 ± 0.22; P = 0.007) and healthy controls (0.97 ± 0.68 versus 0.30 ± 0.11; P < 0.001) (Figure 1E).

Association of PGC- 1β expression in monocytes with bone erosion in RA patients. To investigate the relationship between PGC- 1β expression in circulating oste-oclast precursors and bone erosion in RA, 30 patients with RA were included for statistical analysis, with 43.3% (13 of 30) having erosive disease (Supplementary Table 3, avail-able on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40868/ abstract). Nuclear expression of PGC- 1β as shown by Western blot analysis in peripheral blood CD14+ monocytes was significantly higher

in patients with erosive RA than in patients with nonerosive RA (mean ± SD 1.40 ± 0.76 versus 0.63 ± 0.36; P = 0.006) (Figure  2A). Furthermore, there was a positive correlation between PGC- 1β protein expression in peripheral blood CD14+ monocytes and total modified Sharp/van der Heijde score (r = 0.410, P = 0.025), joint space narrowing subscore (r = 0.396, P = 0.030), and erosion subscore (r = 0.413, P = 0.023) (Figure 2B).

Stronger capacity for osteoclastogenesis in mono-cytes with elevated PGC- 1β. To investigate a possible link between elevated PGC- 1β and increased osteoclastogene-sis, peripheral blood CD14+ monocytes obtained from 10 RA patients with erosive disease, 6 RA patients with nonerosive disease, and 6 healthy controls were incubated with RANKL and M- CSF to obtain osteoclasts. The demographic informa-tion and clinical features of the 16 RA patients are shown in Supplementary Table 4 (available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40868/ abstract). On day 14, the mean ± SD cell of mature osteoclasts in RA patients with or without erosive disease were significantly higher than in healthy controls, as follows: for ero-sive RA, 36 ± 8 versus 18 ± 4 (P = 0.008), and for nonerosive RA, 28 ± 7 versus 18 ± 4 (P = 0.042). On day 21, the mean ± SD cell counts in RA patients with or without erosive disease were significantly higher than in healthy controls, as follows: for erosive RA, 165 ± 27 versus 82 ± 11 (P < 0.001), and for noner-osive RA, 109 ± 22 versus 82 ± 11 (P = 0.037) (Figures 2C and D). On day 21, the mean ± SD bone resorption lacunae were significantly higher in RA patients with or without erosive dis-ease than in healthy controls, as follows: for erosive RA, 252 ± 32 versus 74 ± 12 μm2 (P < 0.001), and for nonerosive RA, 135 ± 49 versus 74 ± 12 μm2 (P = 0.026) (Figures 2C and E). The mean ± SD cell counts of mature osteoclasts were significantly higher in RA patients with erosive disease versus those without erosive disease (P = 0.038 and P = 0.002 on days 14 and 21, respectively), as was for day 21, 165 ± 27 versus 109 ± 22 [P = 0.024]) and the mean ± SD pit area of bone resorption lacunae on day 21 (P = 0.006) (Figures 2C–E). These results indicate that elevated PGC- 1β expression in peripheral blood CD14+ mono-cytes may be involved in the dysregulation of osteoclastogenesis in RA.

Suppression of osteoclastogenesis by inhibition of PGC- 1β. To explore the role of PGC- 1β in osteoclastogen-esis in circulating osteoclast precursors, a lentiviral vector with specific PGC- 1β sequence was used to knock down PGC- 1β gene and protein expression in peripheral blood CD14+ mono-cytes from RA patients (72–86%) (Figure 3A). Expression of DC- STAMP and bone- degrading enzymes cathepsin K, TRAP, and MMP- 9 was detected by Western blot analysis. DC- STAMP is an essential regulator of cell fusion among osteoclast precur-

MA ET AL 1258       |

Figure 3. Suppression of osteoclastogenesis with inhibition of PGC- 1β. PB CD14+ monocytes from RA patients were transfected with short hairpin RNA expression lentivirus for PGC- 1β knockdown (lv- sh- PGC- 1β) or with empty vector expressing green fluorescent protein (lv- sh- GFP) as control. A, Efficiency of PGC- 1β knockdown in PB CD14+ monocytes by lentivirus transfection, detected by quantitative polymerase chain reaction (left and right) and Western blot analysis (middle). B, Expression of DC- STAMP, cathepsin K, tartrate- resistant acid phosphatase (TRAP), and matrix metalloproteinase 9 (MMP- 9) following stable knockdown of PGC- 1β in PB CD14+ monocytes for 21 days, assessed by Western blot analysis. The band intensities of DC- STAMP, cathepsin K, TRAP, and MMP- 9 were normalized to the values for β- tubulin (left) and compared by Student’s t- test (right). C, PB CD14+ monocytes transfected with lv- sh- GFP or lv- sh- PGC- 1β and cultured with RANKL and macrophage colony- stimulating factor (M- CSF). Mature osteoclasts were detected by TRAP staining (original magnification × 40) on day 21 and fluorescein isothiocyanate–phalloidin staining (original magnification × 100) on day 21. Bone resorption activity of osteoclasts was detected by toluidine blue staining (original magnification × 400) on day 21. D and E, Cell counts of mature osteoclasts (D) and pit area of bone resorption lacunae (E) on day 21. Representative results using samples from 6 RA patients are shown. Data were summarized from 3 independent experiments. Bars show the mean ± SD. ** = P < 0.01; *** = P < 0.001, by Student’s t- test. NC = negative control; NS = not significant (see Figure 1 for other definitions). Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40868/abstract.

PGC-­1β/NFATc1­PATHWAY­AND­BONE­DESTRUCTION­IN­RA­ |      1259

Figure 4. Promotion of osteoclastogenesis by overexpression of PGC- 1β. PB CD14+ monocytes from healthy controls were transfected with lentivirus containing human full- length PGC- 1β cDNA vector (lv- PGC- 1β) for PGC- 1β overexpression or with empty vector expressing green fluorescent protein (lv- GFP) as control. A, Efficiency of PGC- 1β overexpression in PB CD14+ monocytes by lentivirus transfection, detected by quantitative polymerase chain reaction (left and right) and Western blot analysis (middle). B, Expression of DC- STAMP, cathepsin K, tartrate- resistant acid phosphatase (TRAP), and matrix metalloproteinase 9 (MMP- 9) following stable overexpression of PGC- 1β in PB CD14+ monocytes for 21 days, assessed by Western blot analysis. The band intensities of DC- STAMP, cathepsin K, TRAP, and MMP- 9 were normalized to the values for β- tubulin (left) and compared by Student’s t- test (right). C, PB CD14+ monocytes transfected with lv- PGC- 1β or lv- GFP and cultured with RANKL and macrophage colony- stimulating factor (M- CSF). Mature osteoclasts were detected by TRAP staining (original magnification × 40) and fluorescein isothiocyanate–phalloidin staining (original magnification × 100) on day 21. Bone resorption activity of osteoclasts was detected by toluidine blue staining (original magnification × 400) on day 21. D and E, Cell counts of mature osteoclasts (D) and pit area of bone resorption lacunae (E) on day 21. Representative results using samples from 6 healthy controls are shown. Data were summarized from 3 independent experiments. Bars show the mean ± SD. ** = P < 0.01; *** = P < 0.001, by Student’s t- test. NC = negative control; NS = not significant (see Figure 1 for other definitions). Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40868/abstract.

MA ET AL 1260       |

sors. Cathepsin K is responsible for the degradation of bone collagen, whereas TRAP is correlated with resorption activity of osteoclasts. As a potent gelatinase, MMP- 9 is required for matrix solubilization by osteoclasts. Peripheral blood CD14+ monocytes were cultured in M- CSF and RANKL for 21 days, and Western blot analysis showed that knockdown of PGC- 1β in these monocytes significantly suppressed the cytoplasmic levels of cathepsin K, TRAP, and MMP- 9 (51–82% reduction) (Figure 3B). Knockdown of PGC- 1β in peripheral blood CD14+ monocytes significantly decreased the number of mature oste-oclasts and inhibited bone  resorption activity of osteoclasts, as shown by the  mean ± SD decreased pit area of bone resorption lacunae on day 21 (for mature osteoclasts, 583 ± 73 versus 69 ± 30 [P < 0.001], and for pit area 725 ± 85 versus 138 ± 21 μm2 [P < 0.001]) (Figures 3C–E). These results indi-cate that elevated PGC- 1β in osteoclast precursors plays an important role in promoting formation of osteoclasts and their bone resorption activity.

Overexpression of PGC- 1β and promotion of osteo-clastogenesis. Confirming our hypothesis that elevated PGC- 1β promotes osteoclastogenesis in circulating osteoclast precur-sors, we found that a lentiviral vector with PGC- 1β increased the expression of the PGC- 1β gene and protein by 1.9–2.6- fold in peripheral blood CD14+ monocytes from healthy controls (Figure 4A). Additionally, peripheral blood CD14+ monocytes were cultured in M- CSF and RANKL for 21 days, and Western blot analysis showed that overexpression of PGC- 1β in these mono-cytes significantly increased the cytoplasmic levels of cathepsin K, TRAP, and MMP- 9, with a 1.5–1.8- fold elevation (Figure 4B). Furthermore,  overexpression of PGC- 1β significantly increased counts of mature osteoclasts on day 21 and significantly increased bone resorption activity of osteoclasts as measured by the mean ± SD area of bone resorption lacunae on day 21 (for mature oste-oclasts, 362 ± 63 versus 184 ± 53 [P = 0.005], and for pit area, 742 ± 53 versus 473 ± 36 μm2 [P = 0.004]) (Figures 4C–E). These results confirmed that PGC- 1β is a critical regulator of osteoclas-togenesis and that overexpression of PGC- 1β leads to excessive osteoclast differentiation and their bone resorption activity.

Promotion of osteoclastogenesis through NFATc1 activation. To explore the signaling pathway of PGC- 1β– regulated osteoclastogenesis in circulating osteoclast precursors, peripheral blood CD14+ monocytes with PGC- 1β knockdown from RA patients or with PGC- 1β overexpression from healthy controls were cultured with M- CSF and RANKL for 24 hours. Western blot analysis showed that knockdown of PGC- 1β in peripheral blood CD14+ monocytes from RA patients significantly decreased the expression of nuclear NFATc1 protein. There was no significant difference in cytoplasmic expression of TRAF6, ERK1/2, p- ERK1/2, p38, p- p38, JNK, and p- JNK or nuclear expression of c- Jun and c- Fos between the PGC- 1β knockdown

and control groups (Figure  5A). Conversely, overexpression of PGC- 1β in peripheral blood CD14+ monocytes, this time from healthy controls, significantly increased the expression of nuclear NFATc1 protein, but not that of other signaling pathway molecules (Figure 5B).

To test whether NFATc1 signaling plays a critical role in PGC- 1β–mediated osteoclastogenesis, activation of NFATc1 was inhibited by VIVIT, which selectively inhibits calcineurin- mediated dephosphorylation of NFAT. Combined with 50 ng/ml of M- CSF and 100 ng/ml of RANKL for 24 hours, short- term treatment with 10 μM of VIVIT significantly inhibited nuclear translocation of NFATc1, but not that of PGC- 1β (Figure 5C). Combined with M- CSF and RANKL for 21 days, long- term treatment with VIVIT significantly inhibited the cytoplasmic levels of cathepsin K, TRAP, and MMP- 9. It also limited the effect of overexpressed PGC- 1β on promoting the expression of cathepsin K, TRAP, and MMP- 9 in peripheral blood CD14+ monocytes from healthy controls (Fig-ure 5D). These results suggest that PGC- 1β promotes osteoclas-togenesis through activation of NFATc1.

Binding of PGC- 1β to NFATc1 promoter and transcriptional activation. In a qPCR analysis to further explore whether PGC- 1β directly regulates NFATc1 transc-ription, PGC- 1β increased the level of NFATc1 messenger RNA in peripheral blood CD14+ monocytes (data not shown). Dual- color immunostaining showed a clear NFATc1 and PGC- 1β colocalization signal in the nucleus of peripheral blood CD14+ monocytes from RA patients, whereas the PGC- 1β nuclear signal was markedly lower in cells from healthy controls, and NFATc1 was localized mostly to cytoplasm (Figure 6A). ChIP assay showed that a markedly higher amount of chromosomal DNA containing the NFATc1 promoter was immunoprecipitated with an anti–PGC- 1β antibody compared to control IgG (Figure 6B). ChIP-qPCR analysis confirmed the immunoprecipitation of PGC- 1β and the NFATc1 promoter (Figure  6C), which indicated that PGC- 1β binds to the NFATc1 promoter region. Dual- luciferase reporter gene assay showed that overexpressed PGC- 1β in the peripheral blood CD14+ monocytes from healthy controls increased the transcriptional activity of NFATc1 in a dose-dependent manner (Figure 6D), which suggested that PGC- 1β activates NFATc1 transcription.

DISCUSSION

Excessive bone resorption by osteoclasts is the major cause of bone erosion in RA. Cytokines, such as tumor necrosis fac-tor, interleukin- 1β (IL- 1β), IL- 6, and IL- 17, are effective triggers of bone resorption and some are now targeted in the clinic with inhibitors showing an effect on bone destruction (29–31). These cytokines induce osteoclast differentiation directly or indirectly by increasing the expression of RANKL, which leads to an increase

PGC-­1β/NFATc1­PATHWAY­AND­BONE­DESTRUCTION­IN­RA­ |      1261

Figure  5. Promotion of osteoclastogenesis by PGC- 1β through the NFATc1 pathway. A, Expression of tumor necrosis factor receptor–associated factor 6 (TRAF6), MAPK, activator protein 1 (AP- 1), and NFATc1 signaling pathways detected by Western blot analysis following stable knockdown of PGC- 1β in PB CD14+ monocytes from RA patients for 24 hours. B, Expression of TRAF6, MAPK, AP- 1, and NFATc1 pathways detected by Western blot analysis following stable overexpression of PGC- 1β in PB CD14+ monocytes from healthy controls for 24 hours. C, Expression of PGC- 1β and NFATc1 detected by Western blot analysis in PB CD14+ monocytes from healthy controls for 24 hours following PGC- 1β overexpression and inhibition of NFATc1. D, DC- STAMP, cathepsin K, tartrate- resistant acid phosphatase (TRAP), and matrix metalloproteinase 9 (MMP- 9) expression detected by Western blot analysis in PB CD14+ monocytes from healthy controls for 21 days following PGC- 1β overexpression and inhibition of NFATc1. Data were summarized from 3 independent experiments. Bars show the mean ± SD. * = P < 0.05; ** = P < 0.01; *** = P < 0.001, by Student’s t- test. NC = negative control; lv- sh- GFP = short hairpin RNA expression lentivirus expressing green fluorescent protein; lv- sh- PGC- 1β = shRNA expression lentivirus for PGC- 1β knockdown; M-CSF = macrophage colony-stimulating factor; NS = not significant (see Figure 1 for other definitions). Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40868/abstract.

MA ET AL 1262       |

in osteoclast differentiation and bone resorption activity, and subsequent bone erosion (32–34). Joint damage can be found in patients who have had RA for only a short time, and stud-ies have shown enhanced levels of bone metabolism markers in RA patients with preclinical disease, which suggests that bone erosion might happen before the onset of clinical inflammation (35,36). Our in vitro study also showed that peripheral blood CD14+ monocytes from RA patients, especially with erosive

disease, had a stronger capability had differentiating into osteo-clasts and higher bone resorption activity than cells from healthy controls. Therefore, monocytes preexposed to inflammation in bone marrow and exposed to circulating cytokines exhibit intra-cellular dysregulation leading to increased osteoclast differentia-tion and activation.

PGC- 1β plays important roles in regulating energy metab-olism and cytokine signaling pathways and is mainly recognized

Figure 6. Activation of transcription by the binding of PGC- 1β to the NFATc1 promoter. A, Representative immunofluorescence staining with NFATc1, PGC- 1β, and DAPI in PB CD14+ monocytes from healthy controls and RA patients. Original magnification × 1,000. B, Analysis, by chromatin immunoprecipitation (ChIP) and electrophoresis, of NFATc promoter status of IgG, RNA polymerase II (RNA Pol- II), and PGC- 1β in PB CD14+ monocytes from healthy controls and RA patients. C, Analysis, by ChIP followed by quantitative polymerase chain reaction, of NFATc promoter status of IgG, RNA polymerase II, and PGC- 1β in PB CD14+ monocytes from RA patients. D, Luciferase activity of the NFATc1 promoter region (shown as fold induction relative to that with the empty pcDNA3.1) after cotransfection with plasmids NFATc1- luc and pcDNA3.1- PGC- 1β in PB CD14+ monocytes from healthy controls. Data in C and D were summarized from 3 independent experiments. Bars show the mean ± SD. ** = P < 0.01; *** = P < 0.001, by Student’s t- test. E, Illustration of a novel PGC- 1β/NFATc1 pathway mediating excessive osteoclastogenesis in RA. M- CSF = macrophage colony- stimulating factor; MMP- 9 = matrix metalloproteinase 9; TRAP = tartrate- resistant acid phosphatase (see Figure 1 for other definitions).

PGC-­1β/NFATc1­PATHWAY­AND­BONE­DESTRUCTION­IN­RA­ |      1263

as a mitochondrial and energy regulatory protein. Earlier stud-ies of PGC-1β focused mainly on metabolic diseases such as hyperlipidemia and diabetes mellitus (37,38). We previously found that elevated PGC- 1β levels in RA fibroblast- like synovio-cytes promoted their proinflammatory effect and RANKL secre-tion (24). We then proposed that PGC- 1β might play important roles in osteoclastogenesis in RA. In the present study, we found elevated nuclear expression of PGC- 1β protein in periph-eral blood CD14+ monocytes from RA patients, especially those patients with erosive disease. This expression was positively correlated with radiographic scores. Further studies showed that elevated PGC- 1β in RA monocytes promoted osteoclast differentiation and their bone resorption activity. These results implied that PGC- 1β in circulating osteoclast precursors might be involved in RA osteoclastogenesis.

The canonical RANKL signaling pathway inducing osteo-clasts involves TRAF6. In the microenvironment of a local joint in RA, large quantities of RANKL bind to RANK on the surface of osteoclast precursors, leading to the activation of adaptor molecules such as TRAF6, which is critical for osteoclast differ-entiation and activation (39). Downstream signaling pathways from TRAF6 finally activate NFATc1, the master regulator of osteoclastogenesis. Deficiency of Nfatc1 results in complete loss of osteoclastic bone resorption (40,41). NFATc1 induces its target genes to regulate differentiation, cell fusion, and func-tion of osteoclasts (39). In this study, knockdown or overex-pression of PGC- 1β in peripheral blood CD14+ monocytes resulted in decreased or increased expression of NFATc1, and of TRAP, cathepsin K and MMP- 9, but not of TRAF6. Further inhibition of NFATc1 activation limited the role of PGC- 1β in the expression of these genes. These results indicate that PGC- 1β might act as an upstream regulator of NFATc1, but not TRAF6.

PGC- 1β positively regulates both mitochondrial biogene-sis and differentiation in osteoclasts. PGC- 1β alone, or NFATc1 co- overexpression with PGC- 1β in RelB−/− cells, allowed osteo-clast differentiation but did not rescue mitochondrial biogenesis (42), suggesting that PGC- 1β/NFATc1 regulation of osteoclast differentiation may occur through a mechanism other than the mitochondrial function of PGC- 1β. The role of PGC- 1β in reg-ulating osteoclastogenesis was confirmed by global deletion of the PGC- 1β gene in mice, leading to increased bone mass and compromised mitochondrial biogenesis in osteoclasts (43). In Tie2- Cre mice with conditionally deleted PGC- 1β in myeloid lin-eage cells, the number of osteoclasts was decreased (12). Con-sistent with these findings, our results clearly showed that, in RA monocytes, PGC- 1β directly binds to the promotor of NFATc1 and regulates its transcription.

In conclusion, our findings provide the first evidence that PGC- 1β in circulating osteoclast precursors regulates osteo-clastogenesis in RA, through mechanisms involving interac-tions in the PGC- 1β/NFATc- 1 pathway. These results indicate that PGC- 1β in peripheral blood CD14+ monocytes might be

a promising therapeutic target for RA and other diseases asso-ciated with osteoclast activation, ranging from arthritis to bone metastasis (Figure 6E).

ACKNOWLEDGMENTS

We thank all patients and medical staff who generously con-tributed to this study. We also thank Professor Liwei Lu (University of Hong Kong) and Professor Frank Pessler (TWINCORE Center for Experimental and Clinical Infection Research and Helmholtz Center for Infection Research, Braunschweig, Germany), who kindly provided valuable suggestions for this study.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final ver-sion to be submitted for publication. Drs. Shao and Dai had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Ma, Jing, Shao, Miossec, Dai.Acquisition of data. Ma, Jing, Wang, Mo, Li, Chen, Shao, Dai.Analysis and interpretation of data. Ma, Jing, Lin, Shao, Miossec, Dai.

REFERENCES 1. Smolen JS, Aletaha D, McInnes IB. Rheumatoid arthritis. Lancet

2016;388:2023–38.

2. Adamopoulos IE, Mellins ED. Alternative pathways of osteoclas-togenesis in inflammatory arthritis. Nat Rev Rheumatol 2015;11: 189–94.

3. Okamoto K, Nakashima T, Shinohara M, Negishi-Koga T, Komatsu N, Terashima A, et al. Osteoimmunology: the conceptual framework unifying the immune and skeletal systems. Physiol Rev 2017;97:1295–349.

4. Komano Y, Nanki T, Hayashida K, Taniguchi K, Miyasaka N. Identification of a human peripheral blood monocyte subset that differentiates into osteoclasts. Arthritis Res Ther 2006;8:R152.

5. Durand M, Boire G, Komarova SV, Dixon SJ, Sims SM, Harrison RE, et al. The increased in vitro osteoclastogenesis in patients with rheumatoid arthritis is due to increased percentage of precursors and decreased apoptosis: the In Vitro Osteoclast Differentiation in Arthritis (IODA) study. Bone 2011;48:588–96.

6. Rana AK, Li Y, Dang Q, Yang F. Monocytes in rheumatoid arthritis: circulating precursors of macrophages and osteoclasts and their heterogeneity and plasticity role in RA pathogenesis. Int Immunopharmacol 2018;65:348–59.

7. O’Neill LA, Kishton RJ, Rathmell J. A guide to immunometabolism for immunologists. Nat Rev Immunol 2016;16:553–65.

8. Weyand CM, Goronzy JJ. Immunometabolism in early and late stages of rheumatoid arthritis. Nat Rev Rheumatol 2017;13:291–301.

9. Murata K, Fang C, Terao C, Giannopoulou EG, Lee YJ, Lee MJ, et al. Hypoxia- sensitive COMMD1 integrates signaling and cellular metabolism in human macrophages and suppresses osteoclas-togenesis. Immunity 2017;47:66–79.

10. Zeisbrich M, Yanes RE, Zhang H, Watanabe R, Li Y, Brosig L, et al. Hypermetabolic macrophages in rheumatoid arthritis and coronary artery disease due to glycogen synthase kinase 3β inactivation. Ann Rheum Dis 2018;77:1053–62.

11. Villena JA. New insights into PGC- 1 coactivators: redefining their role in the regulation of mitochondrial function and beyond. FEBS J 2015;282:647–72.

MA ET AL 1264       |

12. Wei W, Wang X, Yang M, Smith LC, Dechow PC, Sonoda J, et al. PGC1β mediates PPARγ activation of osteoclastogenesis and rosiglitazone- induced bone loss. Cell Metab 2010;11:503–16.

13. Zhang Y, Rohatgi N, Veis DJ, Schilling J, Teitelbaum SL, Zou W. PGC1β organizes the osteoclast cytoskeleton by mitochondrial biogenesis and activation. J Bone Miner Res 2018;33:1114–25.

14. Sitara D, Aliprantis AO. Transcriptional regulation of bone and joint remodeling by NFAT. Immunol Rev 2010;233:286–300.

15. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24.

16. Aletaha D, Neogi T, Silman AJ, Funovits J, Felson DT, Bingham CO III, et al. 2010 rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum 2010;62:2569–81.

17. Ma JD, Zhou JJ, Zheng DH, Chen LF, Mo YQ, Wei XN, et al. Serum matrix metalloproteinase- 3 as a noninvasive biomarker of histological synovitis for diagnosis of rheumatoid arthritis. Mediators Inflamm 2014;2014:179284.

18. Lin JZ, Liang JJ, Ma JD, Li QH, Mo YQ, Cheng WM, et al. Myopenia is associated with joint damage in rheumatoid arthritis: a cross- sectional study. J Cachexia Sarcopenia Muscle 2019;10:355–67.

19. Prevoo ML, van ‘t Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity scores that include twenty- eight–joint counts: development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995;38:44–8.

20. Van der Heijde DM. How to read radiographs according to the Sharp/van der Heijde method. J Rheumatol 1999;26:743–5.

21. Ma JD, Wei XN, Zheng DH, Mo YQ, Chen LF, Zhang X, et al. Continuously elevated serum matrix metalloproteinase- 3 for 3–6 months predict one year radiographic progression in rheumatoid arthritis: a prospective cohort study. Arthritis Res Ther 2015;17:289.

22. Van der Heijde D, van der Helm-van MA, Aletaha D, Bingham CO, Burmester GR, Dougados M, et al. EULAR definition of erosive disease in light of the 2010 ACR/EULAR rheumatoid arthritis classification criteria. Ann Rheum Dis 2013;72:479–81.

23. Gu P, Chen X, Xie R, Han J, Xie W, Wang B, et al. lncRNA HOXD- AS1 regulates proliferation and chemo- resistance of castration- resistant prostate cancer via recruiting WDR5. Mol Ther 2017;25:1959–73.

24. Zhou JJ, Ma JD, Mo YQ, Zheng DH, Chen LF, Wei XN, et al. Down- regulating peroxisome proliferator- activated receptor- γ coactivator- 1 β alleviates the proinflammatory effect of rheumatoid arthritis fibroblast- like synoviocytes through inhibiting extracellular signal- regulated kinase, p38 and nuclear factor- κβ activation. Arthritis Res Ther 2014;16:472.

25. Chen X, Xie W, Gu P, Cai Q, Wang B, Xie Y, et al. Upregulated WDR5 promotes proliferation, self- renewal and chemoresistance in bladder cancer via mediating H3K4 trimethylation. Sci Rep 2015;5:8293.

26. Chen X, Gu P, Xie R, Han J, Liu H, Wang B, et al. Heterogeneous nuclear ribonucleoprotein K is associated with poor prognosis and regulates proliferation and apoptosis in bladder cancer. J Cell Mol Med 2017;21:1266–79.

27. Metzelder SK, Michel C, von Bonin M, Rehberger M, Hessmann E, Inselmann S, et al. NFATc1 as a therapeutic target in FLT3- ITD- positive AML. Leukemia 2015;29:1470–7.

28. Chen X, Xie R, Gu P, Huang M, Han J, Dong W, et al. Long noncoding RNA LBCS inhibits self- renewal and chemoresistance of bladder cancer stem cells through epigenetic silencing of SOX2. Cancer Res 2019;25:1389–403.

29. Noack M, Miossec P. Selected cytokine pathways in rheumatoid arthritis. Semin Immunopathol 2017;39:365–83.

30. Osta B, Roux JP, Lavocat F, Pierre M, Ndongo-Thiam N, Boivin G, et al. Differential effects of IL- 17A and TNF- α on osteoblastic differentiation of isolated synoviocytes and on bone explants from arthritis patients. Front Immunol 2015;6:151.

31. Ndongo-Thiam N, Clement A, Pin JJ, Razanajaona-Doll D, Miossec P. Negative association between autoantibodies against IL- 17, IL- 17/anti- IL- 17 antibody immune complexes and destruction in rheumatoid arthritis. Ann Rheum Dis 2016;75:1420–2.

32. Walsh MC, Takegahara N, Kim H, Choi Y. Updating osteoimmunology: regulation of bone cells by innate and adaptive immunity. Nat Rev Rheumatol 2018;14:146–56.

33. Shim JH, Stavre Z, Gravallese EM. Bone loss in rheumatoid arthritis: basic mechanisms and clinical implications. Calcif Tissue Int 2018;102:533–46.

34. Ikebuchi Y, Aoki S, Honma M, Hayashi M, Sugamori Y, Khan M, et al. Coupling of bone resorption and formation by RANKL reverse signalling. Nature 2018;561:195–200.

35. Van Schaardenburg D, Nielen MM, Lems WF, Twisk JW, Reesink HW, van de Stadt RJ, et al. Bone metabolism is altered in preclinical rheumatoid arthritis [letter]. Ann Rheum Dis 2011;70:1173–4.

36. Van Nies JA, van Steenbergen HW, Krabben A, Stomp W, Huizinga TW, Reijnierse M, et al. Evaluating processes underlying the predictive value of baseline erosions for future radiological damage in early rheumatoid arthritis. Ann Rheum Dis 2015;74:883–9.

37. Chen S, Wen X, Zhang W, Wang C, Liu J, Liu C. Hypolipidemic effect of oleanolic acid is mediated by the miR- 98- 5p/PGC- 1β axis in high- fat diet- induced hyperlipidemic mice. FASEB J 2017;31:1085–96.

38. Villegas R, Williams SM, Gao YT, Long J, Shi J, Cai H, et al. Genetic variation in the peroxisome proliferator- activated receptor (PPAR) and peroxisome proliferator- activated receptor γ co- activator 1 (PGC1) gene families and type 2 diabetes. Ann Hum Genet 2014;78:23–32.

39. Park JH, Lee NK, Lee SY. Current understanding of RANK signaling in osteoclast differentiation and maturation. Mol Cells 2017;40:706–13.

40. Aliprantis AO, Ueki Y, Sulyanto R, Park A, Sigrist KS, Sharma SM, et al. NFATc1 in mice represses osteoprotegerin during osteoclastogenesis and dissociates systemic osteopenia from inflammation in cherubism. J Clin Invest 2008;118:3775–89.

41. Winslow MM, Pan M, Starbuck M, Gallo EM, Deng L, Karsenty G, et al. Calcineurin/NFAT signaling in osteoblasts regulates bone mass. Dev Cell 2006;10:771–82.

42. Zeng R, Faccio R, Novack DV. Alternative NF- κB regulates RANKL- induced osteoclast differentiation and mitochondrial biogenesis via independent mechanisms. J Bone Miner Res 2015;30:2287–99.

43. Ishii KA, Fumoto T, Iwai K, Takeshita S, Ito M, Shimohata N, et al. Coordination of PGC- 1β and iron uptake in mitochondrial biogenesis and osteoclast activation. Nat Med 2009;15:259–66.

1265

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1265–1275DOI 10.1002/art.40877 © 2019 University of Pécs. Arthritis & Rheumatology published by Wiley Periodicals, Inc. on behalf of American College of Rheumatology. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Amelioration of Autoimmune Arthritis in Mice Treated With the DNA Methyltransferase Inhibitor 5′- AzacytidineDániel M. Tóth,1 Timea Ocskó,1 Attila Balog,2 Adrienn Markovics,1 Katalin Mikecz,1 László Kovács,2 Meenakshi Jolly,1 Aleksandra A. Bukiej,1 Andrew D. Ruthberg,1 András Vida,1 Joel A. Block,1 Tibor T. Glant,1 and Tibor A. Rauch3

Objective. Disease- associated, differentially hypermethylated regions have been reported in rheumatoid arthritis (RA), but no DNA methyltransferase inhibitors have been evaluated in either RA or any animal models of RA. The present study was conducted to evaluate the therapeutic potential of 5′- azacytidine (5′- azaC), a DNA methyltrans-ferase inhibitor, and explore the cellular and gene regulatory networks involved in the context of autoimmune arthritis.

Methods. A disease- associated genome- wide DNA methylation profile was explored by methylated CpG island recovery assay–chromatin immunoprecipitation (ChIP) in arthritic B cells. Mice with proteoglycan- induced arthritis (PGIA) were treated with 5′- azaC. The effect of 5′- azaC on the pathogenesis of PGIA was explored by measuring serum IgM and IgG1 antibody levels using enzyme- linked immunosorbent assay, investigating the efficiency of class- switch recombination (CSR) and Aicda gene expression using real- time quantitative polymerase chain reaction, mon-itoring germinal center (GC) formation by immunohistochemistry, and determining alterations in B cell subpopulations by flow cytometry. The 5′- azaC– induced regulation of the Aicda gene was explored using RNA interference, ChIP, and luciferase assays.

Results. We explored arthritis- associated hypermethylated regions in mouse B cells and demonstrated that DNA demethylation had a beneficial effect on autoimmune arthritis. The 5′- azaC–mediated demethylation of the epigenet-ically inactivated Ahr gene resulted in suppressed expression of the Aicda gene, reduced CSR, and compromised GC formation. Ultimately, this process led to diminished IgG1 antibody production and amelioration of autoimmune arthritis in mice.

Conclusion. DNA hypermethylation plays a leading role in the pathogenesis of autoimmune arthritis and its targeted inhibition has therapeutic potential in arthritis management.

INTRODUCTION

Rheumatoid arthritis (RA) is a systemic inflammatory autoim-mune disorder that primarily leads to joint destruction. B cells play an indispensable role in its initiation via the production of high- affinity autoantibodies. It has recently been suggested that dys-regulation of the B cell epigenome can contribute to this antibody production (1).

Currently, there is no cure for RA, but treatments can pro-vide alleviation of symptoms and modify disease progression. In

most cases, the optimal treatment can only be achieved through a combination of different drugs (2). When a combination of traditional disease- modifying antirheumatic drugs, nonsteroidal antiinflammatory drugs, and/or low- dose glucocorticoids does not provide a satisfactory response, highly specific biologic agents are introduced into the treatment regimen (2,3). Highly effective epigenetic enzyme inhibitors have already been devel-oped, and their therapeutic potential has been demonstrated in cancer (4). However, no such drug is currently used to treat RA (5). Epigenetic enzyme targeting drugs represent a novel

Supported by the NIH (grants R01-AR-059356 and R01-AR-062991 to Dr. Glant and R21-AR-064948 to Dr. Rauch). Dr. Balog is supported by the Hungarian Academy of Sciences (the János Bolyai Scholarship).

1Dániel M. Tóth, PhD, Timea Ocskó, MSc, Adrienn Markovics, MD, PhD, Katalin Mikecz, MD, PhD, Meenakshi Jolly, MD, Aleksandra A. Bukiej, MD, Andrew D. Ruthberg, MD, András Vida, PhD, Joel A. Block, MD, Tibor T. Glant, MD, PhD: Rush University Medical Center, Chicago, Illinois; 2Attila Balog, MD, PhD, László Kovács, MD, PhD: Albert Szent-Györgyi Clinical Center, Szeged, Hungary; 3Tibor A. Rauch, PhD: Rush

University Medical Center, Chicago, Illinois, and University of Pécs, Pécs, Hungary.

Dr. Rauch holds a patent for a method of detecting methylated CpG islands (US patent application no. 7,425,415). No other disclosures relevant to this article were reported.

Address correspondence to Tibor A. Rauch, PhD, Department of Medical Biology, School of Medicine, University of Pécs, Pécs, 48-as tér 1, 7622, Hungary. E-mail: Tibor.Rauch@aok.pte.hu.

Submitted for publication October 19, 2018; accepted in revised form February 28, 2019.

TÓTH ET AL 1266       |

approach that might replace or be used in combination with conventional arthritis therapies to achieve more effective arthritis management.

DNA methylation, one of the most common epigenetic modifications, is catalyzed by DNA methyltransferases and associated with gene silencing when it takes place in promoter regions (6). Recent studies have shown that RA pathogenesis is associated with DNA methylome changes similarly to car-cinogenesis. In both diseases characteristic global DNA hypo-methylation (decreased methylation) events affecting intergenic and intragenic regions (7–9) and de novo hypermethylated (increased methylation) promoters have been described (8,10). Induced demethylation of disease- specific hypermethylated promoters using DNA methyltransferase inhibitors, such as 5′- azacytidine (5′- azaC), can rescue the epigenetically inacti-vated genes, resulting in impaired disease progression (11,12). However, the significance of DNA hypermethylation in RA pathogenesis has been shown only indirectly (13). We hypoth-esized that locus- specific hypermethylation in B cells may play a role in arthritis pathogenesis and that DNA methylation inhib-itors may have therapeutic potential in RA.

Herein, we report that low- dose 5′- azaC treatment has a notable effect on disease progression in proteoglycan- induced arthritis (PGIA), a murine model of RA (14). The 5′- azaC–medi-ated suppressive effect is due to transcriptional silencing of activation- induced cytidine deaminase (AID), an enzyme respon-sible for multiple catalytic steps of high- affinity antibody matu-ration encompassing class- switch recombination (CSR) and somatic hypermutation (15,16). Furthermore, AID plays a pivotal role in germinal center (GC) formation (17), the site of autoanti-body production in secondary lymphoid organs (18). We demon-strate that aryl hydrocarbon receptor (AHR) transcription factor, which is known to be involved in B cell differentiation (19) and directly regulates Aicda expression (20), is rescued from arthritis- associated promoter hypermethylation in B cells after 5′- azaC treatment and contributes to the suppression of antibody pro-duction. These data promote consideration of epigenetic drugs in arthritis therapy.

MATERIALS AND METHODS

Methylated CpG island recovery assay (MIRA). MIRA–chromatin immunoprecipitation (ChIP) was performed as previously described. MIRA- enriched DNA fractions were analyzed by quanti-tative polymerase chain reaction (qPCR) wherein the Xist promoter and mouse Gapdh or human TBP promoters were used as positive and negative controls, respectively (Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). Measured Ct values were compared to negative control promoter–related Ct values as 1- fold baseline methylation level. DNA methylation raw data files are acces-sible at NCBI GEO (accession no. GSE98939).

Gene expression analysis using Agilent microar-ray platforms. RNA sample labeling, array hybridization, and primary data analysis were performed by Arraystar. Microarray data are available from the NCBI GEO database (accession no. GSE98932).

Induction and assessment of PGIA in mice and 5′- azaC treatment. To induce PGIA, 3- month old female wild- type BALB/c mice (Charles River) were immunized intraperitoneally with human PG as previously described (14). At disease onset, the degree of arthritis in each paw was visually scored for redness and swelling on a scale of 0–4 every other day (21). The scoring was performed by one individual (TTG) in a blinded manner. Mice with a score of ≥1 on at least 1 limb were divided into 2 groups. The 5′- azaC–treated mice with PGIA received 2 mg/kg of 5′- azaC (Sigma- Aldrich) IP every other day for either 2 weeks or 4 weeks. The 5′- azaC was freshly dissolved at 0.5 mg/ml concentration in saline solution. The control mice with PGIA received saline solu-tion alone. For in vitro tests, samples were obtained from the mice treated for 2 weeks. The animal studies were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Rush University Medical Center (IACUC # 15–065).

Histologic analysis. Tissue sections from mouse hind limbs were prepared, processed, examined, and photographed as previously described (21).

Enzyme- linked immunosorbent assay (ELISA). Briefly, IgM isotype control (catalog no. 02- 6800; ThermoFisher) or mouse purified IgG1 (0.025 μg/well) was coated onto an ELISA plate. IgM horseradish peroxidase (HRP) antibody (cata-log no. 04- 6820; ThermoFisher) was diluted to 1:2,000 in phos-phate buffered saline (PBS) and added to serum samples diluted 1:100. IgG1 HRP antibody (catalog no. 559626; BD Biosciences) was diluted to 1:40,000 and added to serum samples diluted 1:160,000. For no- serum control (NSC), diluted antibodies were incubated in PBS. After a 1- hour incubation at room temper-ature, samples were added to coated wells and incubated for an additional hour at room temperature. After washes, the color reaction was developed and the optical density (OD) was read using a Synergy 2 ELISA reader. The OD of the NSC well was set to 100% binding, and the percentage of inhibition was cal-culated according to the formula % inhibition = 100−[(ODsample × 100)/ODNSC]. To measure serum levels of the cytokine inter-leukin- 4 (IL- 4), a mouse IL- 4 ELISA kit (catalog no. 555232; BD Biosciences) was used according to the recommendations of the manufacturer.

Isolation of B cells and human peripheral blood mononuclear cells (PBMCs). Mouse spleen cells were iso-lated, and an immunomagnetic selection kit (StemCell) was used for negative B cell selection, according to the recommendations

SUPPRESSION OF AUTOIMMUNE ARTHRITIS BY 5′- AZACYTIDINE |      1267

of the manufacturer. Peripheral blood samples were obtained from consenting treatment- naive RA patients and from consent-ing healthy individuals (Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract) at Albert Szent- Györgyi Clinical Center and Rush University Medical Center under Ethics Com-mittee– and IRB- approved protocols (Hungarian ETT TUKEB905/PI/09 and RUMC 13082202- IRB01). PBMCs were separated on a Ficoll density gradient within 1 hour of blood draw, and were then stored in RNAlater solution (ThermoFisher) until RNA and DNA preparation. Human B cells were isolated from peripheral blood using a negative immunomagnetic selection kit (StemCell) according to the recommendations of the manufacturer.

RNA isolation, complementary DNA (cDNA) syn-thesis, and real- time qPCR. Total RNA preparation, cDNA synthesis, and real- time qPCR were conducted as previously described (17). Measured Ct values were normalized to Hprt1 (mouse samples) or β- actin (human samples). Relative expres-sion was calculated using CFX Manager software (Bio- Rad). The primers used are listed in Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract.

Digestion- circulation qPCR. Digestion- circulation qPCR was performed based on the method described by Lumsden et al (22) with modifications. Quantitative PCR was used to quantify circulated DNA products. The qPCR was performed in 25 μl final volume using 5 μl circulated DNA, SsoAdvanced Universal SYBR Green Supermix (Bio- Rad), and primers (Supplementary Table 1) at 0.5 μM final concentration. The qPCR conditions were as fol-lows: 94°C for 2 minutes, 40 cycles at 94°C for 8 seconds, and 62°C for 30 seconds. Circulated nicotinic acetylcholine receptor (nAChR) DNA was used for normalization. Relative quantities of IgG1- specific recombinant DNA (Sμ−Sγ1) were calculated by 2−ΔCt equation, where ΔCt = CtSμ−Sγ1−CtnAChR.

Flow cytometric analysis. Erythrocytes were lysed from mouse spleen and bone marrow (BM) cells by hypotonic buffer, then Fc receptors (FcR) were blocked with TruStain FcX anti- mouse CD16/CD32 (catalog no. 101320; BioLegend) and surface antigens were stained with fluorescence- labeled antibodies (Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). Data acquisition and analy-sis were performed using a FACSCanto II instrument and BD FACSDiva software (version 5.0).

Immunohistochemistry. Mouse spleens and lymph nodes were embedded and snap- frozen in Tissue- Tek OCT compound (Sakura Finetek), cut on a Microm HM 550 cryo-stat (Microm International), and stored at −20°C until used.

Sections were fixed by cold 100% acetone (Sigma- Aldrich) and blocked with 1% normal goat serum (ThermoFisher) in PBS, then immunostained with fluorescence- labeled antibod-ies (Supplementary Table 3) for 1 hour. Stained sections were post- fixed in 10% formalin and images were captured on a Zeiss LSM 700 confocal microscope and analyzed with Zen 2.0 software (Zeiss).

Cell cultures and in vitro 5′- azaC treatment. A20 BALB/c mouse B lymphoma and HEK 293T cell lines were purchased from ATCC and were maintained in Dulbecco’s modified Eagle’s medium (Sigma- Aldrich) supplemented with 10% fetal bovine serum (FBS; HyClone) and antibiotics. The human B cell line GM12878 was purchased from the Coriell Institute of Aging Cell Repository and was cultured in RPMI 1640 medium (Lonza) supplemented with 15% FBS and anti-biotics. All cells were maintained under standard culture con-ditions.

The 5′- azaC was dissolved in DMSO at a 10 mM concen-tration and stored at −20°C until used. A20 cells were plated into 6- well plates (2 × 105 cell/ml) in medium supplemented with 5′- azaC at a 10 μM final concentration. Control samples were treated with DMSO at a 0.1% final concentration. Cell cultures were incu-bated for 24, 48, or 72 hours. Media with the same reagents were replaced with fresh media every 24 hours.

Western blot analyses. A20 cells were treated with 5′- azaC for 48 hours, and then nuclear and cytoplasmic extracts were prepared using a Nuclei EZ Prep kit (Sigma- Aldrich) accord-ing to the manufacturer’s instructions. Protein concentrations were determined using a BCA Protein Assay kit (ThermoFisher) and a Synergy 2 ELISA reader (BioTek Instruments). Protein extracts were separated on 4–20% sodium dodecyl sulfate–poly-acrylamide gels (Bio- Rad), blotted onto Immuno- Blot membranes (Bio- Rad), then blocked with Tris buffered saline–0.05% Tween 20 containing 3% nonfat milk, and incubated with the following primary antibodies at 4°C overnight: anti- Ahr (1:500) (catalog no. GTX22769; GeneTex), anti- Aicda (1:625) (catalog no. ZA001; ThermoFisher), antilamin β (1:300) (catalog no. sc6217; Santa Cruz Biotechnology), and anti- GAPDH (1:10,000) (catalog no. G9545; Sigma- Aldrich). HRP- conjugated secondary antibod-ies were purchased from Santa Cruz Biotechnology. Pierce ECL Western Blotting Substrate (ThermoFisher) was used to generate a chemiluminescent signal, which was detected by radiography and developed in an Alphatek AX200 machine.

RNA interference study. An Ahr- specific short hairpin RNA (shRNA) sequence (Supplementary Table 1) was inserted into the pU6Abase vector (23) downstream of the human U6 promoter (pU6AshRNA). Exponentially growing A20 cells (107) were sus-pended in electroporation buffer (20 mM HEPES, 135 mM KCl, 2 mM MgCl2, and 0.5% Ficoll 400, pH 7.6) together with 20 μg plas-

TÓTH ET AL 1268       |

mid DNA. Electroporation was carried out using 940 volts/cm and 950 μF in a Gene Pulser II machine (Bio- Rad). Cells were plated onto 6- well plates and after a 12- hour incubation they were treated with 10 μM 5′- azaC. Twenty- four hours later the cells were resus-pended in fresh tissue culture medium supplemented with 10 μM 5′- azaC and incubated for an additional 24 hours before isolation of total RNA. The pU6Abase without insert was used as control.

ChIP assay. ChIP assay was performed on A20 cells after treatment with 10 μM 5′- azaC for 48 hours. For IP, 50 μg pre-cleared chromatin was incubated with 20 μg rabbit polyclonal

anti- Ahr antibody (catalog no. sc5579; Santa Cruz Biotechnol-ogy) or incubated without antibody (mock). DNA–Ahr antibody complexes were collected using magnetic beads. DNA–AHR anti-body complexes were eluted and reverse crosslinked as input. One microliter of the eluted DNA fraction was used for qPCR in which region- specific primers were used (Supplementary Table 4, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). Quanti-tative data were calculated by the ΔΔCt method and presented as fold enrichment compared to the values for the mock- treated samples.

Figure 1. Amelioration of autoimmune arthritis in mice treated with 5′- azacytidine (5′- azaC). A, Heatmap of differentially methylated regions (left) and arthritis- associated gene expression changes (right) in B cells from naive mice (n = 3) and mice with proteoglycan- induced arthritis (PGIA; n = 3). Pie charts depict the methylation pattern (left) and gene expression pattern (right) of the annotated sites and genes in samples from mice with PGIA. B and C, Arthritis scores for mice with PGIA treated with vehicle (saline) (n = 19 in B and 24 in C) and mice with PGIA treated with 5′- azaC (n = 21 in B and 28 in C) intraperitoneally every other day for 2 weeks (B) or 4 weeks (C). Values are the mean ± SEM. * = P ≤ 0.05; ** = P ≤ 0.01; *** = P ≤ 0.001; **** = P ≤ 0.0001 versus vehicle, by two- way analysis of variance (ANOVA) with Sidak’s test. D, Body weight of mice injected with dimethyldioctadecylammonium bromide (DDA) adjuvant and treated with vehicle (n = 10), mice injected with DDA adjuvant and treated with 5′- azaC (n = 10), mice with PGIA treated with vehicle (n = 24), and mice with PGIA treated with 5′- azaC (n = 28). Values are the mean ± SEM. # = P ≤ 0.05; ## = P ≤ 0.01, mice injected with DDA treated with vehicle versus mice with PGIA treated with vehicle; * = P ≤ 0.05; *** = P ≤ 0.001, mice injected with DDA treated with 5′- azaC versus mice with PGIA treated with 5′- azaC, by two- way ANOVA with Sidak’s test. E, Representative images of histopathologic features of hematoxylin and eosin–stained ankle joints from mice injected with DDA adjuvant, mice with PGIA treated with vehicle, and mice with PGIA treated with 5′- azaC. Arrows indicate bone- eroding inflammatory tissue.

SUPPRESSION OF AUTOIMMUNE ARTHRITIS BY 5′- AZACYTIDINE |      1269

Luciferase assay. Briefly, the mouse Aicda minimal promoter and the longer variant (including region 13, named pGL4.1- Aicda- p+13r) or the shorter variant (without region 13, named pGL4.1- Aicda- p+Δ13r) of the intronic region of Aicda was cloned into pGL4.1 (Promega). A reporter plasmid with a human AHR- overexpressing plasmid (pCMV- hAHR; OriGene) or with a control vector (pCMV- EGFP; Addgene) were cotrans-fected into HEK 293T cells. Twenty- four hours after transfection, luciferase activity was measured. The relative luciferase activity was calculated by normalization of measured relative lumines-cence units to protein concentrations.

Statistical analysis. All statistical analyses were performed using GraphPad Prism software, version 7.04. Results are reported as the mean ± SEM. If the significance level for the pretest for nor-mality (Shapiro- Wilk) was less than 0.05, the Mann- Whitney U test was used. If the significance level for the pretest was greater than 0.05, Student’s unpaired 2- tailed t- test was used. Comparisons of more than 2 groups were performed by one- way analysis of vari-ance (ANOVA), and groups affected by 2 factors were compared by two- way ANOVA followed by Sidak’s post hoc comparison. P values less than 0.05 were considered significant.

RESULTS

Elimination of PGIA symptoms in mice treated with low doses of a DNA hypomethylating agent. To investigate the role of DNA methylation in RA pathology, we conducted an integrative analysis by exploring genome- wide DNA methylation and accompanying gene expression changes in B cells from mice with PGIA (Figure 1A). We observed dominant arthritis- associated hypomethylation and rare hypermethylation events in the investi-

gated intergenic and intragenic regions in arthritic B cells (Supple-mentary Table 5, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). However, DNA methylation profile changes were infrequently fol-lowed by gene expression changes, implying that most of the methylation events have no direct effects on gene expression.

We recognized that carcinogenesis- and arthritis- associated DNA methylation profile changes are similar in a sense, as both are characterized by dominant genome- wide hypomethyla-tion affecting intergenic and intragenic regions, and infrequent promoter- specific hypermethylation events (24). Accordingly, we hypothesized that a demethylating agent that proved to be effective in cancer treatment could also be effective in arthritis therapy. We treated arthritic mice with low- dose 5′- azaC every other day for 2 weeks, which resulted in gradual reduction of dis-ease severity and halted arthritis progression (Figure 1B). Next, we started 5′- azaC treatment at a more advanced stage of arthri-tis and continued treatment for 4 weeks, which similarly reduced paw swelling (Figure  1C). Moreover, neither mortality nor toxic effect was observed among these mice. This was confirmed by the increased body weight of 5′- azaC–treated arthritic mice at the fourteenth administration of 5′- azaC (Figure  1D). Furthermore, arthritis- specific massive joint inflammation, synovial pannus for-mation, cartilage damage, and bone destruction were undetect-able by histopathologic analysis in 5′- azaC–treated animals by the end of the study (Figure 1E). These results confirmed our working hypothesis that 5′- azaC–induced DNA demethylation can be an effective approach for the management of autoimmune arthritis.

Attenuated IgG1 antibody production in 5′- azaC–treated mice with PGIA. The lack of joint inflammation prompted us to measure serum levels of IgM and IgG1 antibod-

Figure 2. Attenuation of IgG1 production in mice with PGIA treated with 5′- azaC. A and B, Changes in IgM (A) and IgG1 (B) antibody levels in mice with PGIA treated with vehicle (n = 8) and mice with PGIA treated with 5′- azaC (n = 8). Sera were collected before the first and after the last 5′- azaC treatment. Values are the percent of pretreatment level. C, Relative quantity of IgG1- specific class- switched recombinant genomic DNA (Sμ−Sγ1) in mice with PGIA treated with vehicle (n = 5) and mice with PGIA treated with 5′- azaC (n = 5). D and E, Relative (rel) normalized (norm) expression of mRNA for germline (Iμ−Cμ and Iγ1–Cγ1) and post- recombination (Iμ−Cγ1) IgG1 (D) and for Aicda (E) in B cells from mice with PGIA treated with vehicle (n = 8) and mice with PGIA treated with 5′- azaC (n = 8). Bars show the mean ± SEM. * = P ≤ 0.05; ** = P ≤ 0.01, by the Mann- Whitney U test in B and by t- test in C–E. See Figure 1 for definitions.

TÓTH ET AL 1270       |

ies. Samples were collected before the first 5′- azaC treatment and when the experiment was terminated. IgM levels were enhanced in both animal groups, but IgG1 levels were increased only in vehicle- treated mice (Figures 2A and B). This suggested a compromised CSR (25,26) in 5′- azaC–treated mice. Hence, we measured IgG1- specific genome rearrangement in B cells using a modified version of digestion- circulation PCR (22). The level of class- switched DNA

(Figure 2C) and the level of post- recombination IgG1 messenger RNA (mRNA) (Figure 2D) were significantly lower in 5′- azaC–treated mice. Since Aicda regulates CSR (26), we determined its expression level, which was significantly lower in 5′- azaC–treated animals (Fig-ure 2E). These data suggest that 5′- azaC inhibits Aicda expression, leading to diminished CSR and attenuated antibody production in B cells, ultimately resulting in the suppression of arthritis symptoms.

Figure 3. Treatment with 5′- azaC impairs germinal center (GC) formation and shifts former B cell subpopulations in mice with PGIA. A and B, GCs in the joint draining lymph nodes (A) and spleens (B) of mice injected with DDA adjuvant, mice with PGIA treated with vehicle, and mice with PGIA treated with 5′- azaC. Red shows follicular B cells (phycoerythrin- conjugated B220/CD45R), green shows GL- 7+ B cells (Alexa Fluor 488–conjugated GL- 7), and blue shows T cells (biotin- conjugated CD4, biotin- conjugated CD8a, and eFluor 450–conjugated streptavidin). Arrows indicate GCs in B cell follicles. Original magnification × 100. C and D, Percentage of GC B cells, determined by flow cytometry, within B lymphocytes in the lymph nodes (C) and spleen (D) of mice with PGIA treated with vehicle (n = 8) and mice with PGIA treated with 5′- azaC (n = 8). E, Percentage of plasma cells, determined by flow cytometry, within bone marrow cells from mice with PGIA treated with vehicle (n = 5) and mice with PGIA treated with 5′- azaC (n = 5). F–H, Percentage of early prepro–B cells, early pro–B cells, early pre–B cells, and late precursor B cells (F), immature and mature recirculating B cells (G), and transitional type 1 (T1), T2, marginal zone (MZ), and follicular (Fo) B cells (H), determined by flow cytometry, within B lymphocytes in the bone marrow (F and G) or spleen (H) of mice with PGIA treated with vehicle (n = 11) and mice with PGIA treated with 5′- azaC (n = 12). Bars show the mean ± SEM. * = P ≤ 0.05; ** = P ≤ 0.01, *** = P ≤ 0.001; **** = P ≤ 0.0001, by the Mann- Whitney U test in C and E and by t- test in D and F–H. See Figure 1 for other definitions. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40877/abstract.

SUPPRESSION OF AUTOIMMUNE ARTHRITIS BY 5′- AZACYTIDINE |      1271

Treatment with 5′- azaC inhibits GC formation and shifts the proportion of former B cell subpopulations in mice with PGIA. Self- reactive antibodies are mainly generated in GCs of secondary lymphoid organs (18). Thus, we investigated GC formation and determined the proportion of GC B cells in the mouse lymph nodes and spleen. Reduced GC formation and significantly decreased GC B cell frequency were observed in both organs in 5′- azaC–treated mice (Figures 3A–D and Supplementary Figures 1A and B, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). Moreover, as a consequence of diminished GC formation, the proportion of plasma cells was also decreased in the BM of 5′- azaC–treated ani-mals (Figure 3E and Supplementary Figure 1C).

To explore the origin of the altered frequency of GC B cells, we examined B cell subpopulations in mouse BM and spleen. In

both organs, the proportions of B cells were significantly lower in the lymphocyte compartments in 5′- azaC–treated animals (Supplementary Figure 2, available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). The proportions of early precursor B cells were increased (Figure 3F), while the frequencies of late precursor (Figure 3F) and immature B cells (Figure 3G) were reduced upon 5′- azaC treatment. In the periphery, we detected a decreased proportion of transitional type 1 (T1) B cells, while the frequen-cies of follicle- colonizing T2 and marginal zone (MZ) B cells were elevated within the B cell populations of 5′- azaC–treated mice (Figure 3H). Finally, 5′- azaC treatment also elevated the percent-age of the mature recirculating B cells in the BM (Figure 3G). The proportion of follicular B cells, which are essential in GC responses (18), was similar in the vehicle- treated and 5′- azaC–treated mice

Figure 4. Hypermethylated Ahr regulatory regions in mice with PGIA and treatment- naive patients with rheumatoid arthritis (RA). A, Venn diagram of PGIA- associated (up- regulated and down- regulated), marginal zone (MZ)–specific, and germinal center (GC)–specific B cell genes. * = includes the Ahr gene. B, Relative (rel) normalized (norm) Ahr mRNA expression in splenic B cells from naive mice (n = 3), splenic B cells from mice with PGIA (n = 3), and in the mouse A20 B lymphoma cell line. C, Relative DNA methylation level of the Ahr promoter in splenic B cells from naive mice (n = 3) and mice with PGIA (n = 3). D and E, Relative normalized AHR mRNA expression in human peripheral blood mononuclear cells (PBMCs) isolated from healthy individuals (n = 19) and treatment- naive RA patients (n = 46) (D) and in human B cells isolated from healthy individuals (n = 5) and treatment- naive RA patients (n = 10) (E). F, Relative DNA methylation level of differentially methylated region C of AHR in PBMCs from healthy individuals (n = 10) and treatment- naive RA patients (n = 11). In B, C, and F, bars show the mean ± SEM. In D and E, symbols represent individual patients; horizontal lines show the mean. * = P ≤ 0.05; ** = P ≤ 0.01; *** = P ≤ 0.001, by one- way ANOVA with Sidak’s test in B, by t- test in C, and by the Mann- Whitney U test in D–F. See Figure 1 for other definitions.

TÓTH ET AL 1272       |

with PGIA (Figure 3H and Supplementary Figure 3, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract).

Since follicular B cells can evolve GCs in a T cell–dependent manner (27), we investigated the frequency of CD4+ T cells and measured serum levels of IL- 4 cytokine, which is predominantly produced by GC- forming follicular T helper cells (28). Interest-ingly, neither the frequency of CD4+ T cells nor the serum level of IL- 4 was changed in 5′- azaC–treated mice compared to vehicle- treated arthritic animals (data not shown).

Hypermethylated Ahr regulatory regions in arthritic mouse B cells and human samples. DNA hypomethylation induced by 5′- azaC is nonspecific (29), in contrast to the locus- specific Aicda- mediated hypomethylation events in the B cell epige-nome, which are essential for effective GC formation (17). Accordingly, it is conceivable that 5′- azaC–induced demethylation may pro-voke genes that inhibit GC formation. Taking this into account, our aim was to identify genes that could be induced by 5′- azaC and suppress Aicda expression. To this end, we compared arthritis- associated up- regulated and down- regulated genes (Figure 1A) with characteristic sets of MZ and GC B cell genes (30) (Figure 4A) that express Aicda (15,25). We focused on down- regulated genes that were also hypermethylated in arthritis, which can be possible tar-gets of 5′- azaC–triggered activation (Supplementary Table 6, avail-able on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). These down- regulated genes included the Ahr gene, which encodes a transcription factor that is known to suppress Aicda expression after agonist stimula-tion (20). Indeed, Ahr expression was down- regulated (Figure 4B) and its promoter was hypermethylated in B cells isolated from mice with PGIA (Figure 4C and Supplementary Figure 4, available on the

Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract).

Importantly, the disease specificity of down- regulated AHR expression was confirmed in PBMCs and B cells (Figures 4D and E) isolated from treatment- naive RA patients. RA- associated DNA hypermethylation was not observed in the AHR promoter region; however, a distant intergenic region (differentially methylated region C) 155 kb upstream of the AHR transcription start site (TSS) was differentially hypermethylated (Figure 4F and Supplementary Figure 5, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). These results suggest that the AHR gene can be silenced in a DNA methyla-tion–dependent manner and that AHR silencing may be involved in RA pathology.

Demethylation of the Ahr promoter plays a signifi-cant role in the inhibition of Aicda. To verify that 5′- azaC induced Ahr expression, we defined the promoter methylation and expression of Ahr in B cells. Indeed, Ahr promoter demethylation and accompanying elevated Ahr gene expression (Figures 5A and B) were identified in 5′- azaC–treated mice with PGIA. To deter-mine the mechanistic connection between 5′- azaC–induced Ahr and Aicda, we conducted experiments with A20 cells, a BALB/c mouse B lymphoma cell line (31). In these cells, Ahr expression was down- regulated as in arthritic mouse B cells (Figure  4B). Furthermore, 5′- azaC treatment evoked a significant increase in Ahr expression (Figure 5C) due to its promoter demethylation (Figure 5D). Moreover, Ahr was translocated into the nucleus (Fig-ure 5E), indicating its engagement in transcriptional regulation.

Next, to confirm the involvement of 5′- azaC–induced Ahr in down- regulated Aicda expression (Figures  5E and F), we used an shRNA- mediated inhibition approach. Treatment with 5′- azaC

Figure 5. Demethylation induced by 5′- azaC provokes Ahr expression in vivo and in vitro. A, Relative (rel) DNA methylation level of the Ahr promoter in B cells from mice with PGIA treated with vehicle (n = 4) and mice with PGIA treated with 5′- azaC (n = 3). B, Relative normalized (norm) Ahr mRNA expression in B cells from mice with PGIA treated with vehicle (n = 8) and mice with PGIA treated with 5′- azaC (n = 8). C and D, Relative normalized Ahr mRNA expression in A20 cells after 24, 48, and 72 hours of treatment with 10 μM 5′- azaC or 0.1% DMSO as control (C) and relative DNA methylation level of the Ahr promoter in A20 cells after 48 hours of treatment with 5′- azaC or control (D) (n = 3 samples per group). E, Localization of Ahr and Aicda in the cytoplasm and nucleus of A20 cells treated with 5′- azaC for 48 hours. GAPDH and lamin β were used as loading controls. F, Relative normalized Aicda mRNA expression in A20 cells treated with 5′- azaC or control for 48 hours (n = 3 samples per group). In A–D and F, bars show the mean ± SEM. * = P ≤ 0.05; ** = P ≤ 0.01; *** = P ≤ 0.001; **** = P ≤ 0.0001, by t- test in A, B, D, and F and by two- way ANOVA with Sidak’s test in C. See Figure 1 for definitions.

SUPPRESSION OF AUTOIMMUNE ARTHRITIS BY 5′- AZACYTIDINE |      1273

did not increase Ahr expression when Ahr- specific shRNA was expressed in A20 cells, resulting in only a moderate decrease in Aicda expression (Figure  6A). Furthermore, we performed Ahr- specific ChIP focused on such regions (Supplementary Table 4, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract) in the genomic context of the Aicda gene that harbor in silico predicted Ahr bind-ing sites (32). We detected 5′- azaC–induced Ahr binding in the first intron, termed region 13 (Figure 6B). Although region 13 does not contain a canonical Ahr binding site (33), it carries reasonably simi-lar sequences to a ChIP- Seq–defined Ahr binding consensus motif (34), which was capable of binding Ahr effectively (Supplementary Figure 6, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract).

Finally, in transient expression studies, AHR only reduced the activity of an Aicda promoter- driven reporter gene if the inserted Aicda first intronic segment included region 13 (Figure 6C). These data demonstrate that Ahr bound directly to the regulatory region of Aicda after 5′- azaC treatment, but this region did not over-lap with a previously described intronic silencer region (20). Our in vitro and in vivo experiments showed that 5′- azaC–induced demethylation rescues Ahr from hypermethylation- mediated down- regulation. B cell culture–based data demonstrated that reactivated Ahr directly contributed to the suppression of Aicda expression through a novel regulatory region in B cells. Therefore, it is plausible that the Ahr–Aicda regulatory cascade is responsible

for attenuated GC formation and antibody production, which ulti-mately ameliorates arthritis symptoms.

DISCUSSION

Targeted inhibition of DNA methylation has already been proven to be an effective approach to restoring the expression of tumor suppressor genes and their cellular function (12). Herein we demonstrated that in vivo treatment with a DNA methyltransferase inhibitor has therapeutic potential in autoimmune arthritis as well.

We identified a number of differentially hypermethylated regions in arthritic mouse B cells (Figure  1A) and detected genome- wide hypomethylation, consistent with the findings of previous studies (35). However, the hypomethylation cannot be considered disease- specific because DNA demethylation also occurs during “normal” B cell maturation and activation (17,36–38). In contrast, disease- associated, less frequent hypermethyla-tion events probably play more significant roles in arthritis etiology. This concept was supported by the findings of a study investigat-ing fibroblast- like synoviocytes in RA patients (10) and an in vitro study of PBMCs from RA patients (13).

DNA demethylation induced by 5′- azaC halted arthritis pro-gression in mice (Figures 1B and C), which can be attributed to the inhibited production of IgG1 antibodies (Figure 2B), since they can participate in joint destruction by forming immune complexes on the articular cartilage surface (39). Furthermore, high- affinity anti-

Figure 6. Direct regulation of Aicda by 5′- azaC–induced Ahr. A, Relative (rel) normalized (norm) Ahr and Aicda mRNA expression in A20 cells transfected with a control plasmid (pUTAbase) alone or transfected with an anti–Ahr short hairpin RNA (shRNA)–expressing plasmid (pU6AshRNA) or control plasmid and then treated with 5′- azaC for 48 hours (n = 3 samples per group). B, Chromatin immunoprecipitation analysis of Ahr binding to region 13 (Supplementary Table 4, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract) in the first intron of Aicda in A20 cells treated with 5′- azaC or control for 48 hours (n = 3 per group). The intergenic region was used as a negative control and the CYP1a1 promoter region was used as a positive control. Data are presented as fold enrichment compared to the values for control treatment. C, Relative luciferase activity of the mouse Aicda promoter- driven reporter gene alone (pGL4.1- Aicda- p), the Aicda promoter- driven reporter gene with a segment of the first intron of Aicda including region 13 (pGL4.1- Aicda- p+13r), or the Aicda promoter- driven reporter gene with the same segment without region 13 (pGL4.1- Aicda- p+Δ13r). Reporter plasmids were cotransfected with a human AHR- expressing plasmid (pCMV- hAHR) or enhanced green fluorescent protein (EGFP)–expressing plasmid (pCMV- EGFP). Luciferase activity was normalized to cellular protein concentrations (n = 3 samples per group). Bars show the mean ± SEM. * = P ≤ 0.05; ** = P ≤ 0.01; *** = P ≤ 0.001; **** = P ≤ 0.0001, by two- way ANOVA with Sidak’s test. See Figure 1 for other definitions.

TÓTH ET AL 1274       |

bodies, including self- reactive antibodies, are generated in GCs of secondary lymphoid organs (18) wherein CSR occurs due to high Aicda expression (25,26). Thus, the reduced Aicda expression and CSR, which ultimately lead to inhibited IgG1 production, can be attributed to the diminished GC formation in 5′- azaC–treated mice (Figures 3A and B).

To clarify the cause of the decrease in the proportion of GC B cells, we determined the frequencies of former B cell subpop-ulations. The frequencies of these subpopulations were similar to those previously found in anti- CD20–depleted mice (40). In light of the consistency of the findings, our results and those of the previ-ous study can be explained in a similar way. The lower proportions of total B cells in the lymphocyte compartments after 5′- azaC treat-ment (Supplementary Figure 2, available on the Arthritis & Rheu-matology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract) induced a compensatory mechanism which is intended to sustain the constant frequency of follicular B cells and resulted in increased proportions of early precursor B cells (Figure 3F) (40) or follicular B cell–derived (41) mature recirculating B cells (Figure 3G). At the same time, a selective incident with a higher negative selection pressure could be induced, resulting in a decreased frequency of late precursor, immature (Figures 3F and G), and T1 B cells (Figure 3H) (42). Moreover, mature recirculating B cells constitute a reservoir for MZ B cells (41); thus, there might be a positive relationship between the increased frequencies of these cells (Figures 3G and H). Ultimately, these changes did not disturb the proportion of GC- forming (18) follicular B cells (Fig-ure 3H). Taken together with an unchanged frequency of CD4+ T cells and serum IL- 4 level, these data suggest that B cell intrinsic factors are the cause of diminished GC formation.

The contribution of Ahr to autoimmune arthritis has been investigated in two previous studies using the collagen- induced arthritis model. In the first study, Ahr altered T cell differentia-tion and promoted pathogenesis (43), while the other study showed that it exerted a preventive effect by inhibiting mesen-chymal stem cell differentiation (44). Our findings are consist-ent with those of the latter study and demonstrate that DNA hypermethylation–mediated silencing of Ahr contributes to arthritis pathogenesis. This hypermethylated region is located in the Ahr promoter in arthritic mouse B cells (Figure 4C and Supplementary Figure 4, available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract) or in a distant upstream region in RA patients (Supplementary Figure 5, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40877/ abstract). ChIP- Seq data demonstrate that transcription factors can bind to this distant region (45–47), suggesting that this region is a B cell–specific enhancer.

We demonstrated that 5′- azaC–induced demethylation could rescue Ahr from hypermethylation- related down- regulation (Figure 5B). This rescued Ahr contributed to the attenuated IgG1 production through the suppression of Aicda (Figure 5F), which is

an essential regulator of GC formation (17). Ahr bound directly to the intronic regulatory region of Aicda after 5′- azaC treatment (Fig-ure 6B), but this region did not overlap with a previously described silencer region in the first intron ~1.3 kb downstream from the TSS (20). This finding provides evidence of the existence of a differ-ent Ahr regulatory mechanism in 5′- azaC–treated arthritic B cells, which may come from distinct activation methods.

In summary, in this study, we demonstrated that 5′- azaC, a Food and Drug Administration–approved anticancer agent used for the treatment of various blood cell malignancies, effectively treats autoimmune arthritis when administered at low doses beginning at the early phase of the disease. Furthermore, we explored the notion that the beneficial effect of 5′- azaC is owed to compromised GC formation and subsequent reduced antibody production in which Ahr promoter demethylation can play a piv-otal role. Our data provide a foundation for further studies explor-ing the therapeutic potential of low- dose DNA methyltransferase inhibitors in RA and other antibody- dependent diseases.

ACKNOWLEDGMENTS

We thank Ms. Szilvia Pördi for excellent technical assistance. We are indebted to Drs. Lionel Clement and Camille Mace (Rush University Medical Center) for the loan of a luminometer for lucif-erase assays.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Rauch had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Tóth, Glant, Rauch.Acquisition of data. Tóth, Ocskó, Balog, Markovics, Jolly, Bukiej, Ruth-berg, Rauch.Analysis and interpretation of data. Tóth, Ocskó, Balog, Mikecz, Kovács, Vida, Block, Glant, Rauch.

REFERENCES 1. Zan H, Casali P. Epigenetics of peripheral B- cell differentiation and

the antibody response. Front Immunol 2015;6:631.

2. Katchamart W, Trudeau J, Phumethum V, Bombardier C. Metho-trexate monotherapy versus methotrexate combination therapy with non- biologic disease modifying anti- rheumatic drugs for rheumatoid arthritis. Cochrane Database Syst Rev 2010:CD008495.

3. GTEx Consortium. Genetic effects on gene expression across hu-man tissues. Nature 2017;550:204–13.

4. Licht JD. DNA methylation inhibitors in cancer therapy: the immunity dimension. Cell 2015;162:938–9.

5. Bottini N, Firestein GS. Epigenetics in rheumatoid arthritis: a primer for rheumatologists. Curr Rheumatol Rep 2013;15:372.

6. Guillamot M, Cimmino L, Aifantis I. The impact of DNA methylation in hematopoietic malignancies. Trends Cancer 2016;2:70–83.

7. Karouzakis E, Gay RE, Michel BA, Gay S, Neidhart M. DNA hy-pomethylation in rheumatoid arthritis synovial fibroblasts. Arthritis Rheum 2009;60:3613–22.

SUPPRESSION OF AUTOIMMUNE ARTHRITIS BY 5′- AZACYTIDINE |      1275

8. Rauch TA, Zhong X, Wu X, Wang M, Kernstine KH, Wang Z, et al. High- resolution mapping of DNA hypermethylation and hypomethyl-ation in lung cancer. Proc Natl Acad Sci U S A 2008;105:252–7.

9. Richardson B, Scheinbart L, Strahler J, Gross L, Hanash S, John-son M. Evidence for impaired T cell DNA methylation in systemic lupus erythematosus and rheumatoid arthritis. Arthritis Rheum 1990;33:1665–73.

10. Maeshima K, Stanford SM, Hammaker D, Sacchetti C, Zeng L, Ai R, et al. Abnormal PTPN11 enhancer methylation promotes rheu-matoid arthritis fibroblast- like synoviocyte aggressiveness and joint inflammation. JCI Insight 2016;1:e86580.

11. Gnyszka A, Jastrzȩbski Z, Flis S. DNA methyltransferase inhibitors and their emerging role in epigenetic therapy of cancer. Anticancer Res 2013;33:2989–96.

12. Scott LJ. Azacitidine: a review in myelodysplastic syndromes and acute myeloid leukaemia. Drugs 2016;76:899–900.

13. Fu LH, Ma CL, Cong B, Li SJ, Chen HY, Zhang JG. Hypometh-ylation of proximal CpG motif of interleukin- 10 promoter regulates its expression in human rheumatoid arthritis. Acta Pharmacol Sin 2011;32:1373–80.

14. Glant TT, Cs-Szabó G, Nagase H, Jacobs JJ, Mikecz K. Progressive polyarthritis induced in BALB/c mice by aggrecan from normal and osteoarthritic human cartilage. Arthritis Rheum 1998;41:1007–18.

15. Cerutti A, Cols M, Puga I. Marginal zone B cells: virtues of innate- like antibody- producing lymphocytes. Nat Rev Immunol 2013;13:118–32.

16. Mesin L, Ersching J, Victora GD. Germinal center B cell dynamics. Immunity 2016;45:471–82.

17. Dominguez PM, Teater M, Chambwe N, Kormaksson M, Redmond D, Ishii J, et al. DNA methylation dynamics of germinal center B cells are mediated by AID. Cell Rep 2015;12:2086–98.

18. DeFranco AL. Germinal centers and autoimmune disease in humans and mice. Immunol Cell Biol 2016;94:918–24.

19. Stockinger B, Di Meglio P, Gialitakis M, Duarte JH. The aryl hydrocar-bon receptor: multitasking in the immune system. Annu Rev Immu-nol 2014;32:403–32.

20. Vaidyanathan B, Chaudhry A, Yewdell WT, Angeletti D, Yen WF, Wheatley AK, et al. The aryl hydrocarbon receptor controls cell- fate decisions in B cells. J Exp Med 2017;214:197–208.

21. Glant TT, Mikecz K, Arzoumanian A, Poole AR. Proteoglycan- induced arthritis in BALB/c mice: clinical features and histopatholo-gy. Arthritis Rheum 1987;30:201–12.

22. Lumsden JM, McCarty T, Petiniot LK, Shen R, Barlow C, Wynn TA, et al. Immunoglobulin class switch recombination is impaired in ATM- deficient mice. J Exp Med 2004;200:1111–21.

23. Tóth DM, Szoke É, Bölcskei K, Kvell K, Bender B, Bosze Z, et al. No-ciception, neurogenic inflammation and thermoregulation in TRPV1 knockdown transgenic mice. Cell Mol Life Sci 2011;68:2589–601.

24. Arribas AJ, Bertoni F. Methylation patterns in marginal zone lympho-ma. Best Pract Res Clin Haematol 2017;30:24–31.

25. Stavnezer J, Guikema JE, Schrader CE. Mechanism and regulation of class switch recombination. Annu Rev Immunol 2008;26:261–92.

26. Stavnezer J. Complex regulation and function of activation- induced cytidine deaminase. Trends Immunol 2011;32:194–201.

27. Kleiman E, Salyakina D, De Heusch M, Hoek KL, Llanes JM, Cas-tro I, et al. Distinct transcriptomic features are associated with tran-sitional and mature B- cell populations in the mouse spleen. Front Immunol 2015;6:30.

28. Crotty S. Follicular helper CD4 T cells (TFH). Annu Rev Immunol 2011;29:621–63.

29. Grövdal M, Karimi M, Tobiasson M, Reinius L, Jansson M, Ekwall K, et al. Azacitidine induces profound genome- wide hypomethylation in primary myelodysplastic bone marrow cultures but may also reduce histone acetylation. Leukemia 2014;28:411–3.

30. Mabbott NA, Gray D. Identification of co- expressed gene signatures in mouse B1, marginal zone and B2 B- cell populations. Immunology 2014;141:79–95.

31. Bhattacharya P, Grigera F, Rogozin IB, McCarty T, Morse HC, Kenter AL. Identification of murine B cell lines that undergo somat-ic hypermutation focused to A:T and G:C residues. Eur J Immunol 2008;38:227–39.

32. Mathelier A, Fornes O, Arenillas DJ, Chen CY, Denay G, Lee J, et al. JASPAR 2016: a major expansion and update of the open- access database of transcription factor binding profiles. Nucleic Acids Res 2016;44:D110–5.

33. Swanson HI, Chan WK, Bradfield CA. DNA binding specificities and pairing rules of the Ah receptor, ARNT, and SIM proteins. J Biol Chem 1995;270:26292–302.

34. Lo R, Matthews J. High- resolution genome- wide mapping of AHR and ARNT binding sites by ChIP- Seq. Toxicol Sci 2012;130:349–61.

35. Liu Y, Aryee MJ, Padyukov L, Fallin MD, Hesselberg E, Runarsson A, et al. Epigenome- wide association data implicate DNA methyl-ation as an intermediary of genetic risk in rheumatoid arthritis. Nat Biotechnol 2013;31:142–7.

36. Kulis M, Merkel A, Heath S, Queirós AC, Schuyler RP, Castellano G, et al. Whole- genome fingerprint of the DNA methylome during human B cell differentiation. Nat Genet 2015;47:746–56.

37. Oakes CC, Seifert M, Assenov Y, Gu L, Przekopowitz M, Ruppert AS, et al. DNA methylation dynamics during B cell maturation un-derlie a continuum of disease phenotypes in chronic lymphocytic leukemia. Nat Genet 2016;48:253–64.

38. Lai AY, Mav D, Shah R, Grimm SA, Phadke D, Hatzi K, et al. DNA methylation profiling in human B cells reveals immune regulatory ele-ments and epigenetic plasticity at Alu elements during B- cell activa-tion. Genome Res 2013;23:2030–41.

39. Anquetil F, Clavel C, Offer G, Serre G, Sebbag M. IgM and IgA rheu-matoid factors purified from rheumatoid arthritis sera boost the Fc receptor– and complement- dependent effector functions of the disease- specific anti–citrullinated protein autoantibodies. J Immunol 2015;194:3664–74.

40. Shahaf G, Zisman-Rozen S, Benhamou D, Melamed D, Mehr R. B cell development in the bone marrow is regulated by homeostatic feedback exerted by mature B cells. Front Immunol 2016;7:77.

41. Cariappa A, Boboila C, Moran ST, Liu H, Shi HN, Pillai S. The re-circulating B cell pool contains two functionally distinct, long- lived, posttransitional, follicular B cell populations. J Immunol 2007;179:2270–81.

42. Chung JB, Silverman M, Monroe JG. Transitional B cells: step by step towards immune competence. Trends Immunol 2003;24:343–9.

43. Nakahama T, Kimura A, Nguyen NT, Chinen I, Hanieh H, Nohara K, et al. Aryl hydrocarbon receptor deficiency in T cells suppresses the development of collagen- induced arthritis. Proc Natl Acad Sci U S A 2011;108:14222–7.

44. Tong Y, Niu M, Du Y, Mei W, Cao W, Dou Y, et al. Aryl hydrocarbon receptor suppresses the osteogenesis of mesenchymal stem cells in collagen- induced arthritic mice through the inhibition of β- catenin. Exp Cell Res 2017;350:349–57.

45. Ding C, Chen X, Dascani P, Hu X, Bolli R, Zhang H, et al. STAT3 signaling in B cells is critical for germinal center maintenance and contributes to the pathogenesis of murine models of lupus. J Immunol 2016;196:4477–86.

46. Gustems M, Woellmer A, Rothbauer U, Eck SH, Wieland T, Lutter D, et al. C- Jun/c- Fos heterodimers regulate cellular genes via a newly identified class of methylated DNA sequence motifs. Nucleic Acids Res 2014;42:3059–72.

47. Heavey B, Charalambous C, Cobaleda C, Busslinger M. Myeloid lin-eage switch of Pax5 mutant but not wild- type B cell progenitors by C/EBPα and GATA factors. EMBO J 2003;22:3887–97.

1276

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1276–1284DOI 10.1002/art.40869 © 2019, American College of Rheumatology

Disease Burden in Osteoarthritis Is Similar to That of Rheumatoid Arthritis at Initial Rheumatology Visit and Significantly Greater Six Months LaterJacquelin R. Chua , Shakeel Jamal, Mariam Riad , Isabel Castrejon , Anne-Marie Malfait, Joel A. Block, and Theodore Pincus

Objective. To analyze disease burden in osteoarthritis (OA) according to Multidimensional Health Assessment Questionnaire (MDHAQ)/Routine Assessment of Patient Index Data 3 (RAPID3) scores at the initial visit and the 6- month follow- up visit, compared with rheumatoid arthritis (RA) as a benchmark for high disease burden.

Methods. All patients with all diagnoses at the Rush University Medical Center Division of Rheumatology com-plete a paper MDHAQ at all visits, saved as a PDF in the electronic health record. MDHAQ 0–10 scores for physical function, pain, and patient global assessment (compiled into RAPID3 0–30 scores) and additional scales at the initial and 6- month follow- up visits, for new OA and RA patients seen from 2011 to 2017, were compared. OA and RA patients were classified as self-referred or physician- referred, and RA patients were classified as disease- modifying antirheumatic drug (DMARD)–naive or having prior-DMARD treatment. Patient groups were compared using t- tests and analysis of variance, adjusted for age, disease duration, body mass index (BMI), education, and ethnicity.

Results. Compared with RA patients, OA patients had higher age, BMI, and disease duration. At initial visit, the mean RAPID3 did not differ significantly in OA versus DMARD- naive RA patients, whether self- or physician- referred (range 14.8–16.4 [P = 0.38]), or in all OA patients versus DMARD- naive RA patients versus prior-DMARD RA patients (15.0, 15.7, and 15.8, respectively [P = 0.49]). After 6 months, RAPID3 was improved to 13.3, 10.3, and 10.8, re-spectively, which represented substantially greater improvement in RA patients than OA patients (P < 0.001). Similar results were seen for most self- reported measures and in adjusted analyses.

Conclusion. MDHAQ/RAPID3 scores are similar in OA and RA patients at the initial visit, but higher in OA patients than in RA patients 6 months later, reflecting superior RA treatments. The same MDHAQ/RAPID3 allows comparisons of disease burdens in different diseases.

INTRODUCTION

Osteoarthritis (OA) has been viewed traditionally as a highly prevalent but generally mild condition (1), explained in part by uni-versal evidence of “radiographic OA” with aging, which is frequently asymptomatic (2). Even some patients being treated for OA in a rheumatology setting described OA in focus groups as “part of a normal aging process requiring acceptance, not treatment—What do you expect? You’re just getting older” (3). Nonetheless, many reports indicate that OA patients seen by rheumatologists may often have a severe disease, with substantial morbidity (4), costs (5,6), and premature mortality (7–9). Furthermore, the prevalence

of OA is increasing, associated with an aging population, high rates of obesity (10), and additional explanatory variables (11).

Pain and functional disability are the most common symp-toms in both OA and rheumatoid arthritis (RA), the latter of which serves as a benchmark of severe arthritis. These problems, how-ever, have generally been assessed using different measures, the Western Ontario and McMaster Universities OA Index (12) for assessment of OA and the Health Assessment Questionnaire (HAQ) for assessment of RA (13). A few studies in which identi-cal self- report questionnaire measures were used (unlike in most clinical care or research) indicate relatively similar scores in OA patients compared with RA patients (14–16).

Supported by Medical History Services, Inc.Jacquelin R. Chua, MD, Shakeel Jamal, MD, Mariam Riad, MD,

Isabel Castrejon, MD, PhD, Anne-Marie Malfait, MD, PhD, Joel A. Block, MD, Theodore Pincus, MD: Rush University Medical Center, Chicago, Illinois.

Dr. Malfait has received research support from Galápagos NV. Dr. Pincus holds a copyright and trademark on MDHAQ and RAPID3 for which he receives royalties and license fees, all of which are used to support

further development of quantitative questionnaire measurements for patients and doctors in clinical rheumatology care. No other disclosures relevant to this article were reported.

Address correspondence to Theodore Pincus, MD, Division of Rheumatology, Rush University Medical Center, 1161 West Harrison Street, Chicago, IL 60606. E-mail: tedpincus@gmail.com.

Submitted for publication September 21, 2018; accepted in revised form February 21, 2019.

OA VERSUS RA AT FIRST AND 6-MONTH VISITS |      1277

A recent report indicated that disease burden was similar in OA compared with RA (17), according to scores on an MDHAQ (Multidimensional HAQ) (18,19), including RAPID3 (Routine Asse-ssment of Patient Index Data 3), an index within the MDHAQ (20,21). Those data were from a cross- sectional “convenience” sample from 4 sites at which the MDHAQ is assessed in all patients with all diagnoses at all visits (17). It was possible that the clinical status of RA patients at presentation was significantly poorer than the status of OA patients, and the observed relatively poor status of OA patients could be explained primarily or only by superior RA treatments (22). One report on 41 OA patients and 39 RA patients seen at a solo- practice rheumatology site indicated similar RAPID3 scores at the initial visit but poorer status of OA patients 2 months later (23). Therefore, we studied a large cohort of patients seen contemporaneously by several rheumatologists in the same clinical setting. We tested a hypothesis that disease bur-den at the initial visit is high and similar in OA compared to RA, and both OA and RA patients are clinically improved 6 months later, but improvement is substantially greater in RA than OA, resulting in a higher disease burden in OA.

PATIENTS AND METHODS

Study protocol. The study was designed to analyze disease burden in OA patients at their initial visit and a 6- month follow- up visit to a rheumatologist, compared with contemporaneous ini-tial and 6- month follow- up visits in the same setting for patients with RA, as a benchmark for severe arthritis. Disease burden was assessed according to 7 scores on an MDHAQ, i.e., physical function, pain, patient global assessment, RAPID3, fatigue, self- report painful joint count, and symptom checklist.

The study includes MDHAQ/RAPID3 data collected from patients as part of their routine care visits between May 2011 and February 2017, and saved as PDF files in the electronic health record (EHR). A waiver of patient consent for routine care and retrospective study of data with protected information (name, date of birth, and medical record number) not identified in databases for analyses or reports submitted for publication was provided by the Rush Institutional Review Board for the Rheumatology Data Repository (OARA no. 14090502).

Identification of patients with primary OA or RA. Since 2007, all patients (with all diagnoses) seen at Rush University Medical Center (RUMC) for rheumatology care have been asked to complete an MDHAQ/RAPID3 at each visit for review by the rheumatologist in the examination room. Since the introduction of the EHR at RUMC in May 2011, each MDHAQ completed by the patient in paper form has been scanned as a PDF file and saved in the EHR.

In 2017, we conducted a search of the EHR of all new patients seen between May 2011 and February 2017 with a pri-mary diagnosis of OA or RA according to the International Clas-sification of Diseases, Ninth Revision code or the International

Statistical Classification of Diseases and Related Health Prob-lems, Tenth Revision code. Two lists of new patients with OA or RA were provided by the EHR company and reviewed for the following cri teria: 1) the initial visit to RUMC Division of Rheuma-tology occurred between May 2011 and February 2017, 2) the patient was older than 18 years of age at the initial visit, 3) a pri-mary diagnosis of OA or RA was accurate, 4) a follow- up visit to RUMC occurred 6 months after the initial visit (ranging between 3 and 9 months) based on the actual median interval of 5.0–5.7 months, but “6 months” was retained for all groups to simplify the presentation, and 5) an MDHAQ score, which included complete RAPID3 scores, that had been scanned into the EHR database on the dates of both the initial and 6- month visits.

The EHR of each patient who met the above criteria was further reviewed for the following data: 1) the time of onset of symptoms was identified to compute disease duration from the recorded symptom onset to the date of the first visit, 2) the source of patient referral—whether physician- referred or self- referred—was entered, 3) prior therapies were recorded (glu-cocorticoid injections and/or surgery in patients with OA, and oral and/or injectable glucocorticoids, small- molecule disease- modifying antirheumatic drugs [DMARDs], and biologic agents in patients with RA, in order to classify them as “DMARD- naive RA” or “prior-DMARD RA”). This classification was regarded as necessary because of an unexpected observation that 75% of patients newly diagnosed as having RA at RUMC had already been treated with DMARDs at their initial visit, and it appeared inappropriate to pool DMARD- treated RA patients with DMARD- naive patients for comparison with OA patients. All RA patients who had prior-DMARD treatment were regarded as physician- referred, although some may have been self- referred to RUMC. Data on patient height and weight were also recorded to calcu-late body mass index (BMI). Patients were classified as being underweight (BMI <18.5 kg/m2), normal (BMI 18.5–24.9 kg/m2), overweight (BMI 25–29.9 kg/m2), or obese (BMI >30 kg/m2).

Patient self- report MDHAQ. The MDHAQ (18,19) is a 2- page, single- sheet questionnaire developed from the HAQ (13) in a clinical setting over 25 years as a continuous quality improve-ment program for routine care (24). RAPID3, an index on the MDHAQ, was developed initially in studies of patients with RA (20,21,24), but has been found to be sensitive to relevant clinical changes in all rheumatic diseases in which it has been studied (25), including OA (17,23), systemic lupus erythematosus (SLE) (23,26), ankylosing spondylitis (AS) (27–31), psoriatic arthritis (PsA) (23), gout (23), polymyalgia rheumatica (PMR) (32), and vasculitis (33).

RAPID3 is composed of 3 MDHAQ scores for physical function, pain visual analog scale (VAS), and patient global assessment VAS (each on a scale of 0–10), which are compiled into a total index with a scale of 0–30 (20). The MDHAQ also includes scales for fatigue, self- report joint count, and a 60- symptom checklist (19). The phys-ical function scale is composed of 10 activities, 8 from the original

CHUA ET AL 1278       |

standard HAQ (13), and 2 complex activities (“walk 2 miles or 3 kilo-meters” and “participate in sports and recreation as you would like”) (18,19). Each activity is scored 0–3, with 0 = “without any difficulty,” 1 = “with some difficulty,” 2 = “with much difficulty,” and 3 = “unable to do,” for a raw total physical function score of 0–30, recalculated on a scale of 0–10 using a template on the questionnaire.

The 2 MDHAQ VAS to score pain and patient global assess-ment were modified to 21 circles at 0.5 intervals, termed “visual numeric scales” (34), which are more user- friendly for patients and more easily scored by doctors than the traditional 10- cm line (35). The 3 scores for physical function, pain, and patient global assessment (scale 0–10 for each) are compiled into a composite RAPID3 index (scale 0–30), using a template on the MDHAQ (21). The 4 RAPID3 categories of severity (rather than activity), i.e., high (>12), moderate (6.1–12), low (3.1–6), and near- remission (≤3), have been described in RA. Use of these categories provides results similar to those obtained with the Disease Activity Score in 28 joints (36) and Clinical Disease Activity Index (37) in clinical trials (38) and clinical care (39).

The MDHAQ also includes a self- report painful joint count, described initially in 1995 as the RA Disease Activity Index (RADAI) (40), to score 8 symmetric painful joint groups (fingers, wrists, elbows, shoulders, hips, knees, ankles, and toes) as 0 = none, 1 = mild, 2 = moderate, and 3 = severe, for a total of 0–48. Alternatively, each joint may be scored simply 0 (normal) or 1 (abnormal) for a total of 0–16 (comparable with previously described binary joint counts) (41).

The 2- page MDHAQ also includes a fatigue VAS (scale 0–10) (42), 60- symptom checklist (43), and recent medical his-tory (24), as well as patient height and weight, to calculate BMI. Queries concerning sleep quality, anxiety, and depression are included in the patient- friendly HAQ format, coded 0, 1.1, 2.2, and 3.3 rather than 0, 1, 2, and 3, to facilitate a “psycholog-ical” index (score scale 0–9.9), as was proposed in the initial MDHAQ report (18) but is not widely used. Demographic data on the MDHAQ include date of birth, sex, ethnicity (Asian, African American, Hispanic, white, or “Other”), and number of years of formal education.

Table  1. Demographic and clinical characteristics and mean MDHAQ/RAPID3 scores in OA patients versus RA patients with prior-DMARD history versus DMARD- naive RA patients at initial visit to Rush University Medical Center*

OA (n = 149)

Prior-DMARD RA (n = 153)

DMARD- naive RA (n = 50)

Age, years 66.1 ± 10.6 54.2 ± 16.3 51.9 ± 12.2Female sex, no. (%) 124 (83.2) 131 (85.6) 41 (82.0)% white/African American/

Hispanic/Asian27/50/22/2 44/29/25/2 46/32/18/5

Formal education, years 13.5 ± 3.7 13.1 ± 3.8 14.5 ± 3.0Disease duration, median (IQR)

years3.4 (1.0–5.5) 3.2 (0.6–8.6) 1.0 (0.4–3.0)

Interval from initial visit to follow- up visit, median (IQR) months

4.9 (3.4–6.4) 5.7 (4.5–6.7) 5.5 (4.6–6.5)

BMI, kg/m2 33.5 ± 7.6 27.9 ± 6.8 30.7 ± 8.6BMI, %

<18.5/18.5–25/25–30/>30 3/5/28/64 2/37/34/27 2/36/24/38

Physician referral, no. (%) 117 (81) NA 34 (72)RAPID3 (0–30 index) 16.0 ± 5.8 15.5 ± 7.6 15.6 ± 5.9

Physical function (0–10 scale) 3.0 ± 1.9 3.2 ± 2.4 2.9 ± 2.1Pain (0–10 scale) 7.0 ±2.2 6.3 ± 3.0 6.8 ± 2.5Patient global assessment

(0–10 scale)5.9 ± 2.7 6.0 ± 3.1 5.9 ± 2.6

Symptom checklist (0–60 scale) 12.1 ± 8.4 12.0 ± 7.9 10.2 ± 6.9Self- report painful joint count

(0–16 scale)6.9 ± 4.9 7.9 ± 4.7 7.7 ± 5.0

Fatigue (0–10 scale) 4.8 ± 3.3 5.2 ± 3.1 4.1 ± 3.7

* Scores for individual components of the MDHAQ (Multidimensional Health Assessment Ques-tionnaire), including the composite RAPID3 index (Routine Assessment of Patient Index Data 3) and its subscales, as well as 3 other MDHAQ indices are shown. Data on physician referral pattern were missing for 4 patients with osteoarthritis (OA) and for 3 patients with disease- modifying antirheu-matic drug (DMARD)–naive rheumatoid arthritis (RA). Except where indicated otherwise, values are the mean ± SD. IQR = interquartile range; BMI = body mass index; NA = not applicable.

OA VERSUS RA AT FIRST AND 6-MONTH VISITS |      1279

Construction of database. Paper copies of the MDHAQ/RAPID3, which had been scanned as PDF files into the EHR, and medical record information from the initial and 6- month (3–9 months) follow- up visits of OA and RA patients, were entered into a study electronic MDHAQ/RAPID3 (eMDHAQ/RAPID3) database (44). Protected information, i.e., name, date of birth, and medi-cal record number, were not entered into this database, but into a separate Excel database, available only on the RUMC server. A unique identifying number assigned by the eMDHAQ/RAPID3 software was included in the Excel database to allow recording of longitudinal data on each patient. The eMDHAQ/RAPID3 data-base was exported to an Excel file, which was then imported to Stata version 12.0 for Mac (StataCorp LP).

Statistical analysis. All analyses were performed using Stata version 12.0 for Mac. The MDHAQ demographic and clini-cal variables were compared at the initial and 6- month follow- up visits in 3 groups, OA patients versus DMARD-naive RA patients versus prior-DMARD RA patients, and in 2 groups, OA patients versus all RA patients. OA and DMARD- naive RA patients were also classified and compared in 2 subgroups within and between each diagnosis: physician- referred versus self- referred.

Data were computed as the mean ± SD or mean and 95% con-fidence interval (95% CI) for normally distributed data, the median with interquartile range (IQR) for non- normally distributed data, and frequencies and percentages for categorical data. The Stata margins procedure was used to estimate possible con founding of the MDHAQ scores by 5 clinical variables: age, BMI group (nor-mal, overweight, or obese), ethnicity (African American, white, or Hispanic), level of formal education (<12 years or ≥12 years), and disease duration. Statistical significance was examined using t- tests, one- way analysis of variance, chi- square, Kruskal- Wallis non-parametric tests (2 groups), and Wilcoxon- Mann- Whitney tests (3 groups) of OA versus prior-DMARD RA versus DMARD- naive RA.

RESULTS

Baseline demographic and non–clinical status data in OA versus prior-DMARD RA versus DMARD- naive RA patients. Study criteria were met by 149 OA patients and 203 RA patients. An unexpected observation was that 153 (75%) of first- visit RA patients had been treated with prior DMARDs and only 50 (25%) were DMARD naive. Therefore, comparisons of OA patients versus RA patients included 2 RA subgroups, as well as all RA patients.

Another unexpected finding was that 117 (81%) of 145 OA patients were physician- referred compared with 28 (19%) of 145 OA patients who were self- referred similar to 34 (72%) of 47 DMARD-naive RA patients being classified as physician- referred, and 13 (28%) of 47 DMARD-naive RA patients being classified as self- referred (referral source could not be determined in 4 OA and 3 RA patients) (Table 1 and Supple-

mentary Table 1, available on Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40869/ abstract). No clinically significant differences were seen for clinical measures between physician- referred or self- referred OA or RA patients, and both groups of OA and RA patients

were pooled for subsequent analyses.The mean ± SD age of all OA patients was 66.1 ±10.6 years,

compared with 54.2 ± 16.3 years in prior-DMARD RA patients and 51.9 ± 12.2 years in DMARD- naive RA patients (Table 1). Twenty- seven percent of the patients with OA were white and 50% were African American. Among the patients with prior-DMARD RA, 44% were white and 29% were African American. In the DMARD- naive RA patient group, 46% were white and 32% were African American. Among the 3 groups, the percentage of patients who were female ranged from 82% to 86%, and the number of years of formal education ranged from 13.1 to 14.5 (Table 1).

Baseline clinical data in OA versus prior-DMARD RA versus DMARD- naive RA patients. The median duration of disease was 3.4 years (IQR 1.0–5.5) in OA patients compared with 3.2 years (IQR 0.6–8.6) in prior-DMARD RA and 1.0 years (IQR 0.4–3.0) in DMARD- naive RA (Table  1). The mean ± SD BMI was 33.5 ± 7.6 kg/m2 in OA patients compared with 27.9 ± 6.8 kg/m2 in prior-DMARD RA patients and 30.7 ± 8.6 kg/m2 in DMARD- naive RA patients. Among the OA patients, 92% were classified as overweight (BMI 25–29.9 kg/m2) or obese (BMI >30 kg/m2), compared with 61% of prior-DMARD RA and 62% of DMARD- naive RA patients (Table 1).

RAPID3 and component clinical status scores at the initial visit in OA versus prior-DMARD RA versus DMARD- naive RA patients. The mean composite RAPID3 scores at the initial visit were 16.0 in OA patients versus 15.5 in prior- DMARD RA patients versus 15.6 in DMARD- naive RA patients (Table  1). Similar trends were seen for each of the RAPID3 components (physical function, pain, and patient global assessment scores) (Table 1). The pain VAS differed among the 3 groups, with the highest mean score in the OA group (7.0 [of 10], compared with 6.3 in the prior-DMARD RA group and 6.8 in the DMARD- naive RA group).

Change in RAPID3 and component scores from the initial visit to 6- month follow- up visit in OA versus prior- DMARD RA versus DMARD- naive RA patients. As noted above, the median interval between the initial visit and follow- up visit was 5.0–5.7 months, but designated “6 months” to simplify the presentation. At 6- month follow- up, the adjusted RAPID3 scores decreased in all 3 groups (mean 15.0 to 13.3 in OA patients, 15.8 to 10.8 in prior-DMARD RA patients, and 15.7 to 10.3 in DMARD- naive RA patients [P = 0.78] [Table  2]; results summarized in Figure 1). Differences between visits in the RAPID3 component scores were significant, after adjustment for age, BMI,

CHUA ET AL 1280       |

Tab

le 2

. A

djus

ted

mea

n di

ffere

nce

in M

DH

AQ

/RA

PID

3 sc

ores

bet

wee

n th

e fir

st v

isit

and

6- m

onth

vis

it in

OA

pat

ient

s ve

rsus

RA

pat

ient

s w

ith p

rior-

DM

AR

D h

isto

ry v

ersu

s D

MA

RD

- nai

ve

RA

pat

ient

s se

en a

t Rus

h U

nive

rsity

Med

ical

Cen

ter*

OA

(n =

149

)

RA

(n =

203

)

Prio

r D

MAR

D

(n =

153

)D

MAR

D n

aive

(n

= 5

0)

Firs

t vis

it6

mon

ths

Diff

eren

ceFi

rst v

isit

6 m

onth

sD

iffer

ence

Firs

t vis

it6

mon

ths

Diff

eren

ce

RAP

ID3

(0–3

0 in

dex)

15.0

(12.

9, 1

7.1)

13.3

(11.

2, 1

5.3)

−1.7

(−3.

8, 0

.4)

15.8

(14.

3, 1

7.2)

10.8

(8.5

, 13.

1)−4

.3 (−

6.6,

−1.

9)†

15.7

(13.

5, 1

7.9)

10.3

(7.9

, 12.

7)−5

.7 (−

8.2,

−3.

3)†

Func

tion

(0–1

0 sc

ale)

2.7

(2.1

, 3.3

)2.

4 (1

.8, 2

.9)

−0.3

(−0.

9, 0

.2)

2.7

(2.0

, 3.4

)2.

2 (1

.6, 2

.8)

−0.5

(−1.

1, 0

.1)

3.4

(2.6

, 4.1

)2.

1 (1

.6, 2

.8)

−1.3

(−1.

9, −

0.7)

†

Pain

(0–1

0 sc

ale)

7.

0 (6

.2, 7

.8)

6.0

(5.1

, 6.8

)−1

.1 (−

2.0,

−0.

1)†

6.2

(5.3

, 7.1

)4.

5 (3

.6, 5

.5)

−1.7

(−2.

7, −

0.6)

†6.

6 (5

.7, 7

.7)

4.3

(3.3

, 5.3

)−2

.4 (−

3.5,

−1.

3)†

Patie

nt g

loba

l as

sess

men

t (0

- 10

scal

e)

5.3

(4.4

, 6.2

)5.

0 (4

.1, 5

.8)

−0.3

(−1.

1, 0

.7)

6.2

(5.2

, 7.2

)4.

1 (3

.1, 5

.1)

−2.1

(−3.

2, −

0.9)

†6.

0 (5

.0, 7

.1)

4.0

(2.9

, 5.0

)−2

.0 (−

3.2,

−0.

8)†

Sym

ptom

ch

eckl

ist

(0–6

0 sc

ale)

9.1

(5.6

, 12.

6)7.

2 (4

.9, 9

.6)

−1.8

(−4.

3, 0

.7)

9.7

(5.8

, 13.

7)6.

4 (3

.7, 9

.1)

−3.3

(−6.

2, −

0.5)

†11

.0 (7

.3, 1

4.6)

5.3

(2.8

, 7.8

)−5

.7 (−

8.3,

−3.

0)†

Self-

repo

rt p

ain-

ful j

oint

cou

nt

(0–1

6 sc

ale)

5.0

(3.0

, 7.1

)4.

6 (2

.9, 6

.4)

−0.4

(−2.

1, 1

.3)

7.0

(4.6

, 9.4

)5.

3 (3

.2, 7

.3)

−1.7

(−3.

6, 0

.2)

9.0

(6.8

, 11.

2)5.

7 (3

.8, 7

.6)

−3.3

(−5.

0, −

1.5)

†

Fatig

ue (0

–10

scal

e)

3.9

(2.2

, 5.6

)3.

6 (2

.1, 5

.1)

−0.3

(−1.

5, 0

.9)

5.0

(3.1

, 6.9

)3.

4 (1

.7, 5

.1)

−1.6

(−3.

0, −

0.2)

†4.

4 (2

.6, 6

.1)

2.9

(1.4

, 4.5

)−1

.4 (−

2.7,

−0.

2)†

* Va

lues

are

the

mea

n (9

5% c

onfid

ence

inte

rval

) adj

uste

d fo

r ag

e, r

ace,

bod

y m

ass

inde

x, le

vel o

f for

mal

edu

catio

n, a

nd d

isea

se d

urat

ion.

See

Tab

le 1

for

defin

ition

s.

† 95

% c

onfid

ence

inte

rval

doe

s no

t cro

ss 0

.

OA VERSUS RA AT FIRST AND 6-MONTH VISITS |      1281

disease duration, race, and level of formal education, only for pain in the OA group, for pain and patient global assessment in the prior DMARD RA group, and for all 3 components in the DMARD-

naive RA group (Table 2).The mean differences in the adjusted RAPID3 declines (3.3

lower in OA patients compared with all RA patients and 4.0 lower in OA patients compared with DMARD- naive RA patients) were

statistically significant (Table 3), whereas the difference between OA patients and prior-DMARD RA patients (2.6 lower in OA patients) was not statistically significant. This finding suggests that most differences between OA patients and all RA patients were accounted for by DMARD- naive RA patients. Similar patterns

were seen for the RAPID3 component scores (Table 3).

Other MDHAQ scores at the initial and 6- month follow- up visits in OA versus prior-DMARD RA versus DMARD- naive RA patients. At the initial visit, there were no clinically or statistically significant differences in the scores for symptom checklist, self- report painful joint count, or fatigue between OA, prior-DMARD RA, and DMARD- naive RA patients (Table 1). The baseline 0–16 self- report painful joint count of 6.9 in OA patients was only slightly lower than 7.9 in prior-DMARD RA patients and 7.7 in DMARD- naive RA patients (Table 1), sug-gesting that from the patient’s perspective, OA appears almost as polyarticular as RA.

In comparisons over 6 months within each of the 3 groups, differences in symptom checklist scores were significant in adjusted analyses in the 2 RA groups, but not in the OA group (Table 2). Dif-ferences in self- report painful joint counts were significant in adjusted analyses in DMARD- naive RA patients, but not in OA patients or prior- DMARD RA patients (Table 2). Over 6 months, differences in fatigue VAS scores were significant in adjusted analyses of DMARD- naive RA and prior-DMARD RA patients, but not in OA patients (Table 2).

Differences in changes over 6 months in the symptom checklist and fatigue VAS were not clinically or statistically signif-icant in the 3 groups, in either the unadjusted analyses (data not shown) or the adjusted analyses (Table 3). Differences in changes in the self- report joint count were significant only between OA and DMARD- naive RA patients (Table 3).

Figure  1. MDHAQ/RAPID3 (Multidimensional Health Assessment Questionnaire/Routine Assessment of Patient Index Data 3) scores at the initial visit and 6- month follow- up visit in patients with osteoarthritis, patients with rheumatoid arthritis (RA) and prior disease- modifying antirheumatic drug (DMARD) treatment, and patients with RA and no prior-DMARD treatment. Data are shown as box plots. Each box represents the interquartile range (IQR). Lines inside the boxes represent the median. Lines outside the boxes represent the 10th and 90th percentiles. *** = P < 0.001.

Table  3. Adjusted mean difference in MDHAQ/RAPID3 scores between the first visit and 6- month visit in OA patients versus RA patients with prior-DMARD history, DMARD- naive RA patients, and both RA groups combined seen at Rush University Medical Center*

Difference in degree of change between first and 6- month visit

OA versus prior-DMARD RA

OA versus DMARD- naive RA OA versus all RA

RAPID3 (0–30 index) −2.6 (−6.0, 0.9) −4.0 (−7.5, −0.6)† −3.3 (−6.3, −0.3)†Physical function (0–10 scale) −0.2 (−1.1, 0.7) −1.0 (−1.9, −0.1)† −0.6 (−1.4, 0.2)Pain (0–10 scale) −0.6 (−2.2, 1.0) −1.4 (−2.9, 0.2) −1.0 (−2.4, 0.4)Patient global assessment (0–10

scale)−1.7 (−3.4, −0.1)† −1.7 (−3.4, −0.04)† −1.7 (−3.2, −0.3)†

Symptom checklist (0–60 scale) −1.5 (−5.6, 2.6) −3.9 (−7.8, 0.1) −2.8 (−6.3, 0.7)Self- report painful joint count

(0–16 scale)−1.3 (−4.1, 1.4) −2.9 (−5.5, −0.3)† −2.2 (−4.5, 0.2)

Fatigue (0–16 scale) −1.3 (−3.3, 0.7) −1.1 (−3.0, 0.8) −1.2 (−2.9, 0.5)

* Values are the mean (95% confidence interval) adjusted for adjusted for age, race, body mass index, level of formal education, and disease duration. See Table 1 for definitions. † 95% confidence interval does not cross 0.

CHUA ET AL 1282       |

DISCUSSION

This report extends and clarifies previous cross- sectional, observational studies and 1 small 2- month longitudinal study concerning a similar or greater disease burden in OA versus RA patients (14–16,23). The burden of disease according to patient self- report MDHAQ/RAPID3 scores appears to be similar in OA and RA at an initial visit to an academic rheumatology unit, regard-less of prior therapy or whether the patient was physician- referred or self- referred. Improvement in RAPID3 and other MDHAQ scores was seen in all 3 patient groups over 6 months, in general greatest in DMARD- naive RA patients, intermediate in prior-DMARD RA patients, and least in OA, reflecting superior treatments for RA and resulting in a significantly greater disease burden in OA compared with RA.

OA patients were significantly older, and had higher BMI and longer duration of disease compared with RA patients, as might be anticipated (45). However, the significantly greater improvement of RA patients compared with OA patients remained significant when adjusted for these variables, as well as education level or ethnicity. Greater improvement in most variables in DMARD- naive versus prior-DMARD RA patients may suggest that patients pre-viously treated with DMARDs may have been close to maximum improvement prior to their visit to RUMC, or were relatively resis-tant to therapy compared with most DMARD- naive patients. As noted, most differences between OA patients and all RA patients appear to result primarily from DMARD- naive RA patients.

The primary symptoms of both OA and RA are pain and functional disability. However, studies of disease burden in each condition traditionally have used “disease- specific” measures, which do not permit direct comparisons between diseases. The few studies that have included the same instruments to compare pain and function have indicated relatively similar values between patients with OA and patients with RA, particularly for pain scores (14–17,23). From the health professionals’ perspectives of patho-physiology and treatments, OA and RA are very different diseases. From the patients’ viewpoints, however, both diseases appear to present a similar burden on their daily lives.

The MDHAQ/RAPID3 was developed initially to assess patients with RA, but has been reported to be valuable in monitor-ing status and status changes over time in all rheumatic diseases in which it has been studied (25,43,46), including OA (23), SLE (23,26), AS (27–31), PsA (23), gout (23), PMR (32), and vasculitis (33). These observations illustrate the potential value of using the same questionnaire in patients with different rheumatic diseases. Clinical research efforts may require multiple disease- specific questionnaires for different diseases (47), but even research studies may benefit from the use of a questionnaire such as the MDHAQ, which has been found to be informative in many differ-ent rheumatic diseases. The HAQ has been recognized to have generic properties (48), and this appears to also pertain to the

HAQ- derived MDHAQ, as well as to the WOMAC, which can be informative in RA (15).

Scales on the MDHAQ beyond RAPID3 also appear infor-mative concerning disease burden and improvement of clinical status. Symptom checklist scores were improved significantly in all groups, in the usual pattern of highest in DMARD- naive RA, intermediate in prior-DMARD RA, and lowest in OA. Self- report painful joint count data suggest that polyarticular disease is common in OA. Improvement was statistically significant only in DMARD- naive RA patients and higher in OA than in prior-DMARD RA patients, suggesting that prior-DMARD RA patients may have joint damage that is affected minimally, if at all, by antiinflammatory therapies. Fatigue scores were similar at the initial visit and did not change substantially over 6 months in either OA or RA patients, which is consistent with previously reported data demonstrating that fatigue VAS scores appear less likely to be associated with levels of inflammation than most other MDHAQ scores (42).

This study has several limitations. First, only patients who had complete MDHAQ/RAPID3 scores at their initial and 6- month follow- up visits were available for analysis. However, there was neither selection for which patients had complete data nor demo-graphic differences between patients with complete data and other patients with OA or RA seen at RUMC, at other medical centers (17), or reported in the medical literature. Second, the patients were treated at a single tertiary center, and the findings may not be generalizable to a community setting, in which patients with OA and RA may have milder disease. However, the focus of the research was a rheumatology clinical setting, albeit an academic center, and no significant differences were seen in clinical mea-sures between patients who were physician- referred and those who were self- referred. Third, all patients, both OA and RA, had prior treatment at least with nonsteroidal antiinflammatory drugs, and only 25% of RA patients were DMARD naive at the initial visit. Therefore, the patients cannot be considered an inception cohort, as might be desirable and seen in other diseases. However, the RAPID3 and other MDHAQ measures of clinical status did not differ significantly in prior-DMARD RA patients versus those who were DMARD naive, supporting the notion that disease burden is similar in OA and RA. Fourth, data concerning the presence of secondary/concomitant OA in patients with RA and/or other comorbidities such as fibromyalgia were not collected systemat-ically in routine care and therefore were not analyzed, although joint damage likely would raise RA scores. Fifth, missing values were seen for most variables, including possible confounding var-iables (up to 47% for disease duration). Sixth, conventional statis-tical adjustment does not capture potential confounding variables that are not identified, and may further be inadequate to account for differences between OA and RA patients in age, BMI, and/or other variables. A rheumatologist in a busy clinical setting may make mental adjustments for age, BMI, and/or other variables, but does not perform formal statistical adjustments. Nonetheless,

OA VERSUS RA AT FIRST AND 6-MONTH VISITS |      1283

the adjusted results generally did not differ substantially from the unadjusted results.

In the past, RA may have been considerably more severe than OA, and the similarity of OA and RA at the initial visit at this time may be explained in part by 1) a secular trend toward milder RA (49), 2) OA as a more prevalent and severe disease than 50 years earlier (11), 3) a higher proportion of RA patients referred early for specialist care (so that the relative proportion of patients with severe disease is declining), and 4) improved general health and care in the commu-nity beyond specific treatment for RA. Nonetheless, some patients with RA continue to have high disease activity (50,51).

We do not suggest comparing OA with RA as being “more severe” at a group or individual level. Some patients with either diagnosis may have mild, moderate, or severe clinical status. The composite evidence, however, confirms previous findings that, as groups, OA patients and RA patients currently have similar dis-ease burden at presentation to rheumatologists. After treatment, OA is associated with a higher mean burden of disease than RA. The data also indicate an advantage of using the same measure to quantitate disease burden in different rheumatic diseases.

ACKNOWLEDGMENTS

The authors would like to thank Meghan Smith for assis-tance in retrieving patients from the Epic EHR, Ali Zargham for assistance in data retrieval, entry, and organization, and the clin-ical staff at Rush University Medical Center, Division of Rheuma-tology, for assistance in data collection.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final ver-sion to be published. Dr. Chua had full access to all of the data in the study and takes responsibility for the integrity of the data and the accu-racy of the data analysis.Study conception and design. Chua, Castrejon, Malfait, Block, Pincus.Acquisition of data. Chua, Jamal, Riad, Malfait, Pincus.Analysis and interpretation of data. Chua, Castrejon, Malfait, Block, Pincus.

REFERENCES 1. Woolf AD, Pfleger B. Burden of major musculoskeletal conditions.

Bull World Health Organ 2003;81:646–56.

2. Hannan MT, Felson DT, Pincus T. Analysis of the discordance be-tween radiographic changes and knee pain in osteoarthritis of the knee. J Rheumatol 2000;27:1513–7.

3. Gignac MA, Davis AM, Hawker G, Wright JG, Mahomed N, Fortin PR, et al. “What do you expect? You’re just getting older”: a com-parison of perceived osteoarthritis- related and aging- related health experiences in middle- and older- age adults. Arthritis Rheum 2006;55:905–12.

4. Dieppe P, Cushnaghan J, Tucker M, Browning S, Shepstone L. The Bristol ‘OA500 study’: progression and impact of the disease after 8 years. Osteoarthritis Cartilage 2000;8:63–8.

5. Yelin E, Callahan LF, for the National Arthritis Data Work Group. The economic cost and social and psychological impact of musculoskel-etal conditions. Arthritis Rheum 1995;38:1351–62.

6. Gupta S, Hawker GA, Laporte A, Croxford R, Coyte PC. The eco-nomic burden of disabling hip and knee osteoarthritis (OA) from the perspective of individuals living with this condition. Rheumatology (Oxford) 2005;44:1531–7.

7. Hochberg MC. Mortality in osteoarthritis. Clin Exp Rheumatol 2008;26 Suppl 51:S120–4.

8. Nüesch E, Dieppe P, Reichenbach S, Williams S, Iff S, Jüni P. All cause and disease specific mortality in patients with knee or hip os-teoarthritis: population based cohort study. BMJ 2011;342:d1165.

9. Pincus T, Gibson KA, Block JA. Premature mortality: a neglect-ed outcome in rheumatic diseases? Arthritis Care Res (Hoboken) 2015;67:1043–6.

10. Neogi T, Zhang Y. Epidemiology of osteoarthritis. Rheum Dis Clin North Am 2013;39:1–19.

11. Wallace IJ, Worthington S, Felson DT, Jurmain RD, Wren KT, Maijanen H, et al. Knee osteoarthritis has doubled in prevalence since the mid- 20th century. Proc Natl Acad Sci U S A 2017;114:9332–6.

12. Bellamy N, Buchanan WW, Goldsmith CH, Campbell J, Stitt LW. Validation study of WOMAC: a health status instrument for mea-suring clinically important patient relevant outcomes to antirheumat-ic drug therapy in patients with osteoarthritis of the hip or knee. J Rheumatol 1988;15:1833–40.

13. Fries JF, Spitz P, Kraines RG, Holman HR. Measurement of patient outcome in arthritis. Arthritis Rheum 1980;23:137–45.

14. Callahan LF, Smith WJ, Pincus T. Self- report questionnaires in five rheu-matic diseases: comparisons of health status constructs and associa-tions with formal education level. Arthritis Care Res 1989;2:122–31.

15. Wolfe F, Kong SX. Rasch analysis of the Western Ontario Mac-Master questionnaire (WOMAC) in 2205 patients with osteoar-thritis, rheumatoid arthritis, and fibromyalgia. Ann Rheum Dis 1999;58:563–8.

16. Slatkowsky-Christensen B, Mowinckel P, Kvien TK. Health status and perception of pain: a comparative study between female pa-tients with hand osteoarthritis and rheumatoid arthritis. Scand J Rheumatol 2009;38:342–8.

17. El-Haddad C, Castrejon I, Gibson KA, Yazici Y, Bergman M, Pincus T. MDHAQ/RAPID3 scores in patients with osteoarthritis are sim-ilar to or higher than in patients with rheumatoid arthritis: a cross- sectional study from current routine rheumatology care at four sites. RMD Open 2017;3:e000391.

18. Pincus T, Swearingen C, Wolfe F. Toward a Multidimensional Health Assessment Questionnaire (MDHAQ): assessment of advanced activi-ties of daily living and psychological status in the patient- friendly health assessment questionnaire format. Arthritis Rheum 1999;42:2220–30.

19. Pincus T, Sokka T, Kautiainen H. Further development of a physical function scale on a MDHAQ [corrected] for standard care of patients with rheumatic diseases. J Rheumatol 2005;32:1432–9.

20. Pincus T, Swearingen CJ, Bergman MJ, Colglazier CL, Kaell AT, Kunath AM, et al. RAPID3 (Routine Assessment of Patient Index Data) on an MDHAQ (Multidimensional Health Assessment Questionnaire): agree-ment with DAS28 (Disease Activity Score) and CDAI (Clinical Disease Activity Index) activity categories, scored in five versus more than ninety seconds. Arthritis Care Res (Hoboken) 2010;62:181–9.

21. Pincus T, Bergman MJ, Yazici Y. RAPID3- an index of physical function, pain, and global status as “vital signs” to improve care for people with chronic rheumatic diseases. Bull NYU Hosp Jt Dis 2009;67:211–25.

22. Smolen JS, Landewé R, Breedveld FC, Dougados M, Emery P, Gaujoux-Viala C, et al. EULAR recommendations for the manage-ment of rheumatoid arthritis with synthetic and biological disease- modifying antirheumatic drugs. Ann Rheum Dis 2010;69:964–75.

23. Castrejón I, Bergman MJ, Pincus T. MDHAQ/RAPID3 to recognize improvement over 2 months in usual care of patients with osteoarthri-

CHUA ET AL 1284       |

tis, systemic lupus erythematosus, spondyloarthropathy, and gout, as well as rheumatoid arthritis. J Clin Rheumatol 2013;19:169–74.

24. Pincus T, Maclean R, Yazici Y, Harrington JT. Quantitative measure-ment of patient status in the regular care of patients with rheumatic diseases over 25 years as a continuous quality improvement activity, rather than traditional research. Clin Exp Rheumatol 2007;25 Suppl 47:69–81.

25. Castrejón I. The use of MDHAQ/RAPID3 in different rheumatic diseas-es: a review of the literature. Bull NYU Hosp Jt Dis 2017;75:93–100.

26. Askanase AD, Castrejón I, Pincus T. Quantitative data for care of patients with systemic lupus erythematosus in usual clinical set-tings: a patient Multidimensional Health Assessment Questionnaire and physician estimate of noninflammatory symptoms. J Rheumatol 2011;38:1309–16.

27. Danve A, Reddy A, Vakil-Gilani K, Garg N, Dinno A, Deodhar A. Routine Assessment of Patient Index Data 3 score (RAPID3) cor-relates well with Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) in the assessment of disease activity and monitoring pro-gression of axial spondyloarthritis. Clin Rheumatol 2015;34:117–24.

28. Cinar M, Yilmaz S, Cinar FI, Koca SS, Erdem H, Pay S, et al. A patient- reported outcome measures- based composite index (RAP-ID3) for the assessment of disease activity in ankylosing spondylitis. Rheumatol Int 2015;35:1575–80.

29. Park SH, Choe JY, Kim SK, Lee H, Castrejón I, Pincus T. Routine Assessment of Patient Index Data (RAPID3) and Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) scores yield similar in-formation in 85 korean patients with ankylosing spondylitis seen in usual clinical care. J Clin Rheumatol 2015;21:300–4.

30. Michelsen B, Fiane R, Diamantopoulos AP, Soldal DM, Hansen IJ, Sokka T, et al. A comparison of disease burden in rheumatoid arthritis, psoriatic arthritis and axial spondyloarthritis. PLoS One 2015;10:e0123582.

31. Castrejón I, Pincus T, Wendling D, Dougados M. Responsiveness of a simple RAPID- 3- like index compared to disease- specific BAS-DAI and ASDAS indices in patients with axial spondyloarthritis. RMD Open 2016;2:e000235.

32. Castrejón I, Huang A, Everakes SL, Nika A, Sequeira W. Clinical Im-provement according to RAPID3 in patients with polymyalgia rheu-matica: a longitudinal analysis from routine care. J Clin Rheumatol 2018;24:390–2.

33. Annapureddy N, Elsallabi O, Baker J, Sreih AG. Patient- reported outcomes in ANCA- associated vasculitis: a comparison between Birmingham Vasculitis Activity Score and routine assessment of patient index data 3. Clin Rheumatol 2016;35:395–400.

34. Ritter PL, González VM, Laurent DD, Lorig KR. Measurement of pain using the visual numeric scale. J Rheumatol 2006;33:574–80.

35. Pincus T, Bergman M, Sokka T, Roth J, Swearingen C, Yazici Y. Visual analog scales in formats other than a 10 centimeter hori-zontal line to assess pain and other clinical data. J Rheumatol 2008;35:1550–8.

36. Prevoo ML, van ‘t Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity scores that include twenty- eight–joint counts: development and validation in a prospec-tive longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995;38:44–8.

37. Aletaha D, Nell VP, Stamm T, Uffmann M, Pflugbeil S, Machold K, et al. Acute phase reactants add little to composite disease activity indices for rheumatoid arthritis: validation of a clinical activity score. Arthritis Res Ther 2005;7:R796–806.

38. Pincus T, Chung C, Segurado OG, Amara I, Koch GG. An index of patient reported outcomes (PRO- Index) discriminates effectively between active and control treatment in 4 clinical trials of adalimum-ab in rheumatoid arthritis. J Rheumatol 2006;33:2146–52.

39. Pincus T, Swearingen CJ, Bergman M, Yazici Y. RAPID3 (Routine As-sessment of Patient Index Data 3), a rheumatoid arthritis index with-out formal joint counts for routine care: proposed severity categories compared to disease activity score and clinical disease activity index categories. J Rheumatol 2008;35:2136–47.

40. Stucki G, Liang MH, Stucki S, Brühlmann P, Michel BA. A self- administered Rheumatoid Arthritis Disease Activity Index (RADAI) for epidemiologic research: psychometric properties and correlation with parameters of disease activity. Arthritis Rheum 1995;38:795–8.

41. Fuchs HA, Brooks RH, Callahan LF, Pincus T. A simplified twenty- eight–joint quantitative articular index in rheumatoid arthritis. Arthritis Rheum 1989;32:531–7.

42. Castrejón I, Nikiphorou E, Jain R, Huang A, Block JA, Pincus T. Assessment of fatigue in routine care on a Multidimensional Health Assessment Questionnaire (MDHAQ): a cross- sectional study of as-sociations with RAPID3 and other variables in different rheumatic diseases. Clin Exp Rheumatol 2016;34:901–9.

43. Pincus T, Askanase AD, Swearingen CJ. A Multi- dimensional Health Assessment Questionnaire (MDHAQ) and Routine Assess-ment of Patient Index Data (RAPID3) scores are informative in patients with all rheumatic diseases. Rheum Dis Clin North Am 2009;35:819–27.

44. Pincus T. Electronic Multidimensional Health Assessment Question-naire (eMDHAQ): past, present and future of a proposed single data management system for clinical care, research, quality improvement, and monitoring of long- term outcomes. Clin Exp Rheumatol 2016;34 Suppl 101:S17–33.

45. Felson DT, Lawrence RC, Dieppe PA, Hirsch R, Helmick CG, Jordan JM, et al. Osteoarthritis: new insights. Part 1: the disease and its risk factors. Ann Intern Med 2000;133:635–46.

46. Ingegnoli F, Carmona L, Castrejon I. Systematic review of systemic sclerosis- specific instruments for the EULAR Outcome Measures Li-brary: an evolutional database model of validated patient- reported outcomes. Semin Arthritis Rheum 2017;46:609–14.

47. Rolfson O, Wissig S, van Maasakkers L, Stowell C, Ackerman I, Ayers D, et al. Defining an international standard set of outcome measures for patients with hip or knee osteoarthritis: consensus of the International Consortium for Health Outcomes Measurement Hip and Knee Osteoarthritis Working Group. Arthritis Care Res (Ho-boken) 2016;68:1631–9.

48. Fries JF, Ramey DR. “Arthritis specific” global health analog scales assess “generic” health related quality- of- life in patients with rheu-matoid arthritis. J Rheumatol 1997;24:1697–702.

49. Silman A, Davies P, Currey HL, Evans SJ. Is rheumatoid arthritis be-coming less severe? J Chronic Dis 1983;36:891–7.

50. Sokka T, Haugeberg G, Pincus T. Assessment of quality of rheuma-toid arthritis care requires joint count and/or patient questionnaire data not found in a usual medical record: examples from studies of premature mortality, changes in clinical status between 1985 and 2000, and a QUEST- RA global perspective. Clin Exp Rheumatol 2007;25 Suppl 47:86–97.

51. Sokka T, Kautiainen H, Toloza S, Mäkinen H, Verstappen SM, Lund Hetland M, et al. QUEST- RA: quantitative clinical assessment of pa-tients with rheumatoid arthritis seen in standard rheumatology care in 15 countries. Ann Rheum Dis 2007;66:1491–6.

1285

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1285–1296DOI 10.1002/art.40849 © 2019 The Authors. Arthritis & Rheumatology published by Wiley Periodicals, Inc. on behalf of American College of Rheumatology. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Prioritization of PLEC and GRINA as Osteoarthritis Risk Genes Through the Identification and Characterization of Novel Methylation Quantitative Trait LociSarah J. Rice,1 Maria Tselepi,1 Antony K. Sorial,1 Guillaume Aubourg,1 Colin Shepherd,1 David Almarza,1 Andrew J. Skelton,1 Ioanna Pangou,1 David Deehan,2 Louise N. Reynard,1 and John Loughlin1

Objective. To identify methylation quantitative trait loci (mQTLs) correlating with osteoarthritis (OA) risk alleles and to undertake mechanistic characterization as a means of target gene prioritization.

Methods. We used genome- wide genotyping and cartilage DNA methylation array data in a discovery screen of novel OA risk loci. This was followed by methylation, gene expression analysis, and genotyping studies in additional cartilage samples, accompanied by in silico analyses.

Results. We identified 4 novel OA mQTLs. The most significant mQTL contained 9 CpG sites where methylation correlated with OA risk genotype, with 5 of the CpG sites having P values <1 × 10−10. The 9 CpG sites reside in an interval of only 7.7 kb within the PLEC gene and form 2 distinct clusters. We were able to prioritize PLEC and the ad-jacent gene GRINA as independent targets of the OA risk. We identified PLEC and GRINA expression QTLs operating in cartilage, as well as methylation- expression QTLs operating on the 2 genes. GRINA and PLEC also demonstrated differential expression between OA hip and non- OA hip cartilage.

Conclusion. PLEC encodes plectin, a cytoskeletal protein that maintains tissue integrity by regulating intracellular signaling in response to mechanical stimuli. GRINA encodes the ionotropic glutamate receptor TMBIM3 (transmem-brane BAX inhibitor 1 motif–containing protein family member 3), which regulates cell survival. Based on our results, we hypothesize that in a joint predisposed to OA, expression of these genes alters in order to combat aberrant bio-mechanics, and that this is epigenetically regulated. However, carriage of the OA risk–conferring allele at this locus hinders this response and contributes to disease development.

INTRODUCTION

Osteoarthritis (OA) is a common, age- related disease of synovial joints characterized by the thinning of articular cartilage. This thinning ultimately results in focal loss of cartilage and a full- thickness lesion exposing underlying bone (1,2). Cartilage loss pre-vents the joint from withstanding normal mechanical load and leads to severely impaired joint function. The clinical effect is a painful chronic morbidity and, in those with hip and knee OA, an increased risk of premature death due to secondary cardiovascular events resulting from reduced physical activity (3,4). Regarding treatment

options for hip and knee OA, there are no disease- modifying OA drugs, and arthroplasty of hips and knees is a common, but not a risk- free, procedure. As the population ages, the prevalence of OA increases. Novel treatments are therefore urgently required.

The causes of OA are complex and, with a heritability of >40%, genetic susceptibility is a main driver (5). Genome- wide association studies have revealed that the OA genetic component is polygenic, with the association signals typically having modest odds ratios of <1.5 (6).

Of those OA risk–conferring loci that have so far been functionally investigated, the vast majority mediate their effect

Supported by Arthritis Research UK (grant 20771) as part of the MRC and the Centre for Integrated Research into Musculoskeletal Ageing (grants JXR 10641, MR/P020941/1, and MR/R502182/1), the European Union Seventh Framework Program (grant 305815 [D-BOARD]), the Ruth and Lionel Jacobson Charitable Trust, and the Newcastle upon Tyne NHS Charity (BH162140). Patient tissue was provided by the Newcastle Bone and Joint Biobank, supported by the NIHR Newcastle Biomedical Research Centre, and awarded to the Newcastle upon Tyne NHS Foundation Trust and Newcastle University. Drs. Sorial and Aubourg’s work was supported by the NIHR Integrated Academic Training Program.

1Sarah J. Rice, PhD, Maria Tselepi, MRes, Antony K. Sorial, MRCS, Guillaume Aubourg, MBBS, MRes, Colin Shepherd, PhD, David Almarza, PhD,

Andrew J. Skelton, MSc, Ioanna Pangou, BSc, Louise N. Reynard, PhD, John Loughlin, PhD: International Centre for Life, Newcastle University, Newcastle upon Tyne, UK; 2David Deehan, FRCS: Freeman Hospital, Newcastle upon Tyne, UK.

No potential conflicts of interest relevant to this article were reported.Address correspondence to Sarah J. Rice, PhD, or John Loughlin,

PhD, Newcastle University, Skeletal Research Group, Institute of Genetic Medicine, International Centre for Life, Newcastle upon Tyne NE1 3BZ, UK. E-mail: sarah.rice@newcastle.ac.uk or john.loughlin@ncl.ac.uk.

Submitted for publication August 30, 2018; accepted in revised form January 30, 2019.

RICE ET AL 1286       |

by modulating the expression of a nearby gene. Clear exam-ples include the single- nucleotide polymorphisms (SNPs) rs143383, rs225014, and rs3204689, which correlate with differential expression of the OA and non- OA risk alleles in car-tilage of the genes GDF5, DIO2, and ALDH1A2, respectively (for review, see ref. 6).

DNA methylation at CpG dinucleotides is an epigenetic pro-cess used by the cell to regulate gene expression, and it can amplify or attenuate the impact of an expression quantitative trait locus (QTL) (7–9). Our group has previously shown this in the con-text of GDF5, rs143383, and CpG sites located close to the SNP (10). Identifying cis- acting CpG sites, where methylation corre-lates with an association signal, can therefore offer insight into the mechanism by which the genetic risk at that signal operates. Such CpG sites are known as methylation QTLs (mQTLs). Furthermore, if mQTL CpG sites cluster within or close to a particular gene, their mapping prioritizes that gene for further investigation as a plausi-ble target for the association signal (11).

We have shown the potential utility of this approach in our previous analysis of 16 OA loci, in which we correlated cartilage DNA methylation with OA association signals using genome- wide DNA methylation and genotyping array data (12). In this current study, we extended our investigation to 18 novel OA association signals that have been identified in the last 3 years. We found evidence of mQTLs at 4 of these loci, with the most statistically significant effect being on chromosome 8q24.3 at a signal harbor-ing a number of genes. We subjected this signal to in silico and in vitro analyses, which highlighted the plectin gene PLEC and the TMBIM3 (transmembrane BAX inhibitor 1 motif–containing protein family member 3) gene GRINA as the targets of the association. Plectin is a multifunctional cytoskeletal protein that is particularly abundant in tissues subjected to mechanical load and stress, while TMBIM3 is involved in the control of cell death by endoplas-mic reticulum (ER) stress. Our study highlights the high frequency of cartilage mQTLs operating on OA risk loci and the capacity for mQTL analysis to prioritize a gene from within a gene- rich region as a target of the association signal. PLEC, GRINA, and their encoded proteins now merit much more detailed analyses in the context of OA genetic risk and cartilage biology.

MATERIALS AND METHODS

Cartilage methylation and genotyping array data. We used cartilage CpG methylation and genotype data that we had previously generated using an Illumina Infinium HumanMethylation450 array and an Illumina HumanOmniEx-press array, respectively (12). We had both methylation and genotype data available for a total of 87 patients who had undergone knee or hip joint arthroplasty (57 knee OA patients, 14 hip OA patients, and 16 patients who had undergone hip replacement due to a femoral neck fracture (Supplementary Table 1, avail able on the Arthritis & Rheumatology web site at

http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). The cartilage samples from the patients with femoral neck frac-ture lacked OA lesions.

OA loci investigation and mQTL analysis. We inves-tigated 18 novel OA risk loci that had been reported as being associated with the disease at a significance level close to or surpassing the genome- wide threshold of P < 5 × 10−8 (13–20) (Table 1). If the OA- associated SNP was directly genotyped on an Illumina HumanOmniExpress array, we utilized those SNP data. If the association SNP was not on the array, we identified and, where possible, used a proxy SNP that was in perfect or high linkage disequilibrium (pairwise r2 > 0.7) with the asso-ciation SNP. Proxy SNPs were derived from a candidate list using LDlink’s LDproxy tool (https ://analy sisto ols.nci.nih.gov/LDlin k/) (21) and European population data. Where multiple proxy SNPs were identified, we chose the one with the highest

r2 relative to the association SNP.For each locus, we covered a 2- Mb region encompass-

ing 1 Mb upstream of and 1 Mb downstream of the associ-ation SNP. For each CpG site within the 2- Mb region, linear regression was used to measure the relationship between methylation, in the form of M values, and genotype (0, 1, or 2 copies of the minor allele). For the purpose of this analysis, we defined any genotype–methylation correlations identified within this 2- Mb region to be a cis mQTL. CpG sequences that were directly modified by SNPs were not included in the analysis. We used age, sex, and sample type (OA knee, OA hip, and femoral neck fracture) as covariates. All mQTL calcu-lations were performed using Matrix eQTL software (22), which implements a false discovery rate (FDR) estimation based on the Benjamini- Hochberg FDR procedure (23) and accounts for the number of tests performed. Methylation β values were used for the purpose of visualization.

In silico expression QTL analysis. The genotype- tissue expression database GTEx (https ://www.gtexp ortal.org/home/) was searched for expression QTLs (eQTLs) at rs11780978. The search was performed in July 2017.

RNA- sequencing (RNA- Seq) analysis. The expression of genes of interest was assessed using RNA- Seq data that we had previously generated from the cartilage of 10 patients with hip OA and 6 patients with femoral neck fracture (24,25). These patients are unrelated to the 87 patients used in the mQTL analysis. Details regarding the 16 patients can be found in Supplementary Table 1 (available at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). Transcripts per million (TPM) values for each investigated gene were extracted using R (http://www.R-proje ct.org/) and visualized using the ggplot2 library in R. Differential expression analysis between OA and femoral neck fracture cartilage samples was carried out with

EPIGENETIC PRIORITIZATION OF OA RISK GENES PLEC AND GRINA |      1287

a Bioconductor software package DESeq2 (26). Hypothesis testing was performed using a DESeq2 implementation of the Wald test.

New patients. Cartilage tissue samples were obtained from 104 new OA patients who had undergone joint replace-ment surgery at the Newcastle upon Tyne NHS Foundation Trust hospitals. The Newcastle and North Tyneside Research Ethics Committee granted ethical approval for the collection of tissue samples, with each donor providing verbal and writ-ten informed consent (REC reference number 14/NE/1212). Our patient ascertainment criteria and the protocols for extracting nucleic acid from cartilage have been described in detail previously (27–29). Further details regarding the 104 new patients can be found in Supplementary Table 1 (avail-able at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). Nucleic acid from these patients was used for genotyping, complementary DNA (cDNA) synthesis, allelic expression imbalance (AEI) analysis, and targeted CpG meth-ylation analysis.

Genotyping of SNPs. SNPs were genotyped by pyro-sequencing (rs11783799 and rs11136345), by restriction frag-ment length polymorphism (RFLP) analysis (rs11780978 and rs7819099), or by a real- time SNP genotyping assay (rs9100) (cat-alog no. 4351379; ThermoFisher Scientific). The rs11136336 SNP is in perfect linkage disequilibrium (r2 = 1.0) with rs11783799; the rs11783799 assay was therefore used to genotype rs11136336. Pyrosequencing and RFLP polymerase chain reactions (PCRs) were performed using a G- Storm GS4 Q4 Quad Block Thermal Cycler (Somerton Biotechnology) and the primers listed in Sup-plementary Table 2 (available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). Pyrosequencing assays were designed using PyroMark assay design software 2.0 (Qiagen), and the sequencing was per-formed using a PyroMark Q24 Advanced platform (Qiagen) with the recommended kit, following the instructions of the manufac-turer. The RFLP- digested fragments were separated by electro-phoresis through a 3% agarose gel and visualized using ethidium bromide staining. The rs9100 real- time genotyping assay was run on a QuantStudio 3 (Applied Biosystems), using TaqPath ProAmp

Table 1. List of the 18 osteoarthritis association signals in the study*

LocusAssociation

SNP

Allele, major/minor MAF Chr.

Nearest protein- coding gene

Proxy SNP (r2 relative to

association SNP)† Stratum‡ Ref.

CpG probes from the

methylation array, no.§

1 rs2820436 G/T 0.34 1 ZC3H11B rs2605096 (1.0) Any joint site 13 1532 rs2862851 C/T 0.47 2 TGFA rs1807968 (0.92) Hip 14 5263 rs6766414 T/G 0.25 3 STT3B – Hip 15 1864 rs2236995 A/C 0.49 4 SLBP No proxy at r2 > 0.7 Hip 14 –5 rs11335718 C/- (indel) 0.08 4 ANXA3 No proxy at r2 > 0.7 Any joint site 13 –6 rs4867568 T/C 0.48 5 LSP1P3 – Knee 16 187 rs10471753 C/G 0.38 5 PIK3R1 rs6893396 (1.0) Hip 14 2018 rs3850251 T/A 0.29 6 ENPP3 rs4383836 (0.81) Hand 17 2069 rs833058 C/T 0.38 6 VEGF No proxy at r2 > 0.7 Hip 18 –10 rs788748 A/G 0.49 7 IGFBP3 – Hip 19 22911 rs11780978 G/A 0.42 8 PLEC – Hip 13 1,75912 rs10116772 C/A 0.45 9 GLIS3 – Knee and hip 20 10813 rs496547 T/A 0.35 11 TREH rs598373 (0.78) Hip 14 78414 rs4764133 C/T 0.35 12 MGP rs12316046 (0.98) Hand 17 25915 rs754106 C/T 0.50 13 LRCH1 rs9534442 (0.85) Hip 15 22416 rs864839 A/C 0.33 16 JPH3 No proxy at r2 > 0.7 Any joint site 13 –17 rs2521349 G/A 0.40 17 MAP2K6 rs2521348 (0.98) Hip 13 13818 rs6516886 T/A 0.29 21 RWDD2B rs2832155 (1.0) Knee and hip 13 104

* MAF = minor allele frequency; Chr. = chromosome. † For single- nucleotide polymorphisms (SNPs) not present on the HumanOmniExpress array, a proxy SNP that had the highest linkage dis-equilibrium with the association SNP and that was on the array was used to infer genotypes. The threshold was set at r2 > 0.7. Thirteen of the association SNPs required a proxy and, at the threshold, a proxy was available for 9 of the 13 SNPs (rs2820436, rs2862851, rs10471753, rs3850251, rs496547, rs4764133, rs754106, rs2521349, and rs6516886), but a proxy was not available for 4 of the 13 SNPs (rs2236995, rs11335718, rs833058, and rs864839). ‡ Stratum highlights the joint with which the signal shows association from the original genetic study. § This value shows the number of CpG probes from the Illumina Infinium HumanMethylation450 array that are present within the 2- Mb region surrounding the association SNP.

RICE ET AL 1288       |

MasterMix (ThermoFisher) following the instructions of the manu-facturer.

Complementary DNA synthesis and AEI analysis. Complementary DNA was synthesized from 1 μg of cartilage total RNA using a SuperScript First- Strand synthesis system (Invitro-gen) and random hexamers following the standard protocol of the manufacturer after an initial 30- minute treatment with 1 unit of Turbo DNase (Invitrogen) at 37°C.

AEI at transcript SNPs rs11783799 and rs11136345 was quantified by pyrosequencing, using the methodology described in the above genotyping section and the same primers. The sequences were generated automatically, and an output of allelic ratio was produced using PyroMark Advanced software. AEI at transcript SNP rs9100 was quantified using the above real- time SNP genotyping assay. The use of such assays for AEI has been described previously (30). For each cartilage cDNA and DNA sam-ple, PCRs were formed in triplicate. Samples were excluded from the analysis if the values between the PCR replicates differed by >5% for the pyrosequencing assays, or by >0.5 cycle thresholds (Ct) for the genotyping assay. The respective cDNA and DNA were analyzed concurrently, and allelic expression of cDNA was nor-malized to its corresponding DNA.

Targeted CpG methylation analysis. Methylation anal-ysis of CpG sites cg19405177 and cg14598846 was performed by pyrosequencing using the methodology described above and the PCR and sequencing primers listed in Supplementary Table 3 (available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). Cartilage DNA was bisulfite converted using an EZ DNA meth-ylation kit (Zymo Research). For each sample, PCRs were per-formed in duplicate and the mean calculated, with samples being excluded from the analysis if the methylation between the PCR replicates differed by >5%.

Chromatin interactions. The WashU Epigenome Browser (31) was searched to identify long- range chromatin interactions extending from the CpG sites within the PLEC locus. All publicly available Hi- C and long- range chromatin interaction data sets were loaded for all cell types with available data within the genomic region chromosome 8:145,000,040–145,068,742. The region was searched visually to identify interactions stem-ming from either of the 2 clusters of CpG sites. The human breast cancer cell line MCF7 and the chronic myeloid leukemia cell line K562 interaction schema represent protein factor–medi-ated chromatin interaction data measured by RNA polymerase II chromatin interaction analysis using paired- end tag sequencing (ChIA- PET) data (32). These data were produced as part of the ENCODE project (https://www.encodeproject.org/). Data from the human lymphoblastoid cell line GM12878 were produced by RNA CCCTCF–binding factor ChIA- PET (33).

RESULTS

Identification of OA mQTLs. Table 1 provides details of the 18 loci investigated. For 5 of the 18 loci, the SNP that had been used to identify the association signal had been genotyped on the HumanOmniExpress array. For each of the remaining 13 loci, we searched for a proxy SNP (r2 > 0.7). This was successful for 9 loci, but for the remaining 4 loci there was no proxy (loci 4, 5, 9, and 16). Therefore, these 4 loci were not studied further. In total, we analyzed 4,895 CpG sites across the 14 loci. We assessed correlations in all 87 samples combined for knee OA, hip OA, and femoral neck fracture. This analysis identified 4 SNPs that correlated with methylation (FDR P < 0.05): locus 7, SNP rs10471753 and 1 CpG site; locus 11, SNP rs11780978 and 9 CpG sites; locus 14, SNP rs4764133 and 1 CpG site; and locus 18, SNP rs6516886 and 6 CpG sites (Table 2 and Supplementary Table 4, avail-able on the Arthritis & Rheumatology web site at http://onlin e libr ary.wiley.com/doi/10.1002/art.40849/ abstract).

Many OA genetic association signals have only been dis-covered following stratification by joint and/or sex (5,6). We repeated the mQTL analysis of all 14 loci using such stratifi-cation, but did not identify any further genotype–methylation correlations. Furthermore, to assess whether the identified mQTLs were more specific to OA than femoral neck fracture, we repeated the mQTL analysis by removing the 16 samples from the patients with femoral neck fracture. This analysis did not identify any OA- specific or femoral neck fracture–specific genotype–methylation correlations. In summary, we identified 4 OA susceptibility loci demonstrating significant mQTLs, and these are discussed in more detail below.

Locus 7. Genotype at rs10471753 correlated with meth-ylation of 1 CpG site, cg25008444, which is located 233 kb from the association SNP. This CpG site is within the gene body of PIK3R1, coding for phosphatidylinositol 3- kinase reg-ulatory subunit ɑ (Supplementary Figure 1A, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). The OA risk–conferring C allele of rs10471753 correlated with higher methylation of cg25008444 (Supplementary Figure 1B).

Locus 11. Genotype at rs11780978 correlated with methyla-tion of 9 CpG sites located within an interval of only 7.7 kb and posi-tioned between 25.7 kb and 33.4 kb from the association SNP. The 9 CpG sites are all within the gene body of PLEC, which encodes plectin (Figure 1A). They form 2 clusters as follows: cg19405177, cg20784950, and cg01870834 are located within an interval of 1.4 kb, followed by a gap of 5.3 kb (which contains 6 CpG sites covered by the array and not demonstrating an mQTL) and then 6 CpG sites (cg07427475, cg02331830, cg04255391, cg14598846, cg23299254, and cg10299941), which are located within an interval

EPIGENETIC PRIORITIZATION OF OA RISK GENES PLEC AND GRINA |      1289

of 1.0 kb (Figure 1B). For each of the 3 CpG sites within the first cluster, the OA risk–conferring A allele of rs11780978 correlated with higher methylation, while for each of the 6 CpG sites within the sec-ond cluster, the A allele correlated with lower methylation (Figure 1C).

Locus 14. Genotype at rs4764133 correlated with methyla-tion at 1 CpG site, cg20917083, which is located 50 kb from the association SNP. This CpG is within an intron of ARHGDIB, which codes for Rho GDP dissociation inhibitor 2 (Supplementary Figure 2A, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). The OA risk–conferring T allele of rs4764133 correlated with higher meth-ylation of cg20917083 (Supplementary Figure 2B).

Locus 18. Genotype at rs6516886 correlated with meth-ylation at 6 CpG sites. These are located at either side of the SNP, with cg00065302 (27.4 kb from the SNP), cg05468028 (2.3 kb from the SNP), cg18001427 (1.9 kb from the SNP), and cg20220242 (1.5 kb from the SNP) located on the centromeric side of rs6516886, while cg24751378 (2.7 kb from the SNP) and cg16140273 (62 kb from the SNP) are located on its telomeric side (Supplementary Figure 3A, available on the Arthritis & Rheu-matology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). The CpG site cg00065302 is intergenic and located between LTN1, which codes for E3 ubiquitin- protein ligase

listerin, and RWDD2B, which codes for RWD domain– containing protein 2B. The CpG site cg05468028 is located in the gene body of RWDD2, while cg18001427 and cg20220242 are located immediately upstream of this gene. The CpG site cg24751378 is intergenic and located between and upstream of RWDD2B and USP16, which codes for ubiquitin carboxyl- terminal hydrolase 16. The CpG site cg16140273 resides within an intron of MAP-3K7CL, which codes for MAP3K7 C- terminal–like protein. The OA risk–conferring T allele of rs6516886 correlated with increased methylation of cg00065302, cg05468028, cg18001427, cg20220242, and cg24751378, and with decreased methylation of cg16140273 (Supplementary Figure 3B).

Focus on locus 11. Based on the large number of highly significant CpG sites from within locus 11, which all cluster in an interval of only 7.7 kb, we focused our attention on rs11780978 and set out to further investigate this association signal. As noted above, these CpG sites reside within PLEC at chromo-some 8q24.3. This region of the genome is gene rich. By ana-lyzing the GTEx database, we first assessed whether there were any known rs11780978 eQTLs operating on genes within the locus. This was the case in a broad range of cell types for PLEC and also for PARP10, GRINA, and SPATC1, which are located in the immediately telomeric region of PLEC (Figure 1B). The GTEx database does not include data on cartilage tissue samples.

Table 2. List of the significant genotype–methylation associations identified*

Locus [association SNP/proxy SNP if required] (Chr.), CpG site CpG location (hg19) P, uncorrected P, FDR adjusted

7 [rs10471753/rs6893396] (5) 67586258 0.0001 0.03cg25008444

11 [rs11780978] (8)cg19405177 145001428 2.04 × 10−20 3.33 × 10−17

cg20784950 145002522 3.14 × 10−07 0.0001cg01870834 145002835 2.05 × 10−07 0.0001cg07427475 145008110 1.51 × 10−18 1.85 × 10−15

cg02331830 145008288 6.17 × 10−11 4.31 × 10−8

cg04255391 145008397 2.34 × 10−17 2.30 × 10−14

cg14598846 145008909 1.11 × 10−22 2.72 × 10−19

cg23299254 145008957 4.87 × 10−23 2.38 × 10−19

cg10299941 145009137 0.0001 0.0414 [rs4764133/rs12316046] (12) 15114233 0.0002 0.04

cg2091708318 [rs6516886/rs2832155] (21)

cg00065302 30366250 0.0001 0.04cg05468028 30391383 1.53 × 10−5 0.006cg18001427 30391784 2.95 × 10−5 0.01cg20220242 30392188 3.92 × 10−9 2.40 × 10−6

cg24751378 30396349 3.27 × 10−7 0.0001cg16140273 30455616 3.07 × 10−5 0.01

* P values are from all 87 samples combined. SNP = single- nucleotide polymorphism; Chr. = chromosome; FDR = false discovery rate.

RICE ET AL 1290       |

Figure 1. Correlation of genotype at rs11780978 with the methylation status of 9 CpG sites within the gene body of PLEC. A, The association between rs11780978 and methylation levels of CpG probes that are present within the region. The x- axis represents the genomic position of the CpG probes, and the y- axis represents the Benjamini- Hochberg–corrected −log10 P value of the correlation between rs11780978 genotype and M value at each CpG probe. Each circle represents a single CpG probe, with the 9 significant associations highlighted in red. The broken line indicates the location of rs11780978. The genes within the region analyzed are indicated below the association plot, with the gene direction indicated by arrows. B, An enlarged image of PLEC, highlighting the location of the 9 significant CpG associations (red) falling into 2 distinct clusters. Each circle represents a single CpG probe. C, The association between genotype at rs11780978 and methylation levels at the 9 significant CpG probes for all 87 samples. The level of methylation at the CpG probes is shown as the β value. Symbols represent individual patient samples. Horizontal lines and error bars show the mean ± SEM.

EPIGENETIC PRIORITIZATION OF OA RISK GENES PLEC AND GRINA |      1291

Using our RNA- Seq data, we observed expression of PLEC, PARP10, and GRINA in cartilage (TPMs >10) (Supplementary Figure 4A, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract), but not of SPATC1 (TPM <1). Each of these 3 cartilage- expressed genes showed significant differential expression between our OA and femoral neck fracture cartilage samples, with PLEC and PARP10 demonstrating increased expression in OA (P = 0.014 and P = 0.002, respectively) and GRINA demonstrating reduced expression (P = 0.028). Ensembl (http://www.ensem bl.org/index.html) lists 15 transcript isoforms for PLEC, 23 for PARP10, and 10 for GRINA. The majority of these were detectable in our RNA- Seq data for both the OA and femoral neck fracture carti-lage samples (Supplementary Figures 4B–D).

Methylation QTL, eQTL, and methylation- expression QTL analysis in new patients. We assessed whether cartilage expression of PLEC, PARP10, or GRINA correlated with rs11780978 genotype and, if so, whether any such eQTLs correlated with methylation. As a first step, we recruited additional OA patients undergoing arthroplasty and extracted DNA and RNA samples from their cartilage. Using

pyrosequencing, we measured methylation levels at the PLEC CpG sites in the DNA samples. We focused on 2 of the most significant CpG sites: cg19405177 from the cluster of 3 (clus-ter 1) and cg14598846 from the cluster of 6 (cluster 2) (Table 2 and Figure 1). Each pyrosequencing assay captured the rele-vant CpG site plus additional CpG sites located nearby: 7 sites for cg19405177 and 3 sites for cg14598846. Stratification by rs11780978 genotype replicated our initial mQTL result, and the OA risk–conferring A allele of rs11780978 correlated with higher methylation at cg19405177 (and its 7 flanking CpG sites) and with lower methylation at cg14598846 (and its 3 downstream CpG sites) (Figure 2).

We next tested for eQTLs by undertaking AEI analysis in the new patients. For each gene, we selected a transcript SNP that was in linkage disequilibrium with rs11780978, and we identified those rs11780978 heterozygotes that were also het-erozygous for 1 or more of the transcript SNPs. AEI was then performed using cDNA synthesized from the cartilage RNA samples. For PLEC we used the transcript SNP rs11783799 (r2 = 0.93) and analyzed 19 compound heterozygotes. For PARP10 we used the transcript SNP rs11136345 (r2 = 0.85) and analyzed 22 compound het erozygotes. For GRINA we

Figure  2. Association between rs11780978 genotype and methylation for PLEC CpG sites cg19405177 and cg14598846 in the new osteoarthritis patients (n = 104). In cluster 1, the pyrosequencing assay targeting cg19405177 (CpG5) captured 7 additional CpG sites (CpG1–CpG4 and CpG6–CpG8) located between 67 bp upstream and 45 bp downstream of cg19405177. In cluster 2, the pyrosequencing assay targeting cg14598846 (CpG1) captured 3 additional CpG sites (CpG2–CpG4) located up to 22 bp downstream of cg14598846. P values were calculated using the Kruskal- Wallis test. The square of the correlation coefficient (r2) values were calculated using a model of standard least squares. Horizontal lines and error bars show the mean ± SEM. n = the number of patients providing data per CpG.

RICE ET AL 1292       |

used the transcript SNP rs9100 (r2 = 0.78) and analyzed 11 compound heterozygotes. Significant AEI was observed at PLEC and GRINA, but not at PARP10 (Figure 3). For PLEC, the G allele of rs11783799, which correlates with the OA risk–con-ferring A allele of rs11780978, demonstrated reduced expres-sion in all 19 compound heterozygotes combined (G:A ratio <1.0; P = 0.02). For GRINA, the T allele of rs9100, which corre-lates with the A allele of rs11780978, demonstrated increased expression in all 11 compound heterozygotes combined (T:G ratio >1.0; P = 0.005).

We plotted the PLEC and GRINA AEI data for the compound heterozygotes against their individual methylation values. We focused on the PLEC heterozygotes for which we had matched methylation data (up to 15 of the 19 heterozygotes). Addition-ally, we focused on the GRINA heterozygotes for which we also had methylation data (up to 10 of the 11). For PLEC, 3 cluster- 1 CpG sites demonstrated significant correlation between methyl-ation and AEI: CpG1 (P = 0.006), CpG3 (P = 0.031), and CpG6 (P = 0.047) (Figure 4A). Additionally, we identified a significant correlation between methylation and PLEC AEI at a single clus-

Figure 3. Allelic expression imbalance (AEI) analysis in cartilage from new osteoarthritis patients (n = 104). AEI analysis was conducted for PLEC using transcript single- nucleotide polymorphism (SNP) rs11783799 (A), for PARP10 using transcript SNP rs11136345 (B), and for GRINA using transcript SNP rs9100 (C). The risk/nonrisk allelic ratio is plotted, with a ratio of <1 indicating decreased expression of the risk allele. Three technical repeats were performed for each patient’s DNA sample (black) and cDNA sample (gray). Numbers on the x- axis refer to the anonymized identification number assigned to patients at recruitment. The mean values for DNA and cDNA are combined (left panels), with results represented as box plots, in which the lines within the box represent the median, the box represents the 25th to 75th percentile, and the whiskers represent the maximum and minimum values (right). P values were calculated using Wilcoxon’s matched pairs signed rank test.

EPIGENETIC PRIORITIZATION OF OA RISK GENES PLEC AND GRINA |      1293

Figure 4. Methylation expression quantitative trait locus analysis of PLEC (A) and GRINA (B). A, PLEC log2 allelic expression imbalance (AEI) ratios and B, GRINA log2 AEI ratios were plotted against methylation at cg19405177 and its 7 additional CpG sites (cluster 1) and at cg14598846 and its 3 additional CpG sites (cluster 2).

RICE ET AL 1294       |

ter- 2 CpG site (CpG1; P = 0.048). For GRINA, there were no correlations with cluster- 1 CpG sites, but 3 cluster- 2 CpG sites demonstrated significant correlations: cg14598846 (P = 0.014), CpG2 (P = 0.012), and CpG4 (P = 0.038) (Figure 4B).

In summary, this analysis replicated the mQTLs, identified car-tilage eQTLs at PLEC and GRINA, and revealed the existence of cartilage methylation-expression QTLs (meQTLs) operating on both of these genes but correlating with distinct CpG sites. Methylation at both cluster 1 and 2 was associated with PLEC expression, while only methylation at cluster 2 correlated with GRINA expression.

A genetic difference between clusters 1 and 2. We next sought to assess whether there was a genetic basis for this differ-ence in activity between cluster 1 and cluster 2 CpG sites. Using our original 87 patients, we determined whether any SNPs on the array that were part of the same linkage disequilibrium block as the association signal rs11780978 showed a more significant gen-otype–methylation correlation with cg19405177 or cg14598846. For cg19405177, we identified rs7819099 (r2 = 0.83, uncorrected P = 1.66 × 10−28 versus 2.04 × 10−20 for rs11780978), and for cg14598846 we identified rs11136336 (r2 = 0.94, uncorrected P = 3.69 × 10−27 versus 1.11 × 10−22 for rs11780978). Both SNPs are also located within PLEC and have a pairwise r2 value of 0.87. We genotyped these 2 SNPs in our new patients (n = 104) and, as in the 87 patients, we correlated genotype with methylation. In Supplementary Figure 5 (available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract), the data are presented as a heatmap for the 87 array patients and for the 104 new patients. For cg19405177, rs7819099 also showed a stronger effect than rs11780978 in the new patients, and in the 2 patient groups combined. For cg14598846, rs11136336 and rs11780978 were more compa-rable in the new patients and in the 2 groups combined. Overall, these data reveal different polymorphisms as drivers of the methyl-ation–genotype correlations observed at the 2 clusters: rs7819099 for cluster 1 and rs11136336/rs11780978 for cluster 2.

Interaction between the CpG clusters and nearby genes. Finally, we used in silico data and the WashU epige-nome browser to search for chromatin interactions between the 2 clusters and nearby genes. No cartilage or chondrocyte data were available for the locus, but we identified relevant data for the human breast cancer cell line MCF7, the chronic myeloid leukemia cell line K562, and the human lymphoblastoid cell line GM12878. For each of these cell lines there was evidence of interaction between the clusters and upstream sequences of several PLEC isoforms (Supplementary Figure 6, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40849/ abstract). These data indicate that, at least in these cell lines, these regions harboring the CpG sites physically interact with the PLEC gene.

DISCUSSION

Using genome- wide data, we identified 4 additional OA susceptibility loci in which the risk- conferring allele correlates with differential DNA methylation relative to the nonrisk allele in cis. Combined with our previous analyses (10,12), we have now identified 8 such mQTL loci out of a total of 30 OA association signals studied, which is a frequency of >25%. We suspect that this is an underestimation, considering the low CpG coverage of the Illumina methylation array used (~1.5% of the total CpG sites in the human genome) and our relatively modest sample size of 87 patients. Our analysis, therefore, emphasizes the potentially pivotal mechanistic role that DNA methylation plays in the activity of many OA genetic risk loci, and now warrants targeted editing of the epigenome in order to determine causality. Furthermore, it highlights that this effect is operating on cartilage, the tissue cen-tral to the disease process, and is detectable at a clinically relevant point for the patient, namely, when arthroplasty is required.

Of the 4 novel mQTL loci that we report here as significant following FDR correction, we chose to focus on locus 11 because of the large number of highly significant CpG sites at this signal, i.e., 5 of the 9 positive CpG sites had FDR P values of <1 × 10−10. A subsequent analysis revealed eQTLs operating on genes within the locus. There is evidence of eQTLs correlating with the OA association SNPs at the other 3 loci (data not shown), and these signals, therefore, also merit follow- up studies.

Methylation QTLs operating at the locus 11 CpG sites in association with the OA SNP, or other variants in high linkage disequilibrium with it, have been previously reported in a range of tissues including adipose, pancreas, lung, lymphocytes, and fibroblasts (34–39). However, this is the first study of these mQTLs operating in cartilage, and the first study to identify a link between these CpG sites and OA genetic risk. Hypothesizing that the mQTL at this locus signified an eQTL of functional relevance to the association signal, we used a combination of in silico and in vitro analyses to highlight PLEC and GRINA as likely target genes of the rs11780978 association signal.

PLEC codes for plectin, a multifunctional cytoskeletal linker protein that directly interacts with a broad range of cytoplasmic, membrane, and organelle proteins (40,41). Plectin has a range of functions, including acting as a scaffold for signaling proteins and as an organizer of cytoskeletal filaments. The protein is particu-larly abundant in tissues that are subjected to mechanical stress, and most research to date focuses on plectin in skin, muscle, and blood vessels. Mutations of PLEC result in skin blistering and mus-cular dystrophy, with the tissue affected and the severity of the disease being determined by the site of the mutation (41). GRINA is a member of a family of genes coding for 6 proteins named “transmembrane BAX inhibitor motif–containing,” (TMBIM). These genes encode calcium channels that are present in the Golgi complex, ER, and mitochondria, where they control calcium homeostasis (42). By controlling calcium flow, the TMBIMs

EPIGENETIC PRIORITIZATION OF OA RISK GENES PLEC AND GRINA |      1295

regulate cell death, including during ER stress, with the majority of the proteins being antiapoptotic (43). GRINA encodes TMBIM3, which is expressed in most cells, and localizes to the Golgi com-plex and ER, where it suppresses ER stress–induced apoptosis. To the best of our knowledge, plectin and TMBIM3 have not been functionally linked to an arthritis phenotype before.

Our AEI data indicate that the OA risk–conferring A allele of rs11780978 correlates with reduced expression of PLEC but with increased expression of GRINA, while our RNA- Seq data revealed increased PLEC expression but decreased GRINA expression, in OA versus non- OA (femoral neck fracture) cartilage. For PLEC, our interpretation of this is that in an OA joint, additional plectin is required to mitigate the effect of an altered biomechanical load. However, inheritance of the A allele, with its concomitant lower expression of PLEC, attenuates this mitigation. For GRINA, our interpretation is that an alteration in the activity of cells in the OA joint necessitates a reduction in GRINA expression to facilitate controlled cell death. However, the A allele, with its concomitant increased GRINA expression, hinders this response.

One of the most striking outcomes of our investigation was the discovery of 2 clusters of CpG sites that were physically close but had quite separate and distinct characteristics as follows: 1) for clus-ter 1 the risk- conferring A allele of rs11780978 correlated with higher methylation, whereas for cluster 2 it correlated with lower methyl-ation; 2) CpG sites in both cluster 1 and cluster 2 correlated with an meQTL operating on PLEC, whereas only cluster- 2 CpG sites correlated with an meQTL operating on GRINA; and 3) methylation at cluster- 1 CpG sites correlated more strongly with genotype at rs7819099, whereas methylation at cluster- 2 CpG sites correlated more strongly with genotype at rs11136336/rs11780978. Com-bined, these data support the notion that at least 2 functional effects are encoded by polymorphisms at the PLEC locus, mediated by methylation on distinct CpG sites and impacting 2 separate genes.

In conclusion, our analysis has identified novel mQTLs and highlighted the interplay between OA genetic risk, DNA methylation, and gene expression in cartilage. We have dis-covered relevant functional effects on 2 genes from within a single locus and, as such, have elevated these to prime tar-gets for the encoded risk at this locus. These genes and their encoded proteins now merit much more detailed investigation in the context of OA etiology.

ACKNOWLEDGMENTS

We thank the surgeons and research nurses at the Newcastle upon Tyne NHS Foundation Trust for providing us with access to the study samples.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti-cally for important intellectual content, and all authors approved the final version to be submitted for publication. Drs. Rice and Loughlin had full

access to all of the data in the study and take responsibility for the integ-rity of the data and the accuracy of the data analysis.Study conception and design. Rice, Shepherd, Reynard, Loughlin.Acquisition of data. Rice, Tselepi, Sorial, Aubourg, Shepherd, Skelton, Pangou, Deehan, Reynard, Loughlin.Analysis and interpretation of data. Rice, Tselepi, Sorial, Aubourg, Shep-herd, Almarza, Skelton, Reynard, Loughlin.

REFERENCES 1. Brandt KD, Dieppe P, Radin EL. Etiopathogenesis of osteoarthritis.

Rheum Dis Clin North Am 2008;34:531–59.

2. Loeser RF, Goldring SR, Scanzello CR, Scanzello CR, Goldring MB. Osteoarthritis: a disease of the joint as an organ. Arthritis Rheum 2012;64:1697–707.

3. Johnson VL, Hunter DJ. The epidemiology of osteoarthritis. Best Pract Res Clin Rheumatol 2014;28:5–15.

4. Kendzerska T, Jüni P, King LK, Croxford R, Stanaitis I, Hawker GA. The longitudinal relationship between hand, hip and knee osteoarthritis and cardiovascular events: a population- based cohort study. Osteoarthritis Cartilage 2017;25:1771–80.

5. Loughlin J. Genetic indicators and susceptibility to osteoarthritis. Br J Sports Med 2011;45:278–82.

6. Reynard LN. Analysis of genetics and DNA methylation in osteoarthritis: what have we learnt about the disease? Semin Cell Dev Biol 2017;62:57–66.

7. Van Meurs JB. Osteoarthritis year in review 2016: genetics, genomics and epigenetics. Osteoarthritis Cartilage 2017;25:181–9.

8. Simon TC, Jeffries MA. The epigenomic landscape in osteoarthritis. Curr Rheumatol Rep 2017;19:30.

9. Ramos YF, Meulenbelt I. The role of epigenetics in osteoarthritis: current perspective. Curr Opin Rheumatol 2017;29:119–29.

10. Reynard LN, Bui C, Syddall CM, Loughlin J. CpG methylation regulates allelic expression of GDF5 by modulating binding of SP1 and SP3 repressor proteins to the osteoarthritis SNP rs143383. Hum Genet 2014;133:1059–73.

11. Hannon E, Weedon M, Bray N, O’Donovan M, Mill J. Pleiotropic effects of trait- associated genetic variation on DNA methylation: utility for refining GWAS loci. Am J Hum Genet 2017;100:954–9.

12. Rushton MD, Reynard LN, Young DA, Shepherd C, Aubourg G, Gee F, et al. Methylation quantitative trait locus analysis of osteoarthritis links epigenetics with genetic risk. Hum Mol Genet 2015;24:7432–44.

13. Zengini E, Hatzikotoulas K, Tachmazidou I, Steinberg J, Hartwig FP, Southam L, et al. Genome- wide analyses using UK Biobank data provide insights into the genetic architecture of osteoarthritis. Nat Genet 2018;50:549–58.

14. Castaño-Betancourt MC, Evans DS, Ramos YF, Boer CG, Metrustry S, Liu Y, et al. Novel genetic variants for cartilage thickness and hip osteoarthritis. PLoS Genet 2016;12:e1006260.

15. Panoutsopoulou K, Thiagarajah S, Zengini E, Day-Williams AG, Ramos YF, Meessen JM, et al. Radiographic endophenotyping in hip osteoarthritis improves the precision of genetic association analysis. Ann Rheum Dis 2017;76:1199–206.

16. Yau MS, Yerges-Armstrong LM, Liu Y, Lewis CE, Duggan DJ, Renner JB, et al. Genome- wide association study of radiographic knee osteoarthritis in North American Caucasians. Arthritis Rheumatol 2017;69:343–51.

17. Den Hollander W, Boer CG, Hart D, Yau MS, Ramos YF, Metrustry S, et al. Genome- wide association and functional studies identify a role for matrix Gla protein in osteoarthritis of the hand. Ann Rheum Dis 2017;76:2046–53.

18. Rodriguez-Fontenla C, Calaza M, Evangelou E, Valdes AM, Arden N, Blanco FJ, et al. Assessment of osteoarthritis candidate genes in

RICE ET AL 1296       |

a meta- analysis of nine genome- wide association studies. Arthritis Rheumatol 2014;66:940–9.

19. Evans DS, Cailotto F, Parimi N, Valdes AM, Castaño-Betancourt MC, Liu Y, et al. Genome- wide association and functional studies identify a role for IGFBP3 in hip osteoarthritis. Ann Rheum Dis 2015;74:1861–7.

20. Casalone E, Tachmazidou I, Zengini E, Hatzikotoulas K, Hackinger S, Suveges D, et al. A novel variant in GLIS3 is associated with osteoarthritis. Ann Rheum Dis 2018;77:620–3.

21. Machiela MJ, Chanock SJ. LDlink: a web- based application for exploring population- specific haplotype structure and linking correlated alleles of possible functional variants. Bioinformatics 2015;31:3555–7.

22. Shabalin AA. Matrix eQTL: ultra fast eQTL analysis via large matrix operations. Bioinformatics 2012;28:1353–8.

23. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B 1995;57:289–300.

24. Shepherd C, Skelton AJ, Rushton MD, Reynard LN, Loughlin J. Expression analysis of the osteoarthritis genetic susceptibility locus mapping to an intron of the MCF2L gene and marked by the polymorphism rs11842874. BMC Med Genet 2015;16:108.

25. Shepherd C, Zhu D, Skelton AJ, Combe J, Threadgold H, Zhu L, et al. Functional characterization of the osteoarthritis genetic risk residing at ALDH1A2 identifies rs12915901 as a key target variant. Arthritis Rheumatol 2018;70:1577–87.

26. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 2014;15:550.

27. Southam L, Rodriguez-Lopez J, Wilkins JM, Pombo-Suarez M, Snelling S, Gomez-Reino JJ, et al. An SNP in the 5′- UTR of GDF5 is associated with osteoarthritis susceptibility in Europeans and with in vivo differences in allelic expression in articular cartilage. Hum Mol Genet 2007;16:2226–32.

28. Egli RJ, Southam L, Wilkins JM, Lorenzen I, Pombo-Suarez M, Gonzalez A, et al. Functional analysis of the osteoarthritis susceptibility–associated GDF5 regulatory polymorphism. Arthritis Rheum 2009;60:2055–64.

29. Rushton MD, Reynard LN, Barter MJ, Refaie R, Rankin KS, Young DA, et al. Characterization of the cartilage DNA methylome in knee and hip osteoarthritis. Arthritis Rheumatol 2014;66:2450–60.

30. Styrkasdottir U, Thorleifsson G, Helgadottir H, Bomer N, Metrustry S, Bierma-Zeinstra S, et al. Severe osteoarthritis of the hand associates with common variants within the ALDH1A2 gene and with rare variants at 1p31. Nat Genet 2014;46:498–502.

31. Zhou X, Lowdon RF, Li D, Lawson HA, Madden PA, Costello JF, et al. Exploring long- range genome interactions using the WashU Epigenome Browser [letter]. Nat Methods 2013;10:375–6.

32. Li G, Ruan X, Auerbach RK, Sandhu KS, Zheng M, Wang P, et al. Extensive promoter- centered chromatin interactions provide a topological basis for transcription regulation. Cell 2012;148:84–98.

33. Tang Z, Luo OJ, Li X, Zheng M, Zhu JJ, Szalaj P, et al. CTCF- mediated human 3D genome architecture reveals chromatin topology for transcription. Cell 2015;163:1611–27.

34. Grundberg E, Meduri E, Sandling JK, Hedman AK, Keildson S, Buil A, et al. Global analysis of DNA methylation variation in adipose tissue from twins reveals links to disease- associated variants in distal regulatory elements. Am J Hum Genet 2013;93:876–90.

35. Guitierrez-Arcelus M, Lappalainen T, Montgomery SB, Buil A, Ongen H, Yurovsky A, et al. Passive and active DNA methylation and the interplay with genetic variation in gene regulation. Elife 2013;2:e00523.

36. Lemire M, Zaidi SH, Ban M, Ge B, Aissi D, Germain M, et al. Long- range epigenetic regulation is conferred by genetic variation located at thousands of independent loci. Nat Commun 2015;6: 6326.

37. Olsson AH, Volkov P, Bacos K, Dayeh T, Hall E, Nilsson EA, et al. Genome- wide associations between genetic and epigenetic variation influence mRNA expression and insulin secretion in human pancreatic islets. PLoS Genet 2014;10:e1004735.

38. Shi J, Marconett CN, Duan J, Hyland PL, Li P, Wang Z, et  al. Characterizing the genetic basis of methylome diversity in histologically normal human lung tissue. Nat Commun 2014;5:3365.

39. Relton CL, Gaunt T, McArdle W, Ho K, Duggirala A, Shihab H, et al. Data resource profile: accessible resource for integrated epigenomic studies (ARIES). Int J Epidemiol 2015;44:1181–90.

40. Castañón MJ, Walko G, Winter L, Wiche G. Plectin- intermediate filament partnership in skin, skeletal muscle, and peripheral nerve. Histochem Cell Biol 2013;140:33–53.

41. Rezniczek GA, Winter L, Walko G, Wiche G. Functional and genetic analysis of plectin in skin and muscle. Methods Enzymol 2016;569:235–59.

42. Lisak DA, Schacht T, Enders V, Habicht J, Kiviluoto S, Schneider J, et  al. The transmembrane Bax inhibitor motif (TMBIM) containing protein family: tissue expression, intracellular localization and effects on the ER Ca2+- filling state. Biochim Biophys Acta 2015;1853:2104–14.

43. Liu Q. TMBIM- mediated Ca2+ homeostasis and cell death. Biochim Biophys Acta 2017;1864:850–7.

1297

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1297–1307DOI 10.1002/art.40859 © 2019, American College of Rheumatology

Evaluating the Properties of a Frailty Index and Its Association With Mortality Risk Among Patients With Systemic Lupus ErythematosusAlexandra Legge,1 Susan Kirkland,1 Kenneth Rockwood,1 Pantelis Andreou,1 Sang-Cheol Bae,2 Caroline Gordon,3 Juanita Romero-Diaz,4 Jorge Sanchez-Guerrero,4 Daniel J. Wallace,5 Sasha Bernatsky,6 Ann E. Clarke,7 Joan T. Merrill,8 Ellen M. Ginzler,9 Paul Fortin,10 Dafna D. Gladman,11 Murray B. Urowitz,11 Ian N. Bruce,12 David A. Isenberg,13 Anisur Rahman,13 Graciela S. Alarcón,14 Michelle Petri,15 Munther A. Khamashta,16 M. A. Dooley,17 Rosalind Ramsey-Goldman,18 Susan Manzi,19 Kristjan Steinsson,20 Asad A. Zoma,21 Cynthia Aranow,22 Meggan Mackay,22 Guillermo Ruiz-Irastorza,23 S. Sam Lim,24 Murat Inanc,25 Ronald F. van Vollenhoven,26 Andreas Jonsen,27 Ola Nived,27 Manuel Ramos-Casals,28 Diane L. Kamen,29 Kenneth C. Kalunian,30 Soren Jacobsen,31 Christine A. Peschken,32 Anca Askanase,33 and John G. Hanly34

Objective. To evaluate the properties of a frailty index (FI), constructed using data from the Systemic Lupus Inter-national Collaborating Clinics (SLICC) inception cohort, as a novel health measure in systemic lupus erythematosus (SLE).

Methods. For this secondary analysis, the baseline visit was defined as the first study visit at which both organ damage (SLICC/American College of Rheumatology Damage Index [SDI]) and health- related quality of life (Short- Form 36 [SF- 36] scores) were assessed. The SLICC- FI was constructed using baseline data. The SLICC- FI comprises 48 health deficits, including items related to organ damage, disease activity, comorbidities, and functional status. Content, construct, and criterion validity of the SLICC- FI were assessed. Multivariable Cox regression was used to estimate the association between baseline SLICC- FI values and mortality risk, adjusting for demographic and clinical factors.

Results. In the baseline data set of 1,683 patients with SLE, 89% were female, the mean ± SD age was 35.7 ± 13.4 years, and the mean ± SD disease duration was 18.8 ± 15.7 months. At baseline, the mean ± SD SLICC- FI score was 0.17 ± 0.08 (range 0–0.51). Baseline SLICC- FI values exhibited the expected measurement properties and were weakly correlated with baseline SDI scores (r = 0.26, P < 0.0001). Higher baseline SLICC- FI values (per 0.05 increment) were associated with increased mortality risk (hazard ratio 1.59, 95% confidence interval 1.35–1.87), after adjusting for age, sex, steroid use, ethnicity/region, and baseline SDI scores.

Conclusion. The SLICC- FI demonstrates internal validity as a health measure in SLE and might be used to predict future mortality risk. The SLICC- FI is potentially valuable for quantifying vulnerability among patients with SLE, and adds to existing prognostic scores.

The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.

The Hopkins Lupus Cohort is supported by the NIH (grants AR-43727 and AR-69572).  The Montreal General Hospital Lupus Clinic is supported in part by the Singer Family Fund for Lupus Research. Dr. Bae’s work was supported in part by the Ministry of Science & ICT of the Republic of Korea (grant NRF-2017M3A9B4050335). Dr. Gordon’s work was supported by Lupus UK, the Sandwell and West Birmingham Hospitals NHS Trust, and the NIHR/Wellcome Trust Birmingham Clinical Research Facility. Dr. Fortin’s work was supported in part by the Arthritis Society (Distinguished Senior Investigator Award). Dr. Bruce’s work was supported by the NIHR (Senior Investigator Award), Arthritis Research UK, the NIHR Manchester Biomedical Centre, and the NIHR/Wellcome Trust Manchester Clinical Research Facility. Drs. Isenberg and Rahman’s work was supported by the NIHR and University College London Hospitals Biomedical Research Center. Dr. Dooley’s work was supported by the NIH (grant RR00046). Dr. Ramsey-Goldman’s

work was supported by the NIH (grants 5UL-1TR-001422-02 [formerly UL-1TR-000150], UL-1RR-025741, K24-AR-02318, and P60-AR-064464 [formerly P60-AR-48098]). Dr. Ruiz-Irastorza’s work was supported by the Department of Education, Universities, and Research of the Basque Government. Dr. Jacobsen’s work was supported by the Danish Rheumatism Association (grant A3865) and the Novo Nordisk Foundation (grant A05990). Dr. Hanly’s work was supported by the Canadian Institutes of Health Research (grant MOP-88526).

1Alexandra Legge, MD, Susan Kirkland, PhD, Kenneth Rockwood, MD, Pantelis Andreou, PhD: Dalhousie University, Halifax, Nova Scotia, Canada; 2Sang-Cheol Bae, MD, PhD: Hanyang University Hospital for Rheumatic Diseases, Seoul, Republic of Korea; 3Caroline Gordon, MD: University of Birmingham, Birmingham, UK; 4Juanita Romero-Diaz, MD, MSc, Jorge Sanchez-Guerrero, MD, MSc: Instituto Nacional de Ciencias Medicas y Nutrición, Mexico City, Mexico; 5Daniel J. Wallace, MD: Cedars-Sinai Medical Center and David Geffen School of Medicine at University of California, Los

LEGGE ET AL 1298       |

INTRODUCTION

The clinical course of systemic lupus erythematosus (SLE) is highly variable (1) and difficult to predict. Evaluation of SLE patients encompasses 3 core dimensions (1)—disease activity, organ damage, and health- related quality of life (HRQoL)—with each domain providing valuable prognostic information. In par-ticular, the Systemic Lupus International Collaborating Clinics (SLICC)/American College of Rheumatology (ACR) Damage Index (SDI) (2) has consistently predicted adverse outcomes, including future organ damage (3–5) and mortality (3–7), in patients with SLE. However, the associations between disease activity, organ damage, and HRQoL in SLE are complex (3,8,9) and the optimal approach for aggregating data across these domains is unclear. A comprehensive instrument is required to more accurately predict the risk of adverse health outcomes among SLE patients.

In geriatric medicine (10), and increasingly in other disci-plines (11–13), differences in susceptibility to adverse outcomes are quantified using the construct of frailty, which represents a state of increased vulnerability that results in diminished ability to respond to physiologic stressors (14). One approach to oper-ationalizing frailty is the construction of a frailty index (FI) (15), which conceptualizes frailty as a loss of physiologic reserve due to the accumulation of health deficits across multiple systems (16). Individuals with few deficits are considered relatively fit, while those with a greater number of health problems are con-sidered increasingly frail and, thus, more vulnerable to adverse outcomes (17). Prior work in non- lupus populations has identi-fied properties of the FI that remain remarkably consistent across settings (15,18–21), thereby demonstrating the robustness and

generalizability of this approach. Patients with higher FI values have an increased risk of adverse outcomes, including mortality (18–20,22,23). Although utilized in many different clinical con-texts (18,19,24,25), this approach has not been applied previ-ously in SLE.

We hypothesized that evaluating frailty through deficit accu-mulation could help explain the heterogeneous health outcomes in patients with SLE. Using data from the SLICC inception cohort, we constructed an FI for SLE patients, known as the SLICC- FI. Our primary aim was to evaluate the properties of the SLICC- FI, including its ability to predict mortality within the SLICC inception cohort. Secondarily, we assessed whether the SLICC- FI pro-vides additional prognostic information compared to existing SLE measures. To this end, we compared the abilities of the SLICC-FI and the SDI for the prediction of mortality risk.

PATIENTS AND METHODS

Data source. This study was a secondary analysis of longi-tudinal data from the SLICC inception cohort. SLICC comprises 52 investigators at 43 academic centers in 16 countries. From 1999 to 2011, 1,826 SLE patients were recruited from 31 SLICC sites in Europe, Asia, and North America. Patients were enrolled within 15 months of SLE diagnosis, based on the presence of ≥4 revised ACR classification criteria for SLE (26). Data were col-lected in accordance with a standardized protocol and submitted to the coordinating centers at the University of Toronto (Toronto, Ontario, Canada) and Dalhousie University (Halifax, Nova Scotia, Canada). The study was approved by the Institutional Research Ethics Boards of the participating centers, and patients provided written informed consent.

Angeles; 6Sasha Bernatsky, MD, PhD: McGill University, Montreal, Quebec, Canada; 7Ann E. Clarke, MD, MSc: University of Calgary, Calgary, Alberta, Canada; 8Joan T. Merrill, MD: Oklahoma Medical Research Foundation, Oklahoma City; 9Ellen M. Ginzler, MD, MPH: SUNY Downstate Medical Center, Brooklyn, New York; 10Paul Fortin, MD, MPH: CHU de Québec et Université Laval, Quebec City, Quebec, Canada; 11Dafna D. Gladman, MD, Murray B. Urowitz, MD: Toronto Western Hospital and University of Toronto, Toronto, Ontario, Canada; 12Ian  N. Bruce, MD: University of Manchester, NIHR Manchester Musculoskeletal Biomedical Research Centre, Manchester University NHS Foundation Trust, Manchester, UK; 13David A. Isenberg, MD, Anisur Rahman, MD, PhD: University College London, London, UK; 14Graciela S. Alarcón, MD, MPH: University of Alabama at Birmingham; 15Michelle Petri, MD: Johns Hopkins University School of Medicine, Baltimore, Maryland; 16Munther  A.  Khamashta, MD: St. Thomas’ Hospital, King’s College London School of Medicine, London, UK; 17M. A. Dooley, MD, MPH: University of North Carolina, Chapel Hill; 18Rosalind Ramsey-Goldman, MD, DrPH: Northwestern University Feinberg School of Medicine, Chicago, Illinois; 19Susan  Manzi, MD, MPH: Allegheny Health Network, Pittsburgh, Pennsylvania; 20Kristjan  Steinsson, MD: Landspitali University Hospital, Reykjavik, Iceland; 21Asad A. Zoma, MD: Hairmyres Hospital, East Kilbride, Scotland, UK; 22Cynthia Aranow, MD, Meggan Mackay, MD: Feinstein Institute for Medical Research, Manhasset, New York; 23Guillermo Ruiz-Irastorza, MD: Hospital Universitario Cruces, University of the Basque Country, Barakaldo, Spain; 24S. Sam Lim, MD, MPH: Emory University School of Medicine, Atlanta, Georgia; 25Murat Inanc, MD: Istanbul University, Istanbul, Turkey; 26Ronald  F.  van Vollenhoven, MD: Karolinska Institute, Stockholm, Sweden; 27Andreas Jonsen, MD, PhD, Ola Nived, MD, PhD: Lund University, Lund, Sweden; 28Manuel Ramos-Casals, MD: Hospital Clínic de Barcelona,

Barcelona, Spain; 29Diane L. Kamen, MD: Medical University of South Carolina, Charleston; 30Kenneth C. Kalunian, MD: University of California San Diego School of Medicine, La Jolla; 31Soren Jacobsen, MD, DMSc: Copenhagen University Hospital, Copenhagen, Denmark; 32Christine A. Peschken, MD: University of Manitoba, Winnipeg, Manitoba, Canada; 33Anca Askanase, MD, MPH: Hospital for Joint Diseases, New York University, New York, New York; 34John G. Hanly, MD: Queen Elizabeth II Health Sciences Center and Dalhousie University, Halifax, Nova Scotia, Canada.

Dr. Rockwood has received consulting fees, speaking fees, and/or honoraria from Lundbeck (less than $10,000), and is President and Chief Science Officer of DGI Clinical, which has had contracts with Baxter, Baxalta, Shire, Hollister, Nutricia, Roche, and Otsuka. Dr. Wallace has received consulting fees, speaking fees, and/or honoraria from Merck, EMD Serono, Pfizer, Eli Lilly, and Glenmark (less than $10,000 each). Dr. Clarke has received consulting fees from MedImmune/AstraZeneca, Exagen Diagnostics, and Bristol-Myers Squibb (less than $10,000 each). Dr. Inanc has received consulting fees from MSD, AbbVie, Roche, Novartis, Bristol-Myers Squibb, and Pfizer (less than $10,000 each). Dr. van Vollenhoven has received consulting fees, speaking fees, and/or honoraria from AbbVie, Bristol-Myers Squibb, GlaxoSmithKline, Pfizer, UCB, AstraZeneca, Biotest, Biogen, Celgene, Gilead, Janssen, Eli Lilly, Novartis, and Pfizer (less than $10,000 each) and research support from AbbVie, Bristol-Myers Squibb, GlaxoSmithKline, Pfizer, and UCB. No other disclosures relevant to this article were reported.

Address correspondence to John G. Hanly, MD, Division of Rheumatology, Nova Scotia Rehabilitation Center, First Floor, 1341 Summer Street, Halifax, Nova Scotia B3H 4K4, Canada. E-mail: john.hanly@nshealth.ca.

Submitted for publication September 27, 2018; accepted in revised form February 12, 2019.

SLICC-­FI­IN­PATIENTS­WITH­SLE­ |      1299

Clinical and laboratory assessments. Assessments were performed at enrollment and annually thereafter. Demo-graphic features included age, sex, race/ethnicity, geographic location, and postsecondary education. Use of glucocorticoids, antimalarials, and immunosuppressive agents was noted. We documented disease features meeting the ACR classification criteria for SLE (26), neuropsychiatric events (27), and medical comorbidities in each patient at the enrollment visit, and between follow- up visits. SLE disease activity (SLE Disease Activity Index 2000 [SLEDAI- 2K] scores [28]), cumulative organ damage (SDI scores [2]), and HRQoL (Medical Outcomes Study Short- Form 36 [SF- 36] health survey scores [29]) were documented at each visit. Blood pressure (in mm Hg), height (in meters), and weight (in kilo-grams) were also recorded. Laboratory investigations to assess SLE disease activity and organ damage were performed locally at each visit (3).

Construction of the SLICC- FI. A standard procedure for construction of an FI, as described in detail elsewhere (15), was used to generate the SLICC- FI. Briefly, we established a baseline data set, consisting of the first visit for each patient at which both the SDI and the SF- 36 were completed. Varia-bles were selected for the SLICC- FI if they met the criteria for a health deficit, defined as any symptom, disease process, func-tional impairment, or laboratory abnormality that is 1) acquired, 2) associated with chro nologic age, 3) associated with adverse health outcomes, 4) present in ≥1% and ≤80% of the sample, and 5) missing values for <5% of the sample (15). Of 222 can-didate variables, 48 items met the inclusion criteria. SLICC- FI health deficits spanned a range of organ systems and included variables related to organ damage, disease activity, comorbid-ities, and functional status. Each health deficit was assigned a score from 0 (completely absent) to 1 (fully present) using estab-lished cutoff points (2,26–29). More detailed information regard-ing the SLICC- FI health deficits and their scoring can be found in Supplementary Table 1 (available on the Arthritis & Rheu-matology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40859/ abstract).

Calculation of SLICC- FI scores. The SLICC- FI score is the sum of an individual’s health deficit scores divided by the total number of deficits. For example, if 12 of 48 deficits are fully pres-ent, the SLICC- FI score is calculated as 12/48 = 0.25. Changes in FI values of ≥0.03 are clinically relevant. SLICC- FI scores were calculated for each patient using data at baseline and last visit.

Evaluation of the properties of the SLICC- FI. We con-sidered content, construct, and criterion validity (30). Content validity was inherent in the derivation of health deficits from exist-ing, well- validated SLE instruments (2,26–29). The use of a stan-dard procedure (15) for the identification of health deficits further enhanced face validity.

For construct validity, we compared the properties of the SLICC- FI to existing FI measures in non- SLE populations. We esti-mated the relationship between patient age and SLICC- FI values (FI values typically increase by 1% per year on a log scale [19,20]), the distribution of SLICC- FI values (typically, a Gaussian distribu-tion [19]), and its 99th percentile value (typically, a value of <0.7 [15,18–20]). We also estimated the correlation between baseline SLICC- FI scores and baseline scores on the SLEDAI- 2K, SDI, and SF- 36. For criterion validity, we estimated the predictive validity of baseline SLICC- FI scores for predicting the risk of mortality.

Statistical analysis. Descriptive statistics were calculated for demographic and clinical characteristics, and for SLICC- FI val-ues at baseline and last visit. The distributions of SLICC- FI scores were compared to theoretical distributions using goodness- of- fit tests. The association between age and baseline SLICC- FI scores was estimated using Pearson’s correlation coefficients and simple linear regression. Spearman’s rank correlation coefficients were used to estimate the association of SLICC- FI values with SDI, SLEDAI- 2K, and SF- 36 scores at baseline. In addition, Pearson’s correlation coefficients were used to estimate the association of SLICC- FI scores at baseline with SLICC- FI scores at last fol-low- up. All linear regression models met the required assumptions of linearity, homoscedasticity, and normality of errors.

Kaplan- Meier survival curves illustrated the risk of mortality following the baseline visit. The event date was the date of death, with survivors censored at their last visit. To evaluate the associa-tion between baseline frailty and mortality risk, an FI cutoff point that was previously validated in non- SLE populations (22,31,32) was used to dichotomize patients into those who were frail (SLICC- FI >0.21) and those who were not frail (SLICC- FI ≤0.21) at baseline. We then compared mortality risk between these 2 groups using a log- rank test.

The predictive validity of baseline SLICC- FI scores for mor-tality risk was further evaluated using Cox proportional hazards regression. First, a univariable model was constructed with the baseline SLICC- FI (per 0.05 increase in score) as the indepen-dent variable. Demographic and clinical variables were identified as potential confounders, and univariable models for mortality risk were constructed for each of these variables. The full multivaria-ble model included baseline SLICC- FI scores and potential con-founders associated with mortality at a P value of less than 0.1 in univariable analysis. A backwards stepwise procedure was used to remove potential confounders that were no longer statistically significant in multivariable analysis. The final multivariable model included the baseline SLICC- FI and any potential confounders for which removal from the model would result in a statistically signif-icant change (P < 0.05), as determined by the likelihood ratio (LR) test. Age and sex were retained in the final model regardless of statistical significance.

A similar procedure was followed to construct unadjusted and adjusted models for mortality risk, using 1) baseline SDI scores

LEGGE ET AL 1300       |

as the independent variable of interest, and 2) both baseline SLICC- FI and baseline SDI scores as independent variables in the same model. We then used LR tests to compare the goodness- of- fit of the models containing both baseline SLICC- FI and base-line SDI scores to the goodness- of- fit of the models containing 1) the baseline SLICC- FI alone, and 2) the baseline SDI alone. We also compared the relative performance of these alternative models using Akaike’s information criterion (AIC), with smaller AIC values indicating better predictive quality. For all models, the proportional hazards assumption was tested using log–log plots, time- varying covariates, and Schoenfeld residuals. Data analysis was conducted using STATA- IC version 14 (StataCorp).

Sensitivity analyses. The SLICC- FI contains several health deficits related to organ damage that could overlap with items captured by the SDI. To assess for a relationship between base-line SLICC- FI scores and mortality risk independent of organ dam-age, we repeated the above analyses omitting all damage- related items from the SLICC- FI and recalculating SLICC- FI scores using the remaining 33 health deficits.

As many SLE patients have SDI scores of 0, particularly early in the disease (3), we investigated whether the SLICC- FI could predict mortality risk in the subgroup of patients with no organ damage (SDI of 0) at baseline. Finally, to evaluate the influ-ence of disease duration, we repeated these analyses in patients whose baseline visits occurred within 2 years of SLE diagnosis.

RESULTS

Characteristics of the study patients at baseline. There were 1,683 patients with SLE (92.2% of the cohort) with ≥1 visit at which both the SDI and SF- 36 scores were recorded. Each patient’s first such visit was included in the baseline data set. This occurred within 2 years of SLE diagnosis for 1,390 patients (82.6% of the cohort). The mean ± SD disease duration at base-line was 18.8 ± 15.7 months. The demographic and clinical char-acteristics of the patients are shown in Table 1. At baseline, 70.1% of the patients were receiving glucocorticoids, 68.3% were receiv-

ing antimalarials, and 40.5% were receiving immunosuppressives.

Properties of the SLICC- FI scores at baseline. SLICC- FI scores were calculated for 1,682 of 1,683 patients in the baseline data set. A baseline SLICC- FI score could not be calculated for 1 patient because the patient’s data were missing for >20% of the health deficits (33). Among the 1,682 patients, the baseline SLICC- FI scores ranged from 0 to 0.51, with a median score of 0.16 (interquartile range [IQR] 0.11–0.22) and a slightly higher mean score of 0.17 (±SD 0.08).

The distribution of baseline SLICC- FI scores closely approx-imated a beta distribution, with shape parameters of α = 3.51 and β = 17.49 (Figure 1). This was confirmed using goodness- of- fit tests. There was a positive, linear relationship between

patient age and baseline SLICC- FI values (Pearson’s correlation coefficient r = 0.20, P < 0.0001), although age accounted for only 4% of the total variation in baseline SLICC- FI scores. The submaximal limit (99th percentile value) of baseline SLICC- FI scores was 0.39. A significant relationship between age and SLICC- FI scores was absent in this 99th percentile sample.

Baseline SLICC- FI scores were not significantly different (P = 0.12 by t- test) between male patients (mean ± SD score 0.16 ± 0.08) and female patients (mean ± SD score 0.17 ± 0.08). How-ever, male patients were significantly older (mean ± SD age 40.0 ± 16.4 years) than female patients (mean ± SD age 35.1 ± 12.8

Table  1. Demographic and clinical characteristics of the SLE patients in the SLICC inception cohort at the time of their baseline visit and last follow- up visit*

Baseline visit (n = 1,683)

Last follow- up visit (n = 1,507)

Age, mean ± SD years

35.7 ± 13.4 42.8 ± 13.6

SexFemale 1,493 (88.7) 1,337 (88.7)Male 190 (11.3) 170 (11.3)

Race/ethnicityCaucasian 834 (49.6) 742 (49.2)African ancestry 280 (16.6) 246 (16.3)Asian 260 (15.5) 245 (16.3)Hispanic 248 (14.7) 222 (14.7)Other 61 (3.6) 52 (3.5)

Geographic location

US 467 (27.7) 377 (25.0)Canada 395 (23.5) 376 (25.0)Mexico 197 (11.7) 181 (12.0)Europe 461 (27.4) 419 (27.8)Asia 163 (9.7) 154 (10.2)

EducationPostsecondary 847 (50.3) 767 (50.9)Missing data 22 (1.3) 20 (1.3)

SLE duration, median (IQR) years

1.2 (0.9–1.5) 8.5 (5.6–11.3)

SLEDAI- 2K score, median (IQR)

2 (0–6) 2 (0–4)

SDI score of 0 1,270 (75.5) 721 (47.8)SF- 36 score,

median (IQR) PCS 41.7 (32.5–50.8) 43.7 (32.6–52.6)MCS 48.8 (37.4–55.8) 49.8 (39.3–56.1)

* Except where indicated otherwise, values are the number (%) of patients with systemic lupus erythematosus (SLE). SLICC = Systemic Lupus International Collaborating Clinics; IQR = interquartile range; SLEDAI- 2K = SLE Disease Activity Index 2000; SDI = SLICC/American College of Rheumatology Damage Index; SF- 36 = Short- Form 36; PCS = physical component summary; MCS = mental component summary.

SLICC-­FI­IN­PATIENTS­WITH­SLE­ |      1301

years) at baseline (P < 0.0001 by t- test). After adjusting for age, male sex was associated with lower baseline SLICC- FI scores (β = −0.02, P = 0.01).

At baseline, higher SLICC- FI values were associated with higher SDI scores (Spearman’s rank correlation coefficient rs = 0.26, P < 0.0001) and higher SLEDAI- 2K scores (rs = 0.23, P < 0.0001). These associations were weak, despite the presence of overlapping SDI and SLEDAI- 2K variables that were also captured as health deficits in the SLICC- FI. These correlations remained statistically significant after removing overlapping items from the SLICC- FI (rs = 0.15, P < 0.0001 for the SDI; rs = 0.11, P < 0.0001 for the SLEDAI- 2K).

At baseline, there was a moderately strong, negative asso-ciation between SLICC- FI values and SF- 36 physical component summary (PCS) scores (rs = −0.62, P < 0.0001). Higher SLICC- FI scores were also associated with lower SF- 36 mental component summary (MCS) scores (rs = −0.33, P < 0.0001) at baseline. These negative correlations remained statistically significant after remov-ing SF- 36 variables from the SLICC- FI (rs = −0.35, P < 0.0001 for the PCS; rs = −0.12, P < 0.0001 for the MCS).

Properties of the SLICC- FI scores at last follow- up visit. There were 1,507 patients with final study visits after a mean ± SD follow- up time of 7.2 ± 3.7 years from baseline. Demo-graphic characteristics, including race/ethnicity, postsecondary education status, and sex distribution, were similar to those at baseline (Table 1). Compared to baseline, patients at the last fol-low- up visit had less active disease (mean ± SD SLEDAI- 2K score 2.82 ± 3.37 versus 3.98 ± 4.28 at baseline) and more organ dam-age (mean ± SD SDI score 1.19 ± 1.61 versus 0.40 ± 0.84 at baseline), although 721 patients (47.8%) still had no organ dam-age (SDI of 0) at their last visit.

SLICC- FI values at last follow- up ranged from 0.004 to 0.49, with a mean ± SD SLICC- FI score of 0.15 ± 0.08 and a median of 0.14 (IQR 0.09–0.21). Compared to baseline SLICC- FI values, the properties of the final SLICC- FI scores were similar (Figure 1), including their distribution, 99th percentile value (score 0.38), and weakly positive, linear relationship with age (r = 0.26, P < 0.0001). Similar to baseline, SLICC- FI values at last follow- up were posi-tively associated with SDI scores (rs = 0.44, P < 0.0001) and with SLEDAI- 2K scores (rs = 0.26, P < 0.0001) from the same visit. SLICC- FI values also remained negatively associated with SF- 36 PCS scores (rs = −0.68, P < 0.0001) and MCS scores (rs = −0.36, P < 0.0001) at last follow- up. Final SLICC- FI values were mod-erately correlated with baseline SLICC- FI scores (r = 0.57, P < 0.0001).

During follow- up (n = 1,506), 67.8% of patients had a clinically meaningful change (±0.03) in their SLICC- FI scores (Figure 2). Over time, 395 patients (26.2%) had a clinically mean-ingful increase in SLICC- FI scores, while 626 patients (41.6%) had a clinically meaningful decrease in SLICC- FI scores. Longer fol-low- up was weakly associated with more positive changes (i.e., increases) in SLICC- FI scores (r = 0.10, P = 0.0001).

Association of the baseline SLICC- FI score with mor-tality risk. There were 66 deaths after a mean ± SD follow- up of 5.4 ± 3.7 years (Figure  3A). As 117 patients had no avail-able follow- up data after their baseline visit, 1,566 patients were included in the survival analysis for mortality rates, with a mean ± SD follow- up time among censored individuals of 6.7 ± 4.0 years.

At baseline, 431 (27.5%) of 1,566 patients were considered frail (SLICC- FI score >0.21) and frailty was associated with a signif-icant increase in the risk of mortality (P < 0.0001 by log- rank test) (Figure 3B). Specifically, mortality risk was >4 times higher among frail individuals (SLICC- FI score >0.21) when compared to patients classified as nonfrail (SLICC- FI score ≤0.21) at baseline (hazard ratio [HR] 4.37, 95% confidence interval [95% CI] 2.67–7.17).

In unadjusted Cox regression analysis, higher baseline SLICC- FI values (per 0.05 increment) were associated with an increased risk of mortality (HR 1.62, 95% CI 1.41–1.85). Baseline SDI scores (per 1- unit increase) demonstrated a similar associa-tion with mortality risk in unadjusted analysis (HR 1.65, 95% CI 1.38–1.97). In evaluating potential confounders, we found that

Figure 1. Observed distribution of the Systemic Lupus International Collaborating Clinics Frailty Index (SLICC- FI) scores at baseline (n = 1,682) and at last follow- up visit (n = 1,507) among systemic lupus erythematosus patients in the SLICC inception cohort.

LEGGE ET AL 1302       |

older age, male sex, steroid use, immunosuppressive use, and higher disease activity (higher SLEDAI- 2K scores) at baseline were associated with an increased risk of mortality (Table 2). Antima-larial use and postsecondary education were associated with a lower mortality risk. There were also differences in mortality risk based on race/ethnicity and geographic location (Table 2). How-ever, the effects of race/ethnicity and geographic location were not independent of one another. Therefore, for the purposes of multivariable analysis, a combined ethnicity/region variable was

created.In multivariable analysis, higher baseline SLICC- FI values

remained significantly associated with an increased risk of mortal-ity after accounting for potentially confounding variables (Table 3, model 1). Similarly, there was a persistent association between higher baseline SDI scores and increased mortality risk following multivariable adjustment (Table 3, model 2). In comparing these models using AIC values, the multivariable models containing the SLICC- FI (model 1) demonstrated relatively smaller AIC values

than the models containing the SDI (model 2) (Table 3).When baseline SLICC- FI and SDI scores were included in

the same models for prediction of mortality risk, both measures maintained independent associations with the risk of death dur-ing follow- up (Table 3, model 3). Compared to the models con-taining either the baseline SLICC- FI or the baseline SDI alone, the models containing both baseline SLICC- FI and baseline SDI scores demonstrated superiority for predicting mortality risk (Table 3). In particular, the addition of the baseline SLICC- FI to the model containing the baseline SDI alone was associated with significant improvement in model fit (model 2 versus model 3, LR test statistic 30.07 [P < 0.0001] for the final model) as well as significant improvement in the relative predictive quality (model 2 AIC = 796.3 versus model 3 AIC = 768.3 for the final model).

Results of sensitivity analyses. In a subgroup analy-sis including only patients without organ damage (SDI of 0) at baseline (n = 1,187), frailty was still associated with increased mortality risk. In our final multivariable model for this subgroup, an increase in the baseline SLICC- FI score by 0.05 was associ-ated with an increase in mortality risk by ~50% (HR 1.47, 95% CI 1.18–1.83), after adjusting for age, sex, steroid use, and ethnicity/region (see Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40859/ abstract). Similar results were also obtained when we repeated the survival analyses for mortality after removing all health deficits that related to organ damage from the SLICC- FI (see Supplementary Table 3, online at http://onlin elibr ary.wiley.com/doi/10.1002/art.40859/ abstract).

Finally, we repeated the analyses of mortality risk in the sub-group of patients whose baseline visits occurred within 2 years of SLE diagnosis (n = 1,390). In this subset, we found a similar rela-

Figure 2. Distribution of change in the SLICC- FI values from the baseline assessment to the last follow- up visit among systemic lupus erythematosus (SLE) patients in the SLICC inception cohort (n = 1,506). See Figure 1 for other definitions.

Figure  3. Kaplan- Meier survival curves for the risk of mortality during follow- up among systemic lupus erythematosus (SLE) patients in the SLICC inception cohort, overall (A) and stratified by baseline frailty status (B). See Figure 1 for other definitions.

SLICC-­FI­IN­PATIENTS­WITH­SLE­ |      1303

tionship between baseline SLICC- FI values and mortality risk (see Supplementary Table 4, online at http://onlinelibrary.wiley.com/doi/10.1002/art.40859/abstract).

DISCUSSION

In a well- characterized, international cohort of patients with recently diagnosed SLE, we have demonstrated the feasibility of using an FI to quantify vulnerability to adverse outcomes in SLE. The SLICC- FI was correlated with existing measures of SLE disease activity, organ damage, and HRQoL. Higher SLICC- FI values were associated with increased risk of mortality, indepen-dent of other demographic and clinical factors known to predict mortality in SLE.

The SLICC- FI exhibited measurement properties similar to those consistently demonstrated by other frailty indices in non- lupus populations (15,18–21). For example, at all ages, women demonstrated higher mean SLICC- FI scores compared to men (15,20). We found a positive, linear association between chrono-logic age and SLICC- FI values that was very weak. As expected, the relationship between age and SLICC- FI values attenuated to 0 at the highest levels of frailty, as severely frail individuals die rather than accumulate further deficits (34). Finally, although the mean SLICC- FI value (mean 0.17) was high compared to estimates for similarly aged individuals in the general population (22,35,36), the upper limit of SLICC- FI scores was not higher than expected (maximum score of 0.51), suggesting that the SLICC- FI was not overestimating the prevalence of frailty among SLE patients.

After a mean follow- up interval of 7.2 years, the mean SLICC- FI scores, as well as the overall distribution of SLICC- FI scores, remained largely unchanged compared to baseline. This is uncommon in FI studies with such a prolonged follow- up, and may reflect the impact of treatment, as demonstrated by the large number of patients in whom SLICC- FI scores improved during follow- up. This finding may help to provide insight into the relationship between frailty and chronologic age in disease- specific cohorts.

The lack of change in mean SLICC- FI scores during fol-low- up could also reflect a tradeoff between deficits related to SLE disease activity and those related to organ damage. Early in disease, frailty may be driven by disease activity, with mini-mal organ damage. With treatment, disease activity recedes and damage accumulates consequent to the disease, its treatment, and other comorbidities (4,37,38). With longer follow- up, we expect that mean SLICC- FI scores will increase, as deficits con-tinue to accumulate with increasing age (36) and increasing dis-ease duration. This hypothesis is supported by the finding that a longer follow- up time was weakly associated with worsening SLICC- FI scores over time.

While the overall distribution of SLICC- FI values remained largely unchanged between baseline and last follow- up, approxi-mately two- thirds of the patient population had clinically meaning-ful changes in their SLICC- FI scores between the 2 time points. The potential for SLICC- FI scores to decrease, in contrast to SDI scores, supports the view that frailty itself can be reversed (10). To

Table 2. Univariable Cox regression models for the association of baseline demographic and clinical variables with mortality risk during follow- up among SLE patients in the SLICC inception cohort (n = 1,566)*

Independent variable

P, log- rank test†

Hazard ratio

(95% CI)

P, mortality

risk

Baseline age (in years)‡

1.055 (1.040–1.072) <0.0001

Sex Female 0.061 ReferentMale 1.80 (0.96–3.37) 0.065

Race/ethnicityCaucasian 0.025 ReferentHispanic 1.54 (0.86–2.77) 0.146African ancestry 1.11 (0.56–2.19) 0.765Asian 0.25 (0.08–0.82) 0.023Other 0.41 (0.06–3.03) 0.386

Geographic locationUS 0.052 ReferentCanada 1.07 (0.52–2.21) 0.860Mexico 1.71 (0.81–3.64) 0.162Europe 0.86 (0.41–1.81) 0.692Asia 0.26 (0.06–1.18) 0.080

Postsecondary education§

No 0.009 ReferentYes 0.46 (0.27–0.77) 0.003

Glucocorticoid useNo 0.002 ReferentYes 3.12 (1.49–6.55) 0.003

Immunosuppressive use

No 0.002 ReferentYes 2.19 (1.33–3.59) 0.002

Antimalarial useNo 0.007 ReferentYes 0.52 (0.32–0.84) 0.008

SLEDAI- 2K (per 1.0 increment)

1.05 (1.00–1.09) 0.039

SLE disease duration (in years)

1.00 (0.98–1.02) 0.649

* SLE = systemic lupus erythematosus; SLICC = Systemic Lupus In-ternational Collaborating Clinics; 95% CI = 95% confidence interval; SLEDAI- 2K = SLE Disease Activity Index 2000. † For categorical variables only. ‡ Time- varying covariate (proportional hazards assumption not met). § A “missing” indicator was included for the 1.3% of patients for whom these data were lacking.

LEGGE ET AL 1304       |

this end, the SLICC- FI warrants investigation as a possible out-come measure for future intervention studies.

Similar to the findings of FI studies in non- lupus popula-tions (18,19,23), we identified a significant association between baseline SLICC- FI scores and mortality risk. Given that prior work has emphasized the importance of the SDI for predicting mortal-ity in SLE (3,6,7), some may question whether the ability of the SLICC- FI to predict mortality is heavily reliant on the inclusion of deficits related to organ damage. However, the results of our sensitivity analyses demonstrated a persistence of the relation-ship between baseline SLICC- FI values and mortality risk, despite removal of all damage- related deficits from the index. This finding highlights a key strength of the deficit accumulation approach to

frailty: it is the cumulative impact of multiple small effects, rather than specific individual deficits, that is important (17,39). As long as a sufficient number of variables are included in an FI (generally, more than 30 items), its predictive ability for adverse outcomes remains robust, even when a subset of the included deficits are removed (15,20,21,33,40).

Traditionally, the core dimensions of SLE—disease activ-ity, organ damage, and HRQoL—have been evaluated sepa-rately (1). However, this approach does not capture interactions between these domains, thereby potentially missing their impact on prognosis. Conversely, the SLICC- FI combines aspects of all 3 dimensions into a single measure. The relationships that exist between deficits from different domains within the SLICC- FI are

Table 3. Multivariable Cox regression models for the association of baseline SLICC- FI and SDI scores with mortality risk during follow- up among SLE patients in the SLICC inception cohort*

Full multivariable model

(n = 1,556)†

Final multivariable model

(n = 1,565)‡

Mortality riskModel 1: SLICC- FI alone (per 0.05 increment)

HR (95% CI) 1.62 (1.36–1.92) 1.66 (1.42–1.94)P <0.001 <0.001

Model 2: SDI alone (per 1.0 increment)HR (95% CI) 1.45 (1.18–1.78) 1.50 (1.23–1.83)P <0.001 <0.001

Model 3: SLICC- FI and SDI SLICC- FI (per 0.05 increment)

HR (95% CI) 1.55 (1.30–1.84) 1.59 (1.35–1.87)P <0.001 <0.001

SDI (per 1.0 increment)HR (95% CI) 1.27 (1.03 – 1.57) 1.27 (1.03 – 1.57)P 0.025 0.023

Overall model comparisons by LR testModel 1 vs. model 3

LR test statistic 4.34 4.71P 0.037 0.030

Model 2 vs. model 3LR test statistic 27.79 30.07P <0.001 <0.001

AIC values for model predictive qualityModel 1: SLICC- FI 770.2 771.0Model 2: SDI 789.1 796.3Model 3: SLICC- FI and SDI 767.7 768.3

* SLICC- FI = Systemic Lupus International Collaborating Clinics frailty index; SDI = SLICC/American Col-lege of Rheumatology Damage Index; SLE = systemic lupus erythematosus; HR = hazard ratio; 95% CI = 95% confidence interval; LR = likelihood ratio; AIC = Akaike’s information criterion. † Models were adjusted for the following baseline characteristics: age, sex, steroid use, antimalarial use, immunosuppressive use, ethnicity/location, postsecondary education, and SLE Disease Activity Index 2000. ‡ Models were adjusted for the following baseline characteristics: age, sex, steroid use, and ethnicity/location.

SLICC-­FI­IN­PATIENTS­WITH­SLE­ |      1305

critical to its performance as a prognostic tool. For example, the scoring of the “Cerebrovascular Disease” health deficit weighs transient ischemic attacks and debilitating strokes equally, despite clear differences in the likely impact of these events on prognosis. However, an individual with a disabling stroke is likely to have additional deficits related to their functional performance that will be reflected in their SLICC- FI score. As shown in this example, including deficits from different domains ensures that the overall impact of complex health events is accurately repre-sented in the SLICC- FI.

The baseline SLICC- FI and SDI were both significant predic-tors of mortality risk. Despite some overlap in the items captured, these 2 instruments are likely measuring separate constructs, and each provides valuable prognostic information. The SDI can be viewed as a measure of SLE disease severity in 1 of 3 core dimensions (1). In contrast, the SLICC- FI provides a more holistic approach, incorporating both patient and healthcare provider per-spectives with regard to the impact of the disease, its treatment, and other comorbidities on the health of the patient. Prior FI stud-ies in other disease- specific cohorts have yielded similar findings, namely that both the FI and existing measures of disease severity maintain independent associations with the risk of future adverse health outcomes (18,19).

Another important distinction between the SDI and the SLICC- FI is that the SDI does not capture damage accrued prior to SLE diagnosis, and therefore does not consider the likely effects of preexisting organ damage on mortality risk. Conversely, health deficits accrued prior to SLE diagnosis can be included in the SLICC- FI. Damage captured by the SDI often does not occur until several years after the diagnosis of SLE (3,6,7). Thus, the added prognostic value of the SLICC- FI may be highest early in the dis-ease course. This was demonstrated in our subgroup analysis of patients without baseline organ damage (SDI of 0), in whom each 0.05 increase in the baseline SLICC- FI score was associated with a 50% increase in mortality risk.

An alternative approach to the measurement of frailty uses rules- based tools (17) such as the Fried frailty pheno-type (41), which classifies individuals as frail if any 3 of 5 spe-cific criteria (slow gait speed, impaired grip strength, reduced physical activity, weight loss, and exhaustion) are met. The Fried frailty phenotype was recently evaluated in a cohort of 152 women with SLE (42). Similar to our findings, that study demonstrated an association between frailty and increased mortality risk (42). Phenotypic frailty was also associated with significantly worse physical functioning, mea sured using the physical function subscale of the SF- 36 (42). Interestingly, we observed a similar association between concurrent SLICC- FI and SF- 36 PCS scores. Phenotypic frailty at baseline was also associated with significant declines in physical functioning during follow- up (42). Future work will evaluate the association of baseline SLICC- FI values with changes in functional status and quality of life over time.

Our study has some limitations. First, a relatively low num-ber of deaths occurred during follow- up, which limited the sta-tistical power in our analysis of mortality. Although this would increase our Type II error rate, it would not change the direc-tion of our finding that baseline SLICC- FI values are a significant predictor of mortality risk. The low mortality rate in the SLICC inception cohort reflects improved survival among SLE patients compared to that observed in previous eras (43). As such, future work will evaluate the ability of the SLICC- FI to predict other clin-ically meaningful outcomes.

Second, we only evaluated the change in SLICC- FI values between 2 time points. Future work will focus on better under-standing the trajectories of SLICC- FI values over multiple time points.

Third, we were unable to calculate SLICC- FI values for 144 patients (7.9%) due to missing data. However, the char-acteristics of the included patients were very similar to those reported in previous studies from the SLICC cohort (3), sug-gesting that our data set was representative of the overall cohort. Missing data also precluded the use of SLICC enroll-ment visits as baseline visits for many patients. Despite this, >80% of patients had their baseline visit within 2 years of SLE diagnosis, and our results were unchanged in a subgroup analysis including only these individuals.

Last, it should be acknowledged that we evaluated the SLICC- FI in the same cohort used for its initial construction. This is a cohort of relatively young patients with recently diagnosed SLE. External validation of the SLICC- FI in other SLE cohorts is required to confirm our findings and to investigate their general-izability to older patients with more longstanding SLE.

In conclusion, evaluating frailty through deficit accumula-tion provides a holistic approach to prognostication among SLE patients, incorporating aspects of disease activity, organ dam-age, and HRQoL into a single measure. We have demonstrated the SLICC- FI to be a meaningful health measure in SLE with the ability to vary over time and to predict mortality. Although the practical utility of frailty assessment in routine clinical care of SLE patients remains unexplored, the SLICC- FI holds promise as a clinical and research tool for the identification of vulnerable SLE patients. It may also be a valuable outcome measure for future intervention studies.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final ver-sion to be published. Dr. Hanly had full access to all of the data in the study and takes responsibility for the integrity of the data and the accu-racy of the data analysis.Study conception and design. Legge, Kirkland, Rockwood, Andreou, Romero- Diaz, Merrill, Fortin, Gladman, Urowitz, Bruce, Alarcón, Petri, Khamashta, Zoma, Nived, Askanase, Hanly.Acquisition of data. Legge, Kirkland, Rockwood, Andreou, Bae, Gor-don, Romero- Diaz, Sanchez- Guerrero, Wallace, Bernatsky, Clarke, Merrill, Ginzler, Fortin, Gladman, Urowitz, Bruce, Isenberg, Rahman,

LEGGE ET AL 1306       |

Alarcón, Petri, Khamashta, Dooley, Ramsey- Goldman, Manzi, Steins-son, Zoma, Aranow, Mackay, Ruiz- Irastorza, Lim, Inanc, van Vollen-hoven, Nived, Ramos- Casals, Kamen, Kalunian, Jacobsen, Peschken, Askanase, Hanly.Analysis and interpretation of data. Legge, Kirkland, Rockwood, Andreou, Gordon, Romero- Diaz, Wallace, Bernatsky, Clarke, Fortin, Urowitz, Bruce, Khamashta, Ramsey- Goldman, Zoma, Inanc, van Vol-lenhoven, Jonsen, Askanase, Hanly.

REFERENCES 1. Strand V, Chu AD. Measuring outcomes in systemic lupus

erythematosus clinical trials. Expert Rev Pharmacoecon Outcomes Res 2011;11:455–68.

2. Gladman D, Ginzler E, Goldsmith C, Fortin P, Liang M, Urowitz  M,  et  al. The development and initial validation of the Systemic Lupus International Collaborating Clinics/American College of Rheumatology damage index for systemic lupus erythematosus. Arthritis Rheum 1996;39:363–9.

3. Bruce IN, O’Keeffe AG, Farewell V, Hanly JG, Manzi S, Su L, et al. Factors associated with damage accrual in patients with systemic lupus erythematosus: results from the Systemic Lupus International Collaborating Clinics (SLICC) Inception Cohort. Ann Rheum Dis 2015;74:1706–13.

4. Alarcón GS, Roseman JM, McGwin G, Uribe A, Bastian HM, Fessler  BJ, et al. Systemic lupus erythematosus in three ethnic groups. XX. Damage as a predictor of further damage. Rheumatology (Oxford) 2004;43:202–5.

5. Cardoso CRL, Signorelli FV, Papi JA, Salles GF. Initial and accrued damage as predictors of mortality in Brazilian patients with systemic lupus erythematosus: a cohort study. Lupus 2008;17:1042–8.

6. Rahman P, Gladman D, Urowitz MB, Hallett D, Tam LS. Early damage as measured by the SLICC/ACR damage index is a predictor of mortality in systemic lupus erythematosus. Lupus 2001;10:93–6.

7. Nivad O, Jonsen A, Bengtsson AA, Bengtsson C, Sturfelt G. High predictive value of the Systemic Lupus International Collaborating Clinics/American College of Rheumatology damage index for survival in systemic lupus erythematosus. J Rheumatol 2002;29: 1398–400.

8. Schmeding A, Schneider M. Fatigue, health- related quality of life and other patient- reported outcomes in systemic lupus erythematosus. Best Pract Res Clin Rheumatol 2013;27:363–75.

9. Mok CC, Ho LY, Cheung MY, Yu KL, To CH. Effect of disease activity and damage on quality of life in patients with systemic lupus erythematosus: a 2- year prospective study. Scand J Rheumatol 2009;38:121–7.

10. Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. Frailty in elderly people. Lancet 2013;381:752–62.

11. Muscedere J, Waters B, Varambally A, Bagshaw SM, Boyd JG, Maslove D, et al. The impact of frailty on intensive care unit outcomes: a systematic review and meta- analysis. Intensive Care Med 2017;43:1105–22.

12. Abel GA, Klepin HD. Frailty and the management of hematologic malignancies. Blood 2018;131:515–24.

13. Partridge JS, Harari D, Dhesi JK. Frailty in the older surgical patient: a review. Age Ageing 2012;41:142–7.

14. Fulop T, Larbi A, Witkowski JM, McElhaney J, Loeb M, Mitnitski A, et al. Aging, frailty and age- related diseases. Biogerontology 2010;11:547–63.

15. Searle SD, Mitnitski A, Gahbauer EA, Gill TM, Rockwood K. A standard procedure for creating a frailty index. BMC Geriatr 2008;8:24.

16. Mitnitski A, Rockwood K. Aging as a process of deficit accumulation: its utility and origin. Interdiscipl Top Gerontol 2015;40:85–98.

17. Theou O, Walston J, Rockwood K. Operationalizing frailty using the frailty phenotype and deficit accumulation approaches. Interdiscipl Top Gerontol 2015;41:66–73.

18. Rockwood MR, MacDonald E, Sutton E, Rockwood K, the Canadian Scleroderma Research Group, Baron M. Frailty index to measure health status in people with systemic sclerosis. J Rheumatol 2014;41:698–705.

19. Guaraldi G, Brothers TD, Zona S, Stentarelli C, Carli F, Malagoli A, et al. A frailty index predicts survival and incident multimorbidity independent of markers of HIV disease severity. AIDS 2015;29:1633–41.

20. Mitnitski A, Song X, Skoog I, Broe GA, Cox JL, Grunfeld E, et al. Relative fitness and frailty of elderly men and women in developed countries and their relationship with mortality. J Am Geriatr Soc 2005;53:2184–9.

21. Rockwood K, Mitnitski A. Frailty defined by deficit accumulation and geriatric medicine defined by frailty. Clin Geriatr Med 2011;27:17–26.

22. Rockwood K, Song X, Mitnitski A. Changes in relative fitness and frailty across the adult lifespan: evidence from the Canadian National Population Health Survey. CMAJ 2011;183:E487–94.

23. Kojima G, Iliffe S, Walters K. Frailty index as a predictor of mortality: a systematic review and meta- analysis. Age Ageing 2018;47:193–200.

24. Lai JC, Feng S, Terrault NA, Lizaola B, Hayssen H, Covinsky K. Frailty predicts waitlist mortality in liver transplant candidates. Am J Transplant 2014;14:1870–9.

25. Hubbard RE, Peel NM, Smith M, Dawson B, Lambat Z, Bak M, et al. Feasibility and construct validity of a frailty index for patients with chronic kidney disease. Australas J Ageing 2015;34:E9–12.

26. Hochberg MC. Updating the American College of Rheumatology revised criteria for the classification of systemic lupus erythematosus [letter]. Arthritis Rheum 1997;40:1725.

27. ACR Ad Hoc Committee on Neuropsychiatric Lupus Nomenclature. The American College of Rheumatology nomenclature and case definitions for neuropsychiatric lupus syndromes. Arthritis Rheum 1999;42:599–608.

28. Gladman DD, Ibañez D, Urowitz MB. Systemic Lupus Erythematosus Disease Activity Index 2000. J Rheumatol 2002;29:288–91.

29. Ware JE, Sherbourne CD. The MOS 36- item Short- Form health survey (SF- 36). I. Conceptual framework and item selection. Med Care 1992;30:473–83.

30. Streiner DL, Norman GR. Health measurement scales: a practical guide to their development and use. 4th edition. Oxford: Oxford University Press; 2008.

31. Rockwood K, Song X, MacKnight C, Bergman H, Hogan DB, McDowell I, et al. A global clinical measure of fitness and frailty in elderly people. CMAJ 2005;173:489–95.

32. Rockwood K, Andrew M, Mitnitski A. A comparison of two approaches to measuring frailty in elderly people. J Gerontol A Biol Sci Med Sci 2007;62A:738–43.

33. Theou O, Brothers TD, Mitnitski A, Rockwood K. Operationalization of frailty using eight commonly used scales and comparison of their ability to predict all- cause mortality. J Am Geriatr Soc 2013;61:1537–51.

34. Rockwood K, Rockwood MR, Mitnitski A. Physiological redundancy in older adults in relation to the change with age in the slope of a frailty index. J Am Geriatr Soc 2010;58:318–23.

35. Rockwood K, Blodgett JM, Theou O, Sun MH, Feridooni HA, Mitnitski A, et al. A frailty index based on deficit accumulation quantifies mortality risk in humans and in mice. Sci Rep 2017;7:43068.

36. Mitnitski A, Rockwood K. The rate of aging: the rate of deficit accumulation does not change over the adult life span. Biogerontology 2016;17:199–204.

SLICC-­FI­IN­PATIENTS­WITH­SLE­ |      1307

37. Sutton EJ, Davidson JE, Bruce IN. The Systemic Lupus International Collaborating Clinics (SLICC) Damage Index: a systematic literature review. Semin Arthritis Rheum 2013;43:352–61.

38. Gladman DD, Urowitz MB, Rahman P, Ibañez D, Tam LS. Accrual of organ damage over time in patients with systemic lupus erythematosus. J Rheumatol 2003;30:1955–9.

39. Rutenberg AD, Mitnitski AB, Farrell SG, Rockwood K. Unifying aging and frailty through complex dynamical networks. Exp Gerontol 2018;107:126–9.

40. Rockwood K, Mitnitski A, Song X, Steen B, Skoog I. Long- term risks of death and institutionalization of elderly people in relation

to deficit accumulation at age 70. J Am Geriatr Soc 2006;54: 975–9.

41. Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, et al. Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci 2001;56A:M146–56.

42. Katz PP, Andrews J, Yazdany J, Schmajuk G, Trupin L, Yelin  E. Is frailty a relevant concept in SLE? Lupus Sci Med 2017;4: e000186.

43. Urowitz MB, Gladman DD, Tom B, Ibanez D, Farewell VT. Changing patterns in mortality and disease outcomes for patients with systemic lupus erythematosus. J Rheumatol 2008;35:2152–8.

1308

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1308–1318DOI 10.1002/art.40863 © 2019, American College of Rheumatology

Pim- 1 as a Therapeutic Target in Lupus NephritisRong Fu,1 Yong Xia,2 Meirong Li,3 Renxiang Mao,1 Chaohuan Guo,1 Mianjing Zhou,1 Hechang Tan,4 Meiling Liu,2 Shuang Wang,1 Niansheng Yang,1 and Jijun Zhao1

Objective. Lupus nephritis (LN) is a major determinant of morbidity and mortality in systemic lupus erythemato-sus (SLE). Pim- 1 regulates lymphocyte proliferation and activation. The role of Pim- 1 in autoimmune disease remains unclear. This study was undertaken to test the hypothesis that inhibition of Pim- 1 would have therapeutic potential in patients with LN.

Methods. Pim- 1 expression was analyzed in lupus- prone (NZB × NZW)F1 mice (n = 6), human peripheral blood mononuclear cells (PBMCs) from SLE patients (n = 10), and glomeruli from patients with LN (n = 8). The therapeutic effect of the Pim- 1 inhibitor AZD1208 was assessed in the same murine lupus model (n = 10 mice per group). In vitro analysis was conducted to explore the mechanisms of action of Pim- 1 in mouse and human podocytes after Pim- 1 expression had been induced by anti–double- stranded DNA (anti- dsDNA) antibody–positive serum. Finally, MRL/lpr mice were used to confirm the therapeutic effects of Pim- 1 inhibition in vivo (n = 10 mice per group).

Results. Up- regulation of Pim- 1 was seen in renal lysates from diseased (NZB × NZW)F1 mice and in PBMCs from patients with SLE and renal biopsy tissue from patients with LN, relative to their control counterparts (each P < 0.05). The Pim- 1 inhibitor AZD1208 reduced the severity of proteinuria, glomerulonephritis, renal immune complex de-posits, and serum anti- dsDNA antibody levels, concomitant with the suppression of NFATc1 expression and NLRP3 inflammasome activation, in diseased (NZB × NZW)F1 mice (each P < 0.05 versus controls). Moreover, in mouse and human podocytes, Pim- 1 knockdown with targeted small interfering RNA (siRNA) suppressed NFATc1 and NLRP3 inflammasome signaling in the presence of anti- dsDNA–positive serum (each P < 0.05 versus control siRNA). Mech-anistically, Pim- 1 modulated NLRP3 inflammasome activation through intracellular Ca2+ (P < 0.05 versus normal controls). The therapeutic effect of Pim- 1 blockade was replicated in MRL/lpr mice.

Conclusion. These data identify Pim- 1 as a critical regulator of LN pathogenesis in patients with SLE. Targeting of the Pim- 1/NFATc1/NLRP3 pathway might therefore have therapeutic potential in human LN.

INTRODUCTION

Systemic lupus erythematosus (SLE) is a prototypical auto-immune disease that is characterized by generation of numer-ous autoantibodies. Lupus nephritis (LN) is a common and life- threatening manifestation of SLE (1). Despite improvements in therapy and patient outcomes over the past 50 years, the rate of complete remission in patients with severe LN is <50% (2).

The Pim family of serine/threonine kinases, consisting of Pim- 1, Pim- 2, and Pim- 3, controls cell survival, proliferation, and apo-ptosis, and these kinases have been highly conserved throughout evolution (3). In addition to being extensively studied in tumorigen-

esis, the role of Pim kinases has recently been revealed in inflam-matory and autoimmune settings. Pim- 1 phosphorylates RelA/p65 and activates NF- κB signaling in inflammatory conditions (4,5). Furthermore, Pim- 1 phosphorylates the human transcription factor FoxP3 to restrict the immunosuppressive activity of human Treg cells under conditions of inflammation (6). Inhibition of Pim- 1 could skew T cell differentiation toward Treg cells (7) and attenuate Th17 cell differentiation (8). Pim- 1 inhibition was found to ame-liorate colitis in mice via the suppression of excessive Th1- and Th17- type immune responses (7). Moreover, via its association with CD180, the long form of Pim- 1 kinase can transmit inflamma-tory processes into B cell survival programs, suggesting that the

Supported by the National Natural Science Foundation of China (grants 81601404 and 81660122) and the Natural Science Foundation of Guangdong Province (grant 2016A030313186).

1Rong Fu, MD, PhD, Renxiang Mao, MD, Chaohuan Guo, MD, Mianjing Zhou, MD, Shuang Wang, MD, PhD, Niansheng Yang, MD, PhD, Jijun Zhao, MD, PhD: First Affiliated Hospital, Sun Yat-sen University, Guangzhou, China; 2Yong Xia, MD, PhD, Meiling Liu, BS: Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China; 3Meirong Li, BS: Third Affiliated Hospital, Sun Yat-sen University, Guangzhou, China; 4Hechang Tan, MD,

PhD: Fourth Affiliated Hospital of Guangxi Medical University, Guangxi, China.

Drs. Fu and Xia and Ms Li contributed equally to this work.No potential conflicts of interest relevant to this article were reported.Address correspondence to Jijun Zhao, MD, PhD, First Affiliated Hospital,

Sun Yat-sen University, 58 Zhongshan Road II, Guangzhou 510080, China. E-mail: zhjij@mail.sysu.edu.cn.

Submitted for publication June 6, 2018; accepted in revised form February 14, 2019.

PIM- 1 PROMOTION OF LUPUS NEPHRITIS |      1309

pharmacologic inhibitory properties of this Pim kinase may pro-vide a novel therapeutic option for autoimmune diseases involving increased B cell activity (9). However, the functional role of Pim- 1 in LN remains unclear.

The calcineurin inhibitors, such as cyclosporine, which act through inhibition of NFATc signaling, are widely used to treat human LN (10–12). Of interest, Pim- 1 kinase can enhance the expression and activity of NFATc1 (4,13). In addition, we and other researchers have shown that the NLRP3 inflammasome participates in LN (14–17). In a recent study, we demonstrated that NLRP3 inflammasome activation in podocytes contributed to podocyte injuries and development of proteinuria (18). In the pres-ent study, we aimed to investigate the functional role of Pim- 1 in LN and its potential association with NLRP3 and NFATc1 signaling in podocytes.

MATERIALS AND METHODS

Lupus models. The animal study protocol was approved by the Institutional Animal Care Committee of Sun Yat- sen University. Female (NZB × NZW)F1 mice (The Jackson Laboratory), female MRL/lpr mice (Shanghai SLAC Laboratory Animal Company), and normal female control mice (C57BL/6 mice; Sun Yat- sen Univer-sity) were maintained in a specific pathogen–free barrier facility at the Experimental Animal Center at Sun Yat- sen University.

Treatment protocols. Female (NZB × NZW)F1 mice were orally treated with AZD1208 (15 mg/kg; Selleck Chemicals) or vehicle control (0.1% Tween 80 and 0.5% methyl cellulose in water; Sigma- Aldrich) (19,20) for 12 weeks (n = 10 mice per group), starting at age 22 weeks (at the time of onset of protein-uria). Mice were then placed under anesthesia and killed at age 34 weeks.

Twelve- week- old MRL/lpr mice received the selective Pim- 1 inhibitor SMI- 4a (60 mg/kg; Selleck Chemicals) or vehicle control (DMSO/PEG- 400/Tween 80) twice daily, as described previously (21). Oral gavage was administered on 5 of 7 days each week for 8 weeks (n = 10 mice per group). In an independent experiment, survival was observed in mice until age 30 weeks, and the survival rates were compared between 2 groups (n = 15 mice per group).

Human blood samples. For human studies, written informed consent was obtained from all subjects and the study was approved by the Ethics Committee of First Affiliated Hospital, Sun Yat- sen University. All included patients fulfilled the American College of Rheumatology updated classification criteria for SLE (22). Patients with concurrent infection, malignancy, or other auto-immune diseases were excluded from the study. Disease activity was assessed using the SLE Disease Activity Index (SLEDAI) (23). The cohort comprised female patients with active disease (SLE-DAI score ≥10) and those in remission (SLEDAI score ≤4) at the time of study inclusion.

Human peripheral blood mononuclear cells (PBMCs) and anti–double- stranded DNA (anti- dsDNA) antibody–positive serum samples were collected from subjects. Anti- dsDNA antibody–negative serum from healthy volunteers was used as a control.

Immunohistochemical analysis of human kidney specimens. Written informed consent was obtained from all sub-jects, and the study was approved by the Ethics Committee of The Fourth Affiliated Hospital of Guangxi Medical University. Renal specimens were obtained from patients with LN (n = 8 patients, of whom 4 had class IV LN, 2 had class IV plus class V LN, and 2 had class V LN) at the Department of Nephrology of the Fourth Affiliated Hospital of Guangxi Medical University; classification of LN was done according to the International Society of Nephrol-ogy/Renal Pathology Society 2003 classification criteria (24).

For immunohistochemical analyses, paraffin- embedded kid-ney sections (4 μm) obtained from patients with LN were dewaxed in xylene and rehydrated in graded ethanol solutions. Antigen retrieval was enhanced by microwaving the slides in 0.01M citrate buffer (pH 6). Sections were incubated with anti–Pim- 1 primary antibody (dilution 1:100; Abcam) overnight at 4°C. The sections were then incubated with a horseradish peroxidase–conjugated secondary antibody (Beyotime) at room temperature for 30 min-utes.

Antibody binding was visualized with 3,3′- diaminobenzidine tetrahydrochloride before brief counterstaining with hematoxylin. As a negative control, the primary antibody was replaced with phosphate buffered saline containing 1% bovine serum albumin. Nonmalignant renal tissue samples, which had normal findings on light micrography, came from surgical removal of kidney tissue from patients with renal cancer (n = 8); these samples were used as a normal control. The staining intensity of Pim- 1 was scored in a blinded manner by 2 independent pathologists on a scale of 0–3, as previously described (25).

Measurements of urinary protein. Urine samples were collected from patients every 2 weeks. Proteinuria was mea-sured using a dipstick (Multistix 10SG; Bayer Diagnostics), and the severity of urinary proteinuria was scored on a scale of 0–4 as previously reported (15). In addition, the concentration of urinary albumin was assessed by enzyme- linked immunosorbent assay (ELISA) (Bethyl Laboratories), and levels of urinary creatinine were measured using a Creatinine Colorimetric Assay kit (Cayman Chemical) according to the manufacturer’s instructions. The uri-nary albumin- to- creatinine ratio was then calculated (expressed in μg/mg).

Periodic acid–Schiff (PAS) staining and immunofluo-rescence analysis of mouse kidneys. Kidneys harvested from mice were fixed in 10% neutral formalin and embedded in paraffin for sectioning (4 μm). The kidney sections were then stained with PAS using standard methods. The severity of glomerular lesions

FU ET AL 1310       |

was graded on a scale of 0–3, as previously described, by 2 inde-pendent observers who were blinded with regard to the study groups (15).

An immunofluorescence assay of the kidney tissue was con-ducted using fluorescein isothiocyanate–conjugated anti- mouse IgG antibodies (Santa Cruz Biotechnology) and anti- mouse com-plement C3 antibodies (Cedarlane). Results (expressed as fluores-cence intensity) were assessed on a scale of 0–3 and analyzed as described previously (15). Images were obtained with a fluores-cence microscope (Olympus).

Measurement of anti- dsDNA antibodies. Serum anti- dsDNA antibodies were detected by ELISA, as previously described (15). The absorbance at an optical density of 450 nm was determined. Normal IgG was used as a negative control.

Caspase 1 activity assay. Caspase 1 activity was deter-mined using a Caspase- Glo 1 Inflammasome Assay (Promega) according to the manufacturer’s instructions. In brief, cells were seeded in a 96- well plate. After treatment, an equal volume of Caspase- Glo 1 reagent was added to the culture medium, after the medium had been equilibrated to room temperature for 1 hour. Cells were shaken for 5 minutes and incubated at room temperature for 30 minutes. Luminescence was measured using a Synergy 2 Microplate Reader (BioTek).

Single- cell suspension analysis by flow cytometry. Preparations of single- cell suspensions of glomeruli and analy-ses by flow cytometry were performed as described previously (18). Single- cell suspensions were stained for podocytes using an anti- mouse phycoerythrin- conjugated CD26 antibody (Bio-Legend), Alexa Fluor 647–conjugated anti- nephrin antibody (Bioss), eFluor 780/allophycocyanin–conjugated CD45 antibody (eBioscience), and anti–Pim- 1 antibody (Cell Signaling Technol-ogy). All analyses were done on a Gallios flow cytometer (Beck-man Coulter).

Single- cell suspensions were prepared from splenocytes, and the percentages of Th1 (CD4+IFNγ+) and Th17 (CD4+IL- 17+) cells were analyzed using an anti- mouse interleukin- 17A (IL- 17A) antibody and anti–interferon- γ (anti- IFNγ) antibody (both from BD PharMingen), as described previously (15).

Preparation of mouse and human podocytes. Mouse podocytes, conditionally immortalized by introducing a temperature- sensitive SV40T antigen by transfection, were purchased from Shanghai Ruilu Technology (catalog no. FDCC- MSN059). In brief, cells were grown on type I collagen–coated (BD Biosciences) culture plates under growth- permissive con-ditions at 33°C in RPMI 1640 medium supplemented with 10% fetal bovine serum (Gibco), 20 units/ml mouse recombinant IFNγ (Sigma), and 1% penicillin–streptomycin mixture (Sigma) under an atmosphere of 5% CO2. For differentiation, when the cell

density reached 70–80% confluence, podocytes were cultured under non–growth- permissive conditions at 37°C in the absence of IFNγ for 10–14 days.

Conditionally immortalized human podocytes (AB8/13) were kindly provided by Dr. Moin A. Saleem (Bristol, UK) (26) and cultured in RPMI 1640 medium with the addition of 10% fetal bovine serum (Gibco), 1% penicillin–streptomycin (Sigma), and insulin–transferrin–selenium (Life Technologies). The cells proliferated at 33°C and grew to 70–80% conflu-ence, followed by differentiation at 37°C for 10–14 days. Podocytes were incubated in the medium and stimulated with either anti- dsDNA antibody–positive serum from SLE patients or anti- dsDNA antibody–negative control serum (5% final con-centration) for 24–72 hours.

Pim- 1 knockdown. Small interfering RNAs (siRNAs) tar-geting Pim- 1 (siPim- 1) and nontargeting control siRNA (siNC) were synthesized at RiboBio. All cell transfections with siRNA were conducted with Lipofectamine 3000 (ThermoFisher Sci-entific), following the manufacturer’s instructions. After different treatments were added, IL- 1β in the supernatants was quanti-fied by ELISA (R&D Systems) according to the manufacturer’s instructions. Cells were then collected and subjected to Western blot analysis of Pim- 1 expression.

Real- time quantitative polymerase chain reaction (qPCR). Total RNA was extracted using TRIzol reagent (Life Technologies) according to the manufacturer’s protocol. Com-plementary DNA was obtained with a PrimeScript RT reagent kit (TaKaRa) and subjected to real- time qPCR analysis. GAPDH was used as an internal control. Amplification cycles were 95°C for 10 minutes, followed by 40 cycles at 95°C for 15 seconds and 60°C for 1 minute. The primers used for qPCR were syn-thesized at Shanghai Generay Biotech, as follows: human Pim- 1, forward 5′- GAGAAGGACCGGATTTCCGAC- 3′ and reverse 5′- CAGTCCAGGAGCCTAATGACG- 3′; human GAPDH, forward 5′- TGTGGG CATCAATGGATTTGG- 3′ and reverse 5′- ACACCATGTATTCCGGGTCAAT- 3′; mouse Pim- 1, for-ward 5′- TTCGGCTCGGTCTACTCTGG- 3′ and reverse 5′- CAG TTCTCCCCA ATCGGAAATC- 3′; mouse IFNγ, forward 5’- TACACA CTGCATCTTGGCTTTG- 3’ and reverse 5’- CTTCCACATCTATG CCACTTGAG- 3’; mouse IL- 17, forward 5′- GCTCCAGAA GGC CCTCAGA- 3′ and reverse 5′- AGCTTTCCCTCCGCATTGA- 3′; and mouse GAPDH, forward 5′- TTGTCATGGGAG TGAACGA-GA- 3′ and reverse 5′- CAGGCAGTTGGTGGT ACAGG- 3′.

Western blot analysis. Total protein extraction was per-formed using an M- PER mammalian protein extraction reagent (ThermoFisher Scientific), and the concentrations were determined using a BCA protein assay kit (ThermoFisher Scientific). Proteins were separated by 10% sodium dodecyl sulfate–polyacrylamide gel electrophoresis and then transferred to a polyvinylidene diflu-

PIM- 1 PROMOTION OF LUPUS NEPHRITIS |      1311

oride membrane (Millipore). The membrane was treated with 3% bovine serum albumin in Tris buffered saline (TBS) solution, followed by incubation with primary antibodies against Pim- 1 (Cell Signaling Technology), NLRP3 (AdipoGen), caspase 1- p20 (Santa Cruz Bio-technology), NFATc1 (Abcam), and GAPDH (Cell Signaling Tech-nology). After washing 3 times with TBS–Tween 20, horseradish peroxidase–labeled secondary antibodies were used for detection. The signals on the membranes were visualized using an Enhanced Chemiluminescence detection kit (ThermoFisher Scientific).

Intracellular Ca2+ analysis. A cell- permeant Fluo- 3 ace-toxymethylester (Fluo- 3 AM) probe was purchased from Yeasen Corporation and used for intracellular Ca2+ detection according to the manufacturer’s instructions. Fluo- 3 AM was dissolved in DMSO with 20% Pluronic F127 to prevent its aggregation in Hanks’ balanced salt solution (HBSS) (without Ca2+ or Mg2+) and to promote cellular uptake. The Fluo- 3 AM solution was diluted with HBSS to prepare 4 μM working solution. The suspension was incubated in a water bath for 50 minutes at 37°C, and then

Figure 1. Increased Pim- 1 expression in mouse and human systemic lupus erythematosus (SLE). A, Top, Expression of Pim- 1 mRNA in the kidneys of (NZB × NZW)F1 mice, in groups of prediseased mice (pre- D) (no proteinuria, age 12 weeks), diseased mice (D) (3+ proteinuria, age 34 weeks), and age- matched C57BL/6 normal (N) controls (n = 6 mice per group). Bottom, Representative Western blot bands of Pim- 1 protein expression in the mouse kidneys. B, Top, Expression of Pim- 1 mRNA in human peripheral blood mononuclear cells (PBMCs) from normal controls, SLE patients with active disease, and SLE patients in remission (anti–double- stranded DNA antibody negative) (n = 10 subjects per group). Bottom, Representative Western blot bands of Pim- 1 protein expression in human PBMCs. Values in A and B were normalized to the levels of GAPDH. C, Immunohistochemical analysis of Pim- 1 expression in representative glomeruli from 2 patients with lupus nephritis (LN) (panels c and d) in comparison with a normal nephrectomy specimen (panel a) and a negative control (panel b). Right, Pim- 1 staining intensity analysis. Data are the mean ± SD of 8 samples per group. * = P < 0.05 versus normal controls.

FU ET AL 1312       |

centrifuged for 5 minutes. The resulting fluorescence intensity, as the indicator of Ca2+ concentration, was observed by flow cytom-etry (FACScan Flow Cytometer; Beckman- Coulter) at an excita-tion wavelength of 488 nm and emission wavelength of 525 nm.

After treatment with siPim- 1 or siNC for 24 hours, prior to stimulation with anti- dsDNA–positive serum for 24 hours, cells were treated with the intracellular calcium chelator BAPTA- AM (50 μM; Merck Millipore) to assess involvement of intracellular calcium in the Pim- 1/NLRP3 pathway.

Statistical analysis. Student’s t- test or analysis of variance was used for comparison between 2 or more groups. Kaplan- Meier analysis with log- rank tests was used to compare survival rates in mice. Data are expressed as the mean ± SD. P values less than 0.05 were considered significant. All data were processed using SPSS software version 17.0.

RESULTS

Enhanced expression of Pim- 1 in mouse and human SLE. First, in a lupus model using (NZB × NZW)F1 mice, we compared Pim- 1 expression between prediseased mice (no proteinuria, age 12 weeks), diseased mice with severe protein-uria (3+ proteinuria, age 34 weeks), and age- matched normal control mice. As the proteinuria developed, expression levels of Pim- 1 messenger RNA (mRNA) and protein were significantly elevated in the kidneys of diseased mice (n = 6 mice per group) (Figure 1A). In addition, Pim- 1 mRNA and protein levels were also up- regulated in PBMCs from SLE patients with active disease, as compared with PBMCs from SLE patients in remission (anti- dsDNA negative) and PBMCs from healthy subjects (n = 10 sub-jects per group) (Figure 1B).

In patients with LN, the results of immunohistochemical analysis of the glomeruli showed strong nuclear and cytoplas-

Figure  2. Treatment with the Pim- 1 inhibitor AZD1208 reduces lupus- like syndrome and suppresses NFATc1 expression and NLRP3 inflammasome activation in (NZB × NZW)F1 mice compared with vehicle- treated controls. A, Urinary albumin- to- creatinine ratio (ACR) (in μg/mg). B, Left, Representative periodic acid–Schiff–stained images of glomerular areas. Original magnification × 400. Right, Histologic staining intensity scores of the glomeruli. C, Left, Renal deposition of C3 and IgG. Right, Fluorescence intensity analysis of renal deposits. D, Serum anti–double- stranded DNA (anti- dsDNA) antibody levels. E, Western blot bands (representative kidney samples from 1 mouse per band) showing Pim- 1, NFATc1, and NLRP3 inflammasome component expression. F, Caspase 1 activity. Results are shown as the fold change relative to vehicle- treated controls. G, Production of renal interleukin- 1β (IL- 1β) in the mouse kidneys. Data are the mean ± SD of 10 mice per group. * = P < 0.05 versus vehicle- treated controls. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40863/abstract.

PIM- 1 PROMOTION OF LUPUS NEPHRITIS |      1313

mic staining of Pim- 1 (n = 8) (Figure 1C). In contrast, Pim- 1 was poorly expressed in normal nephrectomy specimens (n = 8).

Attenuation of lupus- like syndrome in (NZB × NZW)F1 mice by Pim- 1 inhibition. We next treated (NZB × NZW)F1 mice with a Pim- 1 inhibitor, AZD1208, which had demon-strated preclinical efficacy in several cancers (19,20,27,28). Proteinuria (according to the urinary albumin- to- creatinine

ratio) was remarkably delayed in AZD1208- treated animals over the 12- week observation period (Figure 2A). Furthermore, AZD1208- treated mouse kidneys showed decreased severity of glomerulonephritis relative to vehicle- treated controls (Fig-ure 2B). We also observed a significant decrease in renal depos-its of IgG and C3 after treatment with AZD1208 (Figure 2C). In addition, anti- dsDNA antibody levels in the mouse serum were significantly reduced by AZD1208 (Figure 2D).

Figure  3. Induction of Pim- 1 expression by anti–double- stranded DNA (anti- dsDNA) antibody–positive serum in mouse and human podocytes. A, Left, Flow cytometry analysis of single- cell suspensions of podocytes from mouse kidneys, comparing AZD1208- treated mice with vehicle- treated controls. Right, Quantification of the flow cytometry data. Data are the mean fluorescence intensity (MFI) (mean ± SD) of Pim- 1 expression in podocytes from 10 mice per group. * = P < 0.05 versus controls. B, Pim- 1 induction in mouse podocytes after 0–72 hours of stimulation with anti- dsDNA–positive serum from diseased (NZB × NZW)F1 mice (3+ proteinuria, age 34 weeks). C, Pim- 1 induction in human podocytes after 0–72 hours of stimulation with anti- dsDNA–positive serum from treatment- naive patients with systemic lupus erythematosus. Values were normalized to the levels of GAPDH. SSC = side scatter. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40863/abstract.

Figure  4. Pim- 1 knockdown inhibits NFATc1 and NLRP3 inflammasome signaling in mouse and human podocytes. A and B, Silencing efficacy of Pim- 1 small interfering RNAs (siPIM- 1 #1 and #2) against Pim- 1 mRNA and protein expression in mouse podocytes was assessed at 48 hours by quantitative polymerase chain reaction and Western blotting. A nontargeting siRNA (siNC) was used as the control. C, Suppression of NFATc1 and NLRP3 inflammasome activity in mouse podocytes by Pim- 1 silencing. D, Suppression of caspase 1 activity in mouse podocytes by Pim- 1 silencing. E, Reduction of supernatant interleukin- 1β (IL- 1β) production in mouse podocytes by Pim- 1 silencing. F, Suppression of NFATc1 and NLRP3 inflammasome activity in human podocytes by Pim- 1 silencing. G, Suppression of caspase 1 activity in human podocytes by Pim- 1 silencing. Results in D and G are shown as the fold change relative to controls. H, Reduction of supernatant IL- 1β production in human podocytes by Pim- 1 silencing. Data are the mean ± SD from 3 independent experiments. * = P < 0.05 versus siNC.

FU ET AL 1314       |

Suppression of NFATc1 expression and NLRP3 inflammasome activation in mouse kidneys by Pim- 1 inhibition. We then investigated the potential effect of Pim- 1 inhibition on NFATc1 expression and NLRP3 inflammasome activation. Pim- 1 inhibition significantly reduced renal NFATc1 expression in (NZB × NZW)F1 mice (Figure  2E). Furthermore, renal expression of NLRP3 and caspase 1- p20 was significantly suppressed by AZD1208 treatment (Figure  2E). The inhibitory effects of AZD1208 on caspase 1 activity was further confirmed using the caspase 1 inflammasome assay (Figure 2F). Consist-ent with these results, renal expression of IL- 1β was inhibited by AZD1208 treatment in (NZB × NZW)F1 mice (Figure 2G).

Induction of Pim- 1 expression by anti- dsDNA–posi-tive serum in mouse and human podocytes. In flow cytom-etry analyses of single- cell suspensions of kidneys from (NZB ×

NZW)F1 mice, we found that Pim- 1 expression was significantly decreased in podocytes from AZD1208- treated mice compared with vehicle- treated controls (Figure 3A). Subsequently, we found that anti- dsDNA–positive serum from diseased (NZB × NZW)F1 mice (3+ proteinuria, age 34 weeks) induced Pim- 1 expression in mouse podocytes, reaching its maximum at 24 hours (Fig-ure 3B). In contrast, control serum from prediseased mice had no such effect. Consistent with these findings, anti- dsDNA–positive serum from treatment- naive SLE patients remarkably enhanced the expression of Pim- 1 in human podocytes at 24 hours (Fig-ure 3C).

Hampering of NFATc1 and NLRP3 inflammasome signaling in mouse and human podocytes by Pim- 1 knockdown. The silencing efficacy of Pim- 1 was assessed at 48 hours by qPCR and Western blot analyses, and the more effec-

Figure 5. Pim- 1 promotes NLRP3 inflammasome activation via intracellular Ca2+. A, Effects of Pim- 1 knockdown by a Pim- 1 small interfering RNA (siPim- 1) on the intracellular Ca2+ concentration in human podocytes in the presence or absence of human anti–double- stranded DNA (anti- dsDNA) antibody–positive serum versus normal serum. A nontargeting siRNA (siNC) was used as a control. B, Caspase 1 activity in the presence or absence of anti- dsDNA–positive serum versus normal serum, and inhibitory effects of an intracellular Ca2+ blocker, BAPTA- AM, whose effects were comparable to those of siPim- 1. Results in A and B are shown as the fold change relative to normal serum controls. C, Production of interleukin- 1β (IL- 1β) in the presence or absence of anti- dsDNA–positive serum versus normal serum, and inhibitory effects of an intracellular Ca2+ blocker, BAPTA- AM, whose effects were comparable to those of siPim- 1. Data in A–C are the mean ± SD from 3 independent experiments. D, Proposed diagram of the possible role of Pim- 1 in the development of lupus nephritis. * = P < 0.05 versus the respective controls. ns = not significant. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40863/abstract.

PIM- 1 PROMOTION OF LUPUS NEPHRITIS |      1315

tive siRNAs were selected for further analyses (Figures 4A and B). Results showed that Pim- 1 knockdown significantly suppressed NFATc1 expression as compared with controls (Figure 4C). Pim- 1 silencing also profoundly down- regulated the activity of NLRP3, caspase 1- p20, and caspase 1 and the resulting IL- 1β production in mouse podocytes (Figures 4C–E).

Moreover, human podocytes stimulated with anti- dsDNA–positive serum from human patients with SLE were assessed in the presence or absence of siPim- 1. As expected, Pim- 1 knock-down suppressed autoantibody- induced NFATc1 expression and NLRP3 inflammasome activation (Figures 4F–H). These data sug-gest that Pim- 1 modulates NFATc1 and NLRP3 signaling path-ways in podocytes.

Modulation of NLRP3 inflammasome activation through intracellular Ca2+ in the presence of Pim- 1. Intra-cellular Ca2+ signaling regulates NLRP3 inflammasome activation. Our results showed that, compared with normal serum, anti- dsDNA–positive serum from SLE patients induced a pronounced increase in the intracellular Ca2+ concentration in human podo-cytes, and this was inhibited by Pim- 1 knockdown (Figure 5A).

Moreover, the increased levels of caspase 1 activity and IL- 1β pro-duction induced by anti- dsDNA–positive serum from SLE patients were both attenuated by an intracellular Ca2+ blocker, BAPTA- AM, and such an inhibitory effect was comparable to that of siPim- 1 (Figures  5B and C), suggesting that Pim- 1 regulates NLRP3 inflammasome activation via intracellular Ca2+. In summary, based on these findings, we propose a model in which Pim- 1 activates NFATc1 and NLRP3 inflammasome signaling and plays vital roles in the development of LN (Figure 5D).

Attenuation of renal disease and reduced frequency of mortality in MRL/lpr mice following selective inhibition of Pim- 1. To further confirm our results, we investigated the effect of a selective Pim- 1 inhibitor, SMI- 4a, in MRL/lpr mice. SMI- 4a profoundly attenuated glomerular damage in the mice (Figure 6A). Moreover, treatment with SMI- 4a attenuated the onset of protein-uria as compared with that in vehicle- treated control mice (Fig-ure 6B). Strikingly, SMI- 4a significantly prolonged the survival of MRL/lpr mice (Figure 6C).

In addition, after 6 weeks of AZD1208 treatment, the Th1 and Th17 responses were inhibited in 14-week-old MRL/lpr

Figure 6. Selective inhibition of Pim- 1 attenuates renal disease and improves survival in MRL/lpr mice. A, Left, Attenuating effects of the selective Pim- 1 inhibitor SMI- 4a on glomerular damage, as compared with that in vehicle- treated control mice. Right, Histologic staining intensity scores for glomerular damage. B, Effects of SMI- 4a treatment on the urinary albumin- to- creatinine ratio (ACR) (in μg/mg). Data in A and B are the mean ± SD in 10 mice per group. C, Survival of MRL/lpr mice up to 30 weeks of age (n = 15 mice per group). * = P < 0.05 versus vehicle- treated controls. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40863/abstract.

FU ET AL 1316       |

mice, as demonstrated by a reduced percentage of Th1 and Th17 cells in the mouse spleens and decreased expression of IFNγ and IL-17 in the mouse kidney tissue (see results in Sup-plementary Figures 1A and B, available on the Arthritis & Rheu-matology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40863/ abstract).

DISCUSSION

Initial investigations have largely focused on the oncogenic role of Pim- 1 in human cancers (29). In this study, we have shown that Pim- 1 inhibition attenuates the renal damage and related immunopathology in lupus- prone (NZB × NZW)F1 mice. Mecha-nistic investigations showed that Pim- 1 promoted LN via NFATc1 and NLRP3 signaling in podocytes. Notably, selective inhibition of Pim- 1 conferred renal protection and extended survival benefits in another lupus model involving MRL/lpr mice. Thus, our data extend the current knowledge of Pim- 1 to the field of autoimmune disease, providing evidence that Pim- 1 is a promising therapeutic target in human LN.

A novel and important finding in this study is that Pim- 1 reg-ulates the NLRP3 inflammasome in LN. Immune complexes can activate the NLRP3 inflammasome and induce IL- 1β production from human monocytes (30). Activation of the NLRP3 inflam-masome in human lupus–derived macrophages enhances pro-duction of inflammatory cytokines (31). Caspase 1, the central component of the NLRP3 inflammasome, is essential for autoan-tibody production and renal damage in pristane- induced lupus (32). In addition to findings from our previous studies (14–16,18), Lu and colleagues demonstrated that an overactivated NLRP3 inflammasome in myeloid cells leads to severe organ damage in experimental lupus (33). It was recently reported that primary trophoblast cells transfected with Pim- 1 siRNA decreased IL- 1β production, when the cells were stimulated with lipopolysaccha-ride or tumor necrosis factor α (34). The results of the present study indicate that Pim- 1 inhibition can suppress NLRP3/IL- 1β signaling in lupus- prone mice and in vitro.

An increased intracellular Ca2+ concentration or Ca2+ mobi-lization is critical for NLRP3 inflammasome activation (35,36). Calcium dynamics were significantly enhanced in Pim- 1–overex-pressing mouse transgenic hearts and depressed in Pim- 1–defi-cient mouse hearts (37). Further investigation in the present study revealed that knockdown of Pim- 1 offset the increased intracellu-lar Ca2+ concentration in anti- dsDNA–positive serum–stimulated human podocytes, and intracellular Ca2+ blockade was as effec-tive as Pim- 1 silencing in neutralizing the activation of the NLRP3 inflammasome. These results show that Pim- 1 modulates NLRP3 inflammasome activation via intracellular Ca2+.

Another valuable finding in this study is that Pim- 1 pro-motes LN via NFATc1 induction. NFATc1 is activated in MRL/lpr mouse T cells, and helps B cells to produce autoantibod-ies (38). NFATc1 is activated in anti- CD3+ anti- CD28–activated

PBMCs from SLE patients (39). As a basic maintenance ther-apy for SLE, hydroxychloroquine inhibits intracellular calcium release and NFATc1 expression in SLE patients (40). Impor-tantly, NFATc1 signaling mediates podocyte injury and loss, promoting proteinuria and, eventually, kidney failure (41–43). Conditional NFATc1 activation in podocytes can cause protein-uria in vivo (41). Pertinent to this study, Pim- 1 interacts with NFATc1 and regulates its activity (4,13). Pim- 1 can phospho-rylate NFATc1 and enhance its transcriptional activity directly in a phosphorylation- dependent manner (13). However, unlike the other known NFATc kinases, Pim- 1 enhances NFATc1- dependent transactivation and IL- 2 production in Jurkat T cells, without any effects on the subcellular localization of NFATc1 (13). Kim and colleagues reported that Pim- 1 modulates RANKL- induced osteoclastogenesis via NFATc1 induction (4). In this study, we demonstrate that Pim- 1 acts as a modulator of NFATc1 signaling. The improvement in glomerular inflammation was attributed, at least partly, to the inhibition of Th1 and Th17 responses. Furthermore, decreased B cell activity due to Pim- 1 inhibition may account for the reduction in the titers of anti- dsDNA antibodies (9).

Previous studies showed that pathogenic anti- DNA autoantibodies complexed with high mobility group binding protein 1 activates the receptor for advanced glycation end products (RAGE) (44–46), which activates Pim- 1 (47). Thus, we speculate that anti- dsDNA antibody–positive serum induces Pim- 1 expression via RAGE in podocytes. Taken together, these results establish a pivotal role for Pim- 1 in the patho-genesis of LN. We thus provide novel evidence to support the hypothesis that targeting Pim- 1 may provide a promising strat-egy for the treatment of human LN.

ACKNOWLEDGMENT

We thank Dr. Xinhui Liu for technical support.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Zhao had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Fu, Xia, Li, Zhao.Acquisition of data. Fu, Xia, Li, Mao, Guo, Zhou, Tan, Liu, Wang, Yang, Zhao.Analysis and interpretation of data. Fu, Xia, Li, Zhao.

REFERENCES 1. Dall’Era M. Treatment of lupus nephritis: current paradigms and

emerging strategies. Curr Opin Rheumatol 2017;29:241–7.

2. Davidson A. What is damaging the kidney in lupus nephritis? Nat Rev Rheumatol 2016;12:143–53.

3. Nawijn MC, Alendar A, Berns A. For better or for worse: the role of Pim oncogenes in tumorigenesis. Nat Rev Cancer 2011;11:23–34.

PIM- 1 PROMOTION OF LUPUS NEPHRITIS |      1317

4. Kim K, Kim JH, Youn BU, Jin HM, Kim N. Pim- 1 regulates RANKL- induced osteoclastogenesis via NF- κB activation and NFATc1 induc-tion. J Immunol 2010;185:7460–6.

5. Nihira K, Ando Y, Yamaguchi T, Kagami Y, Miki Y, Yoshida K. Pim- 1 controls NF- κB signalling by stabilizing RelA/p65. Cell Death Differ 2010;17:689–98.

6. Li Z, Lin F, Zhuo C, Deng G, Chen Z, Yin S, et al. PIM1 kinase phos-phorylates the human transcription factor FOXP3 at serine 422 to negatively regulate its activity under inflammation. J Biol Chem 2014;289:26872–81.

7. Shen YM, Zhao Y, Zeng Y, Yan L, Chen BL, Leng AM, et al. Inhibition of Pim- 1 kinase ameliorates dextran sodium sulfate- induced colitis in mice. Dig Dis Sci 2012;57:1822–31.

8. Wang M, Okamoto M, Domenico J, Han J, Ashino S, Shin YS, et al. Inhibition of Pim1 kinase prevents peanut allergy by enhancing Runx3 expression and suppressing T(H)2 and T(H)17 T- cell differen-tiation. J Allergy Clin Immunol 2012;130:932–44.

9. Egli N, Zajonz A, Burger MT, Schweighoffer T. Human CD180 trans-mits signals via the PIM- 1L kinase. PLoS One 2015;10:e0142741.

10. Mehta J, Genin A, Brunner M, Scalzi LV, Mishra N, Beukelman T, et al. Prolonged expression of CD154 on CD4 T cells from pedi-atric lupus patients correlates with increased CD154 transcription, increased nuclear factor of activated T cell activity, and glomerulone-phritis. Arthritis Rheum 2010;62:2499–509.

11. Rao A. Signaling to gene expression: calcium, calcineurin and NFAT. Nat Immunol 2009;10:3–5.

12. Chighizola CB, Ong VH, Meroni PL. The use of cyclosporine A in rheumatology: a 2016 comprehensive review. Clin Rev Allergy Im-munol 2017;52:401–23.

13. Rainio EM, Sandholm J, Koskinen PJ. Cutting edge: transcription-al activity of NFATc1 is enhanced by the Pim- 1 kinase. J Immunol 2002;168:1524–7.

14. Zhao J, Zhang H, Huang Y, Wang H, Wang S, Zhao C, et al. Bay11- 7082 attenuates murine lupus nephritis via inhibiting NLRP3 inflammasome and NF- κB activation. Int Immunopharmacol 2013;17:116–22.

15. Zhao J, Wang H, Dai C, Wang H, Zhang H, Huang Y, et al. P2X7 blockade attenuates murine lupus nephritis by inhibiting acti-vation of the NLRP3/ASC/caspase 1 pathway. Arthritis Rheum 2013;65:3176–85.

16. Zhao J, Wang H, Huang Y, Zhang H, Wang S, Gaskin F, et al. Lupus nephritis: glycogen synthase kinase 3β promotion of renal damage through activation of the NLRP3 inflammasome in lupus- prone mice. Arthritis Rheumatol 2015;67:1036–44.

17. Kahlenberg JM, Kaplan MJ. The inflammasome and lupus: another innate immune mechanism contributing to disease pathogenesis? Curr Opin Rheumatol 2014;26:475–81.

18. Fu R, Guo C, Wang S, Huang Y, Jin O, Hu H, et al. Podocyte acti-vation of NLRP3 inflammasomes contributes to the development of proteinuria in lupus nephritis. Arthritis Rheumatol 2017;69:1636–46.

19. Keeton EK, McEachern K, Dillman KS, Palakurthi S, Cao Y, Grondine MR, et al. AZD1208, a potent and selective pan- Pim kinase inhibitor, demonstrates efficacy in preclinical models of acute mye-loid leukemia. Blood 2014;123:905–13.

20. Brasó-Maristany F, Filosto S, Catchpole S, Marlow R, Quist J, Francesch-Domenech E, et al. PIM1 kinase regulates cell death, tu-mor growth and chemotherapy response in triple- negative breast cancer. Nat Med 2016;22:1303–13.

21. Lin YW, Beharry ZM, Hill EG, Song JH, Wang W, Xia Z, et al. A small molecule inhibitor of Pim protein kinases blocks the growth of precursor T- cell lymphoblastic leukemia/lymphoma. Blood 2010;115:824–33.

22. Hochberg MC. Updating the American College of Rheumatology re-vised criteria for the classification of systemic lupus erythematosus [letter]. Arthritis Rheum 1997;40:1725.

23. Bombardier C, Gladman DD, Urowitz MB, Caron D, Chang CH, and the Committee on Prognosis Studies in SLE. Derivation of the SLEDAI: a disease activity index for lupus patients. Arthritis Rheum 1992;35:630–40.

24. Weening JJ, D’Agati VD, Schwartz MM, Seshan SV, Alpers CE, Appel GB, et al. The classification of glomerulonephritis in system-ic lupus erythematosus revisited. J Am Soc Nephrol 2004;15:241–50.

25. Yellin MJ, D’Agati V, Parkinson G, Han AS, Szema A, Baum D, et al. Immunohistologic analysis of renal CD40 and CD40L expression in lupus nephritis and other glomerulonephritides. Arthritis Rheum 1997;40:124–34.

26. Saleem MA, O’Hare MJ, Reiser J, Coward RJ, Inward CD, Farren T, et al. A conditionally immortalized human podocyte cell line demonstrating nephrin and podocin expression. J Am Soc Nephrol 2002;13:630–8.

27. Bellon M, Lu L, Nicot C. Constitutive activation of Pim- 1 ki-nase is a therapeutic target for adult T- cell leukemia. Blood 2016;127:2439–50.

28. Kirschner AN, Wang J, van der Meer R, Anderson PD, Franco-Coronel OE, Kushner MH, et al. PIM kinase inhibitor AZD1208 for treatment of MYC- driven prostate cancer. J Natl Cancer Inst 2014;107.

29. Warfel NA, Kraft AS. PIM kinase (and Akt) biology and signaling in tumors. Pharmacol Ther 2015;151:41–9.

30. Shin MS, Kang Y, Lee N, Wahl ER, Kim SH, Kang KS, et al. Self double- stranded (ds)DNA induces IL- 1β production from human monocytes by activating NLRP3 inflammasome in the presence of anti- dsDNA antibodies. J Immunol 2013;190:1407–15.

31. Kahlenberg JM, Carmona-Rivera C, Smith CK, Kaplan MJ. Neu-trophil extracellular trap- associated protein activation of the NLRP3 inflammasome is enhanced in lupus macrophages. J Immunol 2013;190:1217–26.

32. Kahlenberg JM, Yalavarthi S, Zhao W, Hodgin JB, Reed TJ, Tsuji NM, et al. An essential role of caspase 1 in the induction of murine lu-pus and its associated vascular damage. Arthritis Rheumatol 2014;66:152–62.

33. Lu A, Li H, Niu J, Wu S, Xue G, Yao X, et al. Hyperactivation of the NLRP3 inflammasome in myeloid cells leads to severe organ dam-age in experimental lupus. J Immunol 2017;198:1119–29.

34. Szydłowski M, Prochorec-Sobieszek M, Szumera-Ciećkiewicz A, Derezińska E, Hoser G, Wasilewska D, et al. Expression of PIM kinases in Reed- Sternberg cells fosters immune privilege and tumor cell survival in Hodgkin lymphoma. Blood 2017;130:1418–29.

35. Lee GS, Subramanian N, Kim AI, Aksentijevich I, Goldbach-Mansky R, Sacks DB, et al. The calcium- sensing receptor regulates the NLRP3 inflammasome through Ca2+ and cAMP. Nature 2012;492:123–7.

36. Murakami T, Ockinger J, Yu J, Byles V, McColl A, Hofer AM, et al. Critical role for calcium mobilization in activation of the NLRP3 in-flammasome. Proc Natl Acad Sci U S A 2012;109:11282–7.

37. Muraski JA, Rota M, Misao Y, Fransioli J, Cottage C, Gude N, et al. Pim- 1 regulates cardiomyocyte survival downstream of Akt. Nat Med 2007;13:1467–75.

38. Kyttaris VC, Zhang Z, Kampagianni O, Tsokos GC. Calcium signaling in systemic lupus erythematosus T cells: a treatment target. Arthritis Rheum 2011;63:2058–66.

39. Lu MC, Lai NS, Yu HC, Hsieh SC, Tung CH, Yu CL. Nifedipine sup-presses Th1/Th2 cytokine production and increased apoptosis of anti- CD3 + anti- CD28- activated mononuclear cells from patients with systemic lupus erythematosus via calcineurin pathway. Clin Im-munol 2008;129:462–70.

40. Wu SF, Chang CB, Hsu JM, Lu MC, Lai NS, Li C, et al. Hydroxychlo-roquine inhibits CD154 expression in CD4+ T lymphocytes of sys-

FU ET AL 1318       |

temic lupus erythematosus through NFAT, but not STAT5, signaling. Arthritis Res Ther 2017;19:183.

41. Wang Y, Jarad G, Tripathi P, Pan M, Cunningham J, Martin DR, et al. Activation of NFAT signaling in podocytes causes glomerulosclero-sis. J Am Soc Nephrol 2010;21:1657–66.

42. Pedigo CE, Ducasa GM, Leclercq F, Sloan A, Mitrofanova A, Hashmi T, et al. Local TNF causes NFATc1- dependent cholesterol- mediated podocyte injury. J Clin Invest 2016;126:3336–50.

43. Zhang H, Liang S, Du Y, Li R, He C, Wang W, et al. Inducible ATF3- NFAT axis aggravates podocyte injury. J Mol Med (Berl) 2018;96:53–64.

44. Qing X, Pitashny M, Thomas DB, Barrat FJ, Hogarth MP, Putterman C. Pathogenic anti- DNA antibodies modulate gene expression in mesan-

gial cells: involvement of HMGB1 in anti- DNA antibody- induced renal injury. Immunol Lett 2008;121:61–73.

45. Urbonaviciute V, Voll RE. High- mobility group box 1 represents a potential marker of disease activity and novel therapeutic target in systemic lupus erythematosus. J Intern Med 2011;270:309–18.

46. Yung S, Chan TM. Anti- dsDNA antibodies and resident renal cells: their putative roles in pathogenesis of renal lesions in lupus nephritis. Clin Immunol 2017;185:40–50.

47. Meloche J, Paulin R, Courboulin A, Lambert C, Barrier M, Bonnet P, et al. RAGE- dependent activation of the oncoprotein Pim1 plays a critical role in systemic vascular remodeling processes. Arterioscler Thromb Vasc Biol 2011;31:2114–24.

Clinical Images: Tracheobronchial cobblestone in relapsing polychondritis

The patient, a 30- year- old man previously in good health, developed a high fever, hoarseness, joint pain in the left hand, and redness of the eyes. Computed tomography (CT) of the chest revealed diffuse circumferential thickening of the trachea and both main bronchi (A) (arrow). Symptoms did not respond to empirical antibiotics, and after they had persisted for 2 months, he was referred to our hospital. At presentation to us, his temperature was 38.4°C. Physical examination revealed bilateral conjunctival injection, tenderness along the nasal bridge, and rhonchi without crackles on auscultation. The high- sensitivity C- reactive protein (hsCRP) level was elevated (17.43 mg/dl [normal <0.5]). Rheumatoid factor, anti–cyclic citrullinated peptide antibodies, and antineutrophil cytoplasmic antibodies (ANCAs) were absent. Bronchoscopy revealed cobblestoning of the tracheal mucosa (B). CT of the head did not show sinusitis. Positron emission tomography revealed extensive uptake along the tracheobronchial tree and cartilage of the nose and right ear (C). Biopsy showed chronic inflammation of the trachea and submucosal fibrosis without granuloma. The patient later developed swelling and tenderness in cartilage of the right ear, sparing the lobe. The absence of sinusitis, granulomatous inflammation, and ANCAs ruled out a diagnosis of granuloma-tosis with polyangiitis. Relapsing polychondritis was diagnosed, and high- dose corticosteroid therapy was initiated. Symptoms promptly resolved, and the patient’s hsCRP level decreased to 0.03 mg/dl.

Eun Bong Lee, MD, PhDJin Kyun Park, MD, PhDSeoul National University College of MedicineSeoul, Republic of Korea

DOI: 10.1002/art.40922

FU ET AL 1318       |

temic lupus erythematosus through NFAT, but not STAT5, signaling. Arthritis Res Ther 2017;19:183.

41. Wang Y, Jarad G, Tripathi P, Pan M, Cunningham J, Martin DR, et al. Activation of NFAT signaling in podocytes causes glomerulosclero-sis. J Am Soc Nephrol 2010;21:1657–66.

42. Pedigo CE, Ducasa GM, Leclercq F, Sloan A, Mitrofanova A, Hashmi T, et al. Local TNF causes NFATc1- dependent cholesterol- mediated podocyte injury. J Clin Invest 2016;126:3336–50.

43. Zhang H, Liang S, Du Y, Li R, He C, Wang W, et al. Inducible ATF3- NFAT axis aggravates podocyte injury. J Mol Med (Berl) 2018;96:53–64.

44. Qing X, Pitashny M, Thomas DB, Barrat FJ, Hogarth MP, Putterman C. Pathogenic anti- DNA antibodies modulate gene expression in mesan-

gial cells: involvement of HMGB1 in anti- DNA antibody- induced renal injury. Immunol Lett 2008;121:61–73.

45. Urbonaviciute V, Voll RE. High- mobility group box 1 represents a potential marker of disease activity and novel therapeutic target in systemic lupus erythematosus. J Intern Med 2011;270:309–18.

46. Yung S, Chan TM. Anti- dsDNA antibodies and resident renal cells: their putative roles in pathogenesis of renal lesions in lupus nephritis. Clin Immunol 2017;185:40–50.

47. Meloche J, Paulin R, Courboulin A, Lambert C, Barrier M, Bonnet P, et al. RAGE- dependent activation of the oncoprotein Pim1 plays a critical role in systemic vascular remodeling processes. Arterioscler Thromb Vasc Biol 2011;31:2114–24.

Clinical Images: Tracheobronchial cobblestone in relapsing polychondritis

The patient, a 30- year- old man previously in good health, developed a high fever, hoarseness, joint pain in the left hand, and redness of the eyes. Computed tomography (CT) of the chest revealed diffuse circumferential thickening of the trachea and both main bronchi (A) (arrow). Symptoms did not respond to empirical antibiotics, and after they had persisted for 2 months, he was referred to our hospital. At presentation to us, his temperature was 38.4°C. Physical examination revealed bilateral conjunctival injection, tenderness along the nasal bridge, and rhonchi without crackles on auscultation. The high- sensitivity C- reactive protein (hsCRP) level was elevated (17.43 mg/dl [normal <0.5]). Rheumatoid factor, anti–cyclic citrullinated peptide antibodies, and antineutrophil cytoplasmic antibodies (ANCAs) were absent. Bronchoscopy revealed cobblestoning of the tracheal mucosa (B). CT of the head did not show sinusitis. Positron emission tomography revealed extensive uptake along the tracheobronchial tree and cartilage of the nose and right ear (C). Biopsy showed chronic inflammation of the trachea and submucosal fibrosis without granuloma. The patient later developed swelling and tenderness in cartilage of the right ear, sparing the lobe. The absence of sinusitis, granulomatous inflammation, and ANCAs ruled out a diagnosis of granuloma-tosis with polyangiitis. Relapsing polychondritis was diagnosed, and high- dose corticosteroid therapy was initiated. Symptoms promptly resolved, and the patient’s hsCRP level decreased to 0.03 mg/dl.

Eun Bong Lee, MD, PhDJin Kyun Park, MD, PhDSeoul National University College of MedicineSeoul, Republic of Korea

DOI: 10.1002/art.40922

1319

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1319–1328DOI 10.1002/art.40864 © 2019, American College of Rheumatology

Diagnostic Accuracy of Positron Emission Tomography/Computed Tomography of the Head, Neck, and Chest for Giant Cell Arteritis: A Prospective, Double- Blind, Cross- Sectional StudyAnthony M. Sammel,1 Edward Hsiao,2 Geoffrey Schembri,3 Katherine Nguyen,2 Janice Brewer,2 Leslie Schrieber,3 Beatrice Janssen,2 Peter Youssef,4 Clare L. Fraser,5 Elizabeth Bailey,2 Dale L. Bailey,3 Paul Roach,3 and Rodger Laurent3

Objective. Positron emission tomography/computed tomography (PET/CT) has not been well studied as a first- line test for giant cell arteritis (GCA), due, in part, to historical limitations in visualizing the cranial arteries. The Giant Cell Arteritis and PET Scan (GAPS) study was therefore carried out to assess the accuracy of a newer generation PET/CT of the head, neck, and chest for determining a diagnosis of GCA.

Methods. In the GAPS study cohort, 64 patients with newly suspected GCA underwent time- of- flight PET/CT (1- mm slice thickness from the vertex to diaphragm) within 72 hours of starting glucocorticoids and before undergo-ing temporal artery biopsy (TAB). Two physicians with experience in PET reviewed the patients’ scans in a blinded manner and reported the scans as globally positive or negative for GCA. Tracer uptake was graded across 18 artery segments. The clinical diagnosis was confirmed at 6 months’ follow- up.

Results. In total, 58 of 64 patients underwent TAB, and 12 (21%) of the biopsies were considered positive for GCA. Twenty- one patients had a clinical diagnosis of GCA. Compared to TAB, the sensitivity of PET/CT for a diag-nosis of GCA was 92% (95% confidence interval [95% CI] 62–100%) and specificity was 85% (95% CI 71–94%). The negative predictive value (NPV) was 98% (95% CI 87–100%). Compared to clinical diagnosis, PET/CT had a sensitivity of 71% (95% CI 48–89%) and specificity of 91% (95% CI 78–97%). Interobserver reliability was moderate (κ = 0.65). Among the enrolled patients, 20% had a clinically relevant incidental finding, including 7 with an infection and 5 with a malignancy. Furthermore, 5 (42%) of 12 TAB- positive GCA patients had moderate or marked aortitis.

Conclusion. The high diagnostic accuracy of this PET/CT protocol would support its use as a first- line test for GCA. The NPV of 98% indicates the particular utility of this test in ruling out the condition in patients considered to be at lower risk of GCA. PET/CT had benefit over TAB in detecting vasculitis mimics and aortitis.

INTRODUCTION

Giant cell arteritis (GCA), also known as temporal arter­itis, is a medium­ to­ large vessel vasculitis of the elderly that can cause acute irreversible vision loss, aortic aneurysm, and peripheral artery stenosis. Rapid and accurate diagnosis is critical to allow the urgent introduction of glucocorticoid­

based therapy in those with the disease and prevent morbidity resulting from inappropriate treatment in those with alternative diagnoses.

Temporal artery biopsy (TAB) is the traditional test to confirm the diagnosis of GCA and remains widely used in clinical prac­tice (1). Although a positive finding of GCA on TAB definitively confirms the diagnosis, a negative finding on biopsy does not

ClinicalTrials.gov identifier: NCT02771483.Supported by Arthritis Australia.1Anthony M. Sammel, MBBS: Royal North Shore Hospital, University of

Sydney, and Prince of Wales Hospital, Sydney, New South Wales, Australia; 2Edward Hsiao, MBChB, Katherine Nguyen, MD, Janice Brewer, MBBS, Beatrice Janssen, MD, Elizabeth Bailey, PhD: Royal North Shore Hospital, Sydney, New South Wales, Australia; 3Geoffrey Schembri, MBBS, Leslie Schrieber, MD, Dale L. Bailey, PhD, Paul Roach, MBBS, Rodger Laurent, MD: Royal North Shore Hospital and University of Sydney, Sydney, New South Wales, Australia; 4Peter Youssef, PhD: University of Sydney and Royal Prince

Alfred Hospital, Sydney, New South Wales, Australia; 5Clare L. Fraser, MMed: Save Sight Institute, Faculty of Health and Medicine, University of Sydney, Sydney, New South Wales, Australia.

Dr. Youssef has received consulting fees from Roche and Novartis (less than $10,000 each). No other disclosures relevant to this article were reported.

Address correspondence to Anthony M. Sammel, MBBS, Royal North Shore Hospital, Department of Rheumatology, St. Leonards, Sydney, New South Wales 2065, Australia. E-mail: anthony.sammel@health.nsw.gov.au.

Submitted for publication December 4, 2018; accepted in revised form February 19, 2019.

SAMMEL ET AL 1320       |

exclude it, since the disease has a propensity to skip artery seg­ments or primarily affect large vessels (2–6). In the absence of a better gold standard, the true sensitivity of TAB is unknown, but it is likely to be 85–90% at best (1,4).

Positron emission tomography (PET)/computed tomogra­phy (CT) scans can detect large vessel vasculitis in the aorta and subclavian, carotid, iliac, and/or femoral arteries in ~80% of patients with TAB­ positive GCA (7,8). Its diagnostic accuracy, however, has not been robustly studied in unselected cohorts of patients with newly suspected GCA (9,10).

Recent studies have provided better guidance as to how to use PET/CT in GCA. Scans should ideally be performed within 72 hours from the initiation of glucocorticoid treatment, as the intensity and distribution of vascular tracer uptake decreases with treatment duration. A study published in 2018 demonstrated that the diag­nostic sensitivity of PET/CT for large vessel vasculitis was main­tained at 3 days after the commencement of glucocorticoids, but was decreased at 10 days (11). Newer generation time­ of­ flight PET/CT scanners can more precisely calculate the arrival time of photons and provide improved image resolution compared to con­ventional scanners. Time­ of­ flight PET/CT has recently been shown to detect arteritis in the smaller temporal, occipital, maxillary, and vertebral arteries, all of which are typically involved in GCA but have not been well visualized on older PET/CT scanners (12–14). A ret­rospective case–control study demonstrated that PET/CT scans of these vessels provided good discrimination between GCA patients and controls, with a sensitivity of 82% and specificity of 100% (14).

In the present study, we hypothesized that a newer generation time­ of­ flight PET/CT scan of the head, neck, and chest would be able to detect the presence of inflammation in vascular sites typi­cally affected by GCA, including both the cranial and large arteries, and thus provide high diagnostic accuracy for the condition.

PATIENTS AND METHODS

Study design and participants. The Giant Cell Arteritis and PET Scan (GAPS) study was designed as a prospective, double­ blind, cross­ sectional study conducted at Royal North Shore Hospital, a tertiary referral center in Sydney, Australia, between May 2016 and July 2018. Ethics approval was granted by the Northern Sydney Local Health District Human Research Ethics Committee (approval no. HREC/16/HAWKE/68).

Patients who were newly suspected of having GCA were referred by specialist rheumatologists, ophthalmologists, neurolo­gists, or clinical immunologists. Hospitalized and ambulatory clinic patients were eligible for inclusion. Consecutively referred patients were screened and enrolled, after provision of informed consent, if they met the following prespecified inclusion criteria: 1) age >50 years, 2) having fulfilled at least 2 of the 5 American College of Rheumatology (ACR) 1990 classification criteria for GCA (15), 3) having been scheduled for, but not yet undergone, TAB, and 4) having received glucocorticoids for <72 hours at the time of the

PET/CT scan. Patients were excluded if they had known active malignancy, had a history of connective tissue disease or vascu­litis, or had been taking glucocorticoids for a single period of >1 week during the preceding 6 months.

Clinical information was prospectively recorded at enrollment by means of a standardized data collection template. Responses were entered after conducting a detailed patient history review and physical examination. Results from ancillary ophthalmologic examinations and vascular imaging studies (performed at the discretion of the treating clinician and not standardized for this study) were also recorded at that time, along with determination of the C­ reactive protein (CRP) level and erythrocyte sedimenta­tion rate (ESR) at a time point most closely corresponding to the commencement of glucocorticoids.

Test methods. PET/CT scan as the index test. PET/CT scans were performed prior to TAB on a single Siemens Bio­graph mCT time­ of­ flight scanner. Patients fasted for at least 4 hours before they received an intravenous injection of 100 MBq fluorine­ 18 fluoro­ 2­ deoxyglucose (FDG). At 60 minutes after the infusion of the FDG tracer, patients were scanned from the ver­tex of the head to the diaphragm with 1­ mm CT reconstruction. Arms were positioned by the side to allow better visualization of the head and neck vessels.

The full set of scans was independently read between July and August 2018 by 2 nuclear medicine physicians (EH and GS) who were experienced in PET scans. These physician read­ers were blinded with regard to all clinical, imaging, and biopsy data. To calibrate reporting parameters, the physicians co­ read 6 unblinded training scans from non–study patients at 14 months prior to the scan readings. In addition, at that time, they performed a limited blinded reading of 20 study scans to ensure that the study protocol was providing interpretable results.

The primary reporting outcome was a subjective global assessment of the scan as being positive or negative for GCA. Specific criteria were not provided to the readers, and the assess­ment was made based on the intensity and distribution of FDG vascular uptake. In general, the readers were more confident in reporting a scan as being positive for GCA in the following circum­stances: 1) artery wall FDG uptake markedly increased compared to the blood pool, 2) artery wall FDG uptake increased diffusely along its length (in contrast to focal uptake in regions such as the carotid bifurcation, which are known to have a high atherosclerotic burden), or 3) cranial artery involvement (as these vessels are less likely to demonstrate the mild physiologic uptake that can be seen in the aorta and primary branches).

The 2 physician readers also reported the intensity of FDG uptake in 18 artery segments: the bilateral temporal, occipital, maxillary, vertebral, carotid, subclavian, and axillary arteries, the brachiocephalic artery, and the ascending, arch, and descend­ing aorta. Vascular wall FDG uptake was compared to the back­ground intensity of the blood pool in the superior vena cava (16).

DIAGNOSTIC ACCURACY OF PET/CT FOR GCA |      1321

The grading system for vascular tracer uptake was based on established definitions (17) as follows: 0 = no FDG uptake (less than or equal to the blood pool), 1 = minimal/equivocally increased uptake, 2 = moderate/clearly increased uptake, and 3 = very marked uptake. The FDG grading schema for the right superficial temporal artery are illustrated in Figure 1. These results were used for an exploratory analysis of the diagnostic accuracy of PET/CT, using a cutoff value of either grade 1 or grade 2 uptake in any vessel to define a GCA­positive scan. Comparison of vascular to

liver FDG uptake was not performed because the scan field did not routinely include liver parenchyma.

Discordant results relating to each patient’s global scan assessment and the maximum FDG uptake grade in any artery segment were resolved by a consensus reading by the 2 nuclear medicine physicians. A consensus FDG uptake grade was not determined for all artery segments.

Clinically guided unilateral TAB as the reference standard. Clinically guided unilateral TAB was the prespecified reference

Figure 1. Axial head positron emission tomography/computed tomography scans illustrating the grading system for fluorine­ 18 fluoro­ 2­ deoxyglucose (FDG) uptake in the right superficial temporal artery (marked by circles). Arrows point to FDG uptake in the maxillary arteries. Grade 0 = no uptake (less than or equal to the blood pool); grade 1 = minimal/equivocally increased uptake; grade 2 = moderate/clearly increased uptake; grade 3 = very marked uptake.

SAMMEL ET AL 1322       |

standard. In order to minimize the chance of missing skip lesions, we cut through each temporal artery specimen at 0.25­ mm increments. One of 2 anatomic pathologists reviewed all sections for inflammation. The pathologists were blinded with regard to the PET/CT vascular findings but not to clinical details. In accordance with the majority of published literature, we defined a positive biopsy finding for GCA as having evidence of inflammation through one or more layers of the main artery wall (intima, media, and/or adventitia) (18). The presence of iso­lated vasa vasorum or periadventitial small vessel vasculitis (19) was noted but classified as a negative biopsy finding for GCA. A sequential contralateral biopsy was performed in a small number of patients at the discretion of the treating clinician.

Clinical diagnosis as the secondary reference test. Given that TAB has imperfect sensitivity for GCA, we also compared PET/CT to a secondary reference test, the 6­ month clinical diagnosis. This time point was chosen to allow time for clinicians to judge the response to steroid weaning and to confirm relevant alternative diagnoses. The 2016 TABUL ultrasound study (The Role of Ultrasound compared to Biopsy of Temporal Arteries in the Diagnosis and Treatment of Giant Cell Arter itis) also used this time point for confirmation of the final clinical diagnosis (5). Clinical follow­ up data were prospectively obtained at 2 weeks and 3 and 6 months after diagnosis by means of a standardized

phone survey, and by testing CRP and ESR levels. GCA disease flares were determined by the treating clinician.

Patients, treating clinicians, and reviewers were blinded with regard to the PET/CT vascular findings, to ensure that the clini­cal diagnosis was independent from the index test. They were, however, made aware of PET/CT detection of incidental findings, such as infection or cancer, to allow the timely confirmation and management of vasculitis mimics.

Patient diagnoses were confirmed according to the protocol shown in Figure 2. Treating clinicians were contacted in writing in June 2018 and asked if they thought their patient had GCA. They were also asked to choose the “most likely diagnosis for the clinical presentation leading to suspicion of GCA/temporal biopsy” from a prepopulated list of 28 options. Diagnosis by the treating clinician as well as the biopsy result and the level of glucocorticoid dosing at 3 months’ follow­ up were used to define “definite pos­itive” or “definite negative” cases of GCA, in accordance with the criteria for a definite diagnosis (as listed in Figure 2).

Twenty­ four patients were given a “definite positive” or “definite negative” diagnosis of GCA, and the remaining 40 underwent external case review. The reviewer panel comprised 5 rheumatologists and 1 neuro­ ophthalmologist who were not involved in the patients’ clinical care. Reviewers were provided a standardized set of data from enrollment to 6 months for each

Figure 2. Protocol to arrive at the clinical diagnosis (top), and criteria for a definite diagnosis of giant cell arteritis (GCA) (bottom). PET/CT = positron emission tomography/computed tomography.

DIAGNOSTIC ACCURACY OF PET/CT FOR GCA |      1323

patient. Enrollment data included details of symptoms, exam­ination findings, glucocorticoid dosing, inflammation markers, results of vascular imaging when performed (not including PET/CT), biopsy results, incidental PET/CT findings, and ophthalmo­logic assessments. Follow­ up data for the 2­ week, 3­ month, and 6­ month time points included prednisone dose, use of steroid­ sparing agents, active GCA symptoms, inflammation markers, details of GCA flares, and other objective clinical, imaging, histo­pathology, and serology findings that were available to the treat­ing clinician, such as a histologically confirmed malignancy. The final diagnosis required consensus between the treating clinician and at least 1 of 2 reviewers. For the 8 patients for whom con­sensus was not reached, a panel of 4 rheumatologist reviewers decided on the diagnosis.

Statistical analysis. Data analysis was performed using IBM SPSS version 25. The 95% confidence intervals (95% CIs) were calculated using the exact Clopper­ Pearson method. Patients who did not undergo TAB were excluded from the anal­ysis comparing PET/CT to TAB but were included in the assess­ment comparing PET/CT to clinical diagnosis. Interobserver reliability was assessed using the kappa statistic.

The initial target sample size was 69 patients, calculated to achieve a 10% CI around the anticipated test sensitivity of 90% (20). This assumed a 50% rate of TAB­ positive GCA. We expected our cohort to have a higher rate of TAB­ positive GCA compared to recent cohorts of patients with “suspected GCA,” given that we required the presence of 2 or more of the ACR 1990 classification criteria for GCA to be met for enroll­ment (1,17,21,22). After reaching 60 patients, we recalculated our required sample size as 170 based on our actual rate of TAB­ positive GCA of 20%. This was beyond the resources of our study, and we closed enrollment at the end of 2017 with 64 patients enrolled.

RESULTS

In total, 96 patients were referred to the study from 13 sites in Sydney, Australia, of whom 64 met the inclusion criter ia and underwent PET/CT. Of the 64 patients, 58 underwent TAB at a median of 4 days (range 0–21 days) after the PET/CT scan, and 95% of patients completed 6 months of fol­low­ up. The distribution of the study participants is presented in Figure 3.

Figure 3. Flow chart showing the distribution of the study participants. GCA = giant cell arteritis; ACR = American College of Rheumatology; TAB = temporal artery biopsy; PET/CT = positron emission tomography/computed tomography.

SAMMEL ET AL 1324       |

The median age of enrolled patients was 69 years, and 70% were female. At enrollment, 91% of patients had head­ache, 33% had polymyalgia rheumatica (PMR) symptoms, 33% had vision disturbance, and 28% had jaw claudication. The median CRP level was 21 mg/liter (mean 46 mg/ liter), and the median ESR was 41 mm/hour (mean 48 mm/hour). In total, 56 patients had a clinically guided unilateral TAB, and 2 had sequential bilateral biopsies. The median length of the first biopsy was 19 mm.

In total, 12 (21%) of the 58 patients who underwent TAB had evidence of mural inflammation consistent with GCA. Of the 46 patients with a negative biopsy finding, 35 (60%) had no sig­nificant inflammation and 11 (19%) had limited vasa vasorum or periadventitial small vessel vasculitis. The baseline demographic and clinical characteristics of the patients and the TAB results are presented in Table 1.

Among the 64 enrolled patients, 21 (33%) had a clinical diagnosis of GCA, and 43 were diagnosed as having other con­ditions. In total, 42 patients (66%) met the ACR 1990 classifica­tion criteria for GCA, with the caveat that these are not diagnostic criteria (15). A list of the final clinical diagnoses for the 64 enrolled

patients is presented in Table 2. Among the 5 patients who were ultimately diagnosed as having PMR, all 5 reported having head­ache and/or scalp tenderness at enrollment and 2 reported hav­ing transient visual disturbance. Their mean CRP level was 11 mg/liter and the mean ESR was 29 mm/hour. None of these patients had evidence of inflammatory change on unilateral TAB. Only 1 patient underwent dedicated large vessel vascular imag­ing (by ultrasound), the findings of which were negative for arteri­tis 2 weeks after commencement of glucocorticoids.

Levels of inflammation markers were highest in the biopsy­ positive GCA cohort. The mean CRP level and ESR was 98 mg/liter and 72 mm/hour, respectively, in TAB­ positive GCA patients (n = 12), 71 mg/liter and 62 mm/hour, respectively, in patients with clinically diagnosed GCA (n = 21), and 33 mg/liter and 41 mm/hour, respectively, in patients with alternative diagnoses (n = 43).

Among the enrolled patients, 11 of 12 with TAB­ positive GCA were positive for GCA by PET/CT, and 39 of 46 patients with a negative TAB were negative for GCA by PET/CT. For a diagnosis of GCA, this indicated that the PET/CT protocol had a sensitivity of 92%, specificity of 85%, positive predictive value (PPV) of 61%, negative predictive value (NPV) of 98%, and area under the receiver operating

Table 1. Baseline characteristics and temporal artery biopsy (TAB)results in the 64 enrolled study patients*

Age, median (range) years 69 (50–90)Sex, female 45 (70)Jaw claudication 18 (28)Polymyalgia rheumatica 21 (33)Headache 58 (91)Vision disturbance 21 (33)Temporal arteries, tender or reduced

pulse35 (55)

Occipital artery tenderness 12 (19)CRP, median (range) mg/liter 21 (1–280)ESR, median (range) mm/hour 41 (2–130)Meeting the ACR criteria for GCA† 42 (66)Undergoing TAB 58 (91)Days between PET/CT and TAB, median

(range)4 (0–21)

TAB length, median (range) mm 19 (7–32)Positive for GCA by TAB among those

undergoing TAB12 (21)

Negative for GCA by TAB among thoseundergoing TAB

46 (79)

No inflammation 35 (60)Limited VV or periadventitial SVV 11 (19)

* Except where indicated otherwise, values are the number (%) of patients. CRP = C- reactive protein; ESR = erythrocyte sedimenta-tion rate; PET/CT = positron emission tomography/computed to-mography; VV = vasa vasorum; SVV = small vessel vasculitis. † The American College of Rheumatology (ACR) 1990 classification criteria for giant cell arteritis (GCA) (15).

Table  2. Final clinical diagnosis accounting for the GCA presentation*

Clinical diagnosisNo. (%) of patients

GCA 21 (33)Cervicogenic headache 9 (24)PMR 5 (8)Self- limited ophthalmologic disease, other 4 (6)Infection, other 3 (5)Malignancy 3 (5)Pneumonia 3 (5)Dental abscess 2 (3)Headache, not otherwise specified 2 (3)Herpes zoster 2 (3)Inflammatory ocular disease (uveitis,

scleritis)2 (3)

Rheumatic disease† 2 (3)Chronic ophthalmologic disease, other 1 (2)Neurologic disease, other 1 (2)Sinusitis 1 (2)Stroke/TIA 1 (2)Thyroiditis 1 (2)Unknown 1 (2)

* GCA = giant cell arteritis; PMR = polymyalgia rheumatica; TIA = transient ischemic attack. † Rheumatic disease was defined as rheumatoid arthritis, connec-tive tissue disease, spondyloarthropathy, or sarcoidosis. No other definitions were provided to treating clinicians.

DIAGNOSTIC ACCURACY OF PET/CT FOR GCA |      1325

characteristic curve (AUC) of 0.88. Three of the 7 TAB­ negative, PET/CT­ positive GCA patients had a clinical diagnosis of GCA. Two of these patients had convincing GCA disease flares when treatment was withdrawn, with recrudescence of headache and elevated inflammation markers. These symptoms as well as the exacerbation of inflammation markers resolved with the reintroduction of glucocor­ticoid treatment. The third patient had hypoechoic wall thickening in the bilateral temporal and left axillary arteries on ultrasound, sugges­tive of arteritis. Details of the 8 discordant TAB and PET/CT cases are presented in Supplementary Table 1 (available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40864/ abstract). Patient­ level data (with 95% CIs) assessing the diagnostic performance of PET/CT compared to both TAB and the 6­ month clinical diagnosis are presented in Table 3.

The 2 readers disagreed on the PET/CT diagnosis for 9 patients on their initial scan review. These cases were subse­quently co­ read to arrive at the consensus diagnosis, with 6 of 9 scans being judged as positive for GCA. The clinical diagnoses of these 9 patients were as follows: biopsy­ positive GCA (3 patients), biopsy­ negative GCA (1 patient), PMR (1 patient), zoster ophthal­micus (1 patient), cervical osteomyelitis (1 patient), cervicogenic headache (1 patient), and headache not otherwise specified (1 patient). The diagnostic accuracy of the initial scan assessment was higher for reader 1 (EH) (sensitivity 83%, specificity 87%) than for reader 2 (GS) (sensitivity 75%, specificity 85%). Interobserver reliability was moderate (κ = 0.65).

Global PET/CT assessment provided higher overall diagnos­tic accuracy compared to use of an FDG uptake grade cutoff to define a positive scan. When grade 1 or higher FDG uptake in any vessel was used for the GCA­ positive scan cutoff, the sensitivity

was 100% and specificity was 46%. Interobserver reliability was poor (κ = 0.19). When grade 2 or higher FDG uptake in any ves­sel was used for the GCA­ positive scan cutoff, the sensitivity was 83% and specificity 83%, and interobserver reliability was moder­ate (κ = 0.65).

The distribution of FDG vascular uptake differed between patients. Some patients had predominantly cranial involvement, while others had predominantly large vessel involvement. Both readers noted that 3 (25%) of 12 TAB­ positive GCA patients had moderate FDG uptake limited to the head and neck (temporal, occipital, maxillary, and vertebral arteries), and 2 (17%) of 12 had moderate FDG uptake limited to the larger thoracic and/or carotid vessels. The distribution of vascular uptake by vascular territory for those with biopsy­ positive GCA, those with clinically positive GCA, and those with alternative diagnoses is presented in Sup­plementary Table 2 (http://onlin elibr ary.wiley.com/doi/10.1002/art.40864/ abstract).

Six patients underwent a dedicated vascular ultrasound study for GCA within 1 week of PET/CT scan. For the 2 patients who had negative findings on ultrasound, 1 had a negative TAB finding and the other did not undergo biopsy. Both patients had negative findings for GCA on their PET/CT scans, and both were given a final clinical diagnosis of cervicogenic headache. Four patients had positive findings of GCA on ultrasound, with hypoechoic tem­poral artery wall thickening. All 4 had globally positive findings on their PET/CT scans and had grade 1 or higher FDG uptake in at least one temporal artery. Three of the 4 patients were positive for GCA on biopsy, and 1 had a negative biopsy finding (limited periadventitial small vessel vasculitis). All 4 of these patients had a final clinical diagnosis of GCA.

Table 3. Diagnostic performance of PET/CT compared to TAB and clinical diagnosis

PET/CT index test

Reference test

TAB Clinical diagnosis

Positive for GCA

Negative for GCA

Positive for GCA

Negative for GCA

No. of patientsPositive for GCA 11 7 15 4Negative for GCA 1 39 6 39Total assessed 12 46 21 43

Performance*Sensitivity 92 (62–100) 71 (48–89)Specificity 85 (71–94) 91 (78–97)PPV 61 (36–83) 79 (54–94)NPV 98 (87–100) 87 (73–95)AUC 0.88 (0.79–0.98) 0.81 (0.70–0.92)

* Values for the performance of positron emission tomography/computed tomography (PET/CT), compared to temporal artery biopsy (TAB) and clinical diagnosis, in the diagnosis of giant cell arteritis (GCA), including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic curve (AUC), are the percentage (95% confidence interval).

SAMMEL ET AL 1326       |

In addition to the vascular findings, 13 (20%) of 64 patients had a clinically relevant incidental finding on PET/CT. This included 7 patients with an acute infection (3 cases of pneumonia, 3 cases of sinusitis, and 1 case of cervical osteomyelitis), 5 with a malig­nancy (4 with lung cancer and 1 with thyroid cancer), and 1 with subacute thyroiditis. None of these patients had a positive biopsy finding for GCA, while 1 of the 13 patients had a clinical diagnosis of GCA in addition to acute sinusitis that was detected on the PET/CT scan.

Moderate or marked FDG uptake was detected by at least one reader in the aorta of 11 (17%) of 64 patients. Eight of these patients had a clinical diagnosis of GCA and 5 had a positive TAB finding of GCA. The remaining 3 patients had other clinical diag­noses, including PMR (n = 1), zoster ophthalmicus (n = 1), and metastatic lung cancer (n = 1).

There were no reported adverse events from either the TAB or PET/CT scan. The mean ionizing radiation exposure for the PET/CT scan was 3.4 mSv (range 2.8–4.6 mSv).

DISCUSSION

This is the first study to assess the diagnostic accuracy of time­ of­ flight PET/CT scan of the head, neck, and chest in patients having newly suspected GCA. The technique had high diagnostic accuracy against both TAB (sensitivity 92%, specificity 85%) and the 6­ month clinical diagnosis (sensitivity 71%, specific­ity 91%). The NPV of PET/CT compared to TAB was particularly high, at 98%.

In contrast to previous PET/CT studies, which have generally included preselected patients with positive findings of GCA on TAB or with suspected large vessel disease (7,17), we included a real­ world, heterogenous cohort of consecutive patients who required confirmation of GCA or an alternative diagnosis. The positive biopsy rate of 21% and modest baseline elevations in the CRP level and ESR are in keeping with the data from con­temporary cohorts in Canada and the UK (5,18).

Our study is also unique in that it utilized a time­ of­ flight scanner with 1­ mm CT reconstruction. This allowed more detailed assessment of the smaller temporal, maxillary, occip­ital, and vertebral arteries, which are known to be involved in GCA but have previously been considered beyond the resolution of PET/CT. We prespecified a 72­ hour window from the com­mencement of glucocorticoids to PET/CT. Given the imperative to commence therapy immediately in high­ risk patients (23), this provided a clinically practical window to arrange a scan while minimizing the diagnostic impact of glucocorticoid­ related atten­uation of FDG vessel wall uptake (11).

While TAB remains the traditional gold standard test for GCA, there has been a trend toward the use of imaging to support the diagnosis. In 2018, a European League Against Rheumatism group released recommendations on imaging in large vessel vasculitis (24). Ultrasound and high­ resolution

scalp magnetic resonance imaging (MRI) were listed as suitable first­ line tests to confirm a diagnosis of GCA. PET was recom­mended for evaluating large vessel extracranial arteritis but not for the assessment of cranial arteries or for use as a general first­ line imaging modality. Our study challenges this view. The diagnostic performance of PET/CT of the head, neck, and chest compares well with recent large, high­ resolution scalp MRI and ultrasound studies. Our protocol, when compared against TAB, performed similarly to that shown in a 2016 study of high­ resolution scalp MRI (18) for the diagnosis of GCA (MRI versus TAB, sensitivity 94%, specificity 78%). The performance of our PET/CT protocol (relative to both TAB and clinical diagnosis) was superior to that in the TABUL ultrasound study (5) (ultra­sound versus TAB, sensitivity 73%, specificity 69%); ultrasound versus clinical diagnosis, sensitivity 54%, specificity 81%). It is important to note that the TABUL study had a lower diagnostic accuracy than many other ultrasound studies in GCA (25,26). While only 6 of our patients underwent dedicated cranial ultra­sound as part of their diagnostic workup, all cases had diag­nostic agreement between the ultrasound and PET/CT scan.

This study also clarifies how best to report PET/CT scans in GCA. We found that the global scan assessment by a PET­ experienced nuclear medicine physician provided better accuracy than using a predefined vascular FDG uptake grade cutoff. In part, this may be a reflection of the ability of readers to vary the diagnostic weight of FDG uptake by location, taking into consideration the particular vessels involved, the diffuse versus localized nature of vascular uptake, and the potential for inflammatory atheroma at particular vascular sites such as the carotid bifurcation.

PET/CT had utility in diagnosing vasculitis mimics. In this study, a clinically relevant incidental finding was present in 1 in 5 patients. One patient with cervical osteomyelitis may have had a serious adverse outcome if he had been treated with high­ dose glucocorticoids while awaiting TAB.

Aortitis was a common finding in our biopsy­ positive GCA cohort. Moderate or marked aortic wall FDG uptake was detected in 5 (42%) of 12 TAB­ positive GCA patients. It was also detected in a further 3 patients with a clinical diagnosis of GCA. These patients may be at higher risk of developing aneurysms, with implications for the duration and frequency of aortic surveillance (27).

Despite the promising diagnostic performance and ability to detect mimicking conditions, PET/CT does have shortcomings, including its cost and availability. PET/CT is generally considered to be more expensive than ultrasound or MRI. The low­ dose, 100­ MBq FDG head, neck, and chest protocol used in this study was less costly than a standard full­ body, 250–300­ MBq PET/CT scan.

Our study has a number of limitations. First, due to the fact that the percentage of TAB­ positive GCA cases was lower than predicted, the 95% CI around the sensitivity was wider than our  desired CI of 10%. Second, our primary PET/CT reporting

DIAGNOSTIC ACCURACY OF PET/CT FOR GCA |      1327

measure, the global assessment for GCA, is subjective. Although we have specified some general principles that our physician read­ers used to increase their confidence in reporting a scan as being positive for GCA, accuracy ultimately requires the readers to be experienced in interpreting vessel wall changes on PET/CT. This includes an awareness that a high atheroma burden can result in increased FDG uptake and mimic arteritis (28). Diagnostic accu­racy was higher with the final consensus diagnosis compared to that achieved by either of the 2 independent readers. This sug­gests the need for careful assessment of all artery segments and the potential benefit of dual reporting of equivocal scans. The moderate interobserver reliability (κ = 0.65) in our study is similar to that described in the TABUL study (5) for ultrasound and TAB reporting (intraclass correlation coefficient 0.61 and 0.62, respec­tively) and to that of superficial cranial MRI (κ = 0.68) (29).

Moving forward, we will consider testing a number of techni­cal modifications that may improve scan resolution. These include increasing the dose of FDG to 150 mSv and delaying the scan to 90 minutes after FDG injection. A longer delay between injection and scan may reduce background blood pool FDG activity and potentially allow better determination between vessel wall and blood pool uptake.

In summary, time­ of flight PET/CT scan of the head, neck, and chest with 1­ mm CT reconstruction had high diagnos­tic accuracy compared to TAB for the diagnosis of GCA. The results compare well with recent ultrasound and MRI studies. This study would support a first­ line role for PET/CT in the assessment of patients newly suspected of having GCA. The high NPV indicates particular value in excluding the diagnosis in lower risk patients. Given the significant morbidity associ­ated with a misdiagnosis, we believe that TAB remains indi­cated when the scan is inconclusive or discordant with the pretest probability of GCA. PET/CT had additional benefits over TAB, including detection of clinically relevant incidental findings in 20% of patients, and identification of aortitis in more than 40% of biopsy­ positive GCA patients.

ACKNOWLEDGMENTS

The authors thank the following people for assistance with the study: Dr. H. Soh, Mr. N. Hall, Ms R. Asher, and the Royal North Shore Hospital nuclear medicine and ophthalmology department nurses, registrars, and consultants. The authors also acknowledge the generous translational research grant pro­vided by Arthritis Australia.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti­cally for important intellectual content, and all authors approved the final version to be published. Dr. Sammel had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Sammel, Hsiao, Schembri, Brewer, Schrieber, Janssen, Youssef, Fraser, D. L. Bailey, Roach, Laurent.Acquisition of data. Sammel, Hsiao, Schembri, Nguyen, Brewer, Schrie­ber, Janssen, Youssef, Fraser, E. Bailey.Analysis and interpretation of data. Sammel, Hsiao, Schembri, Nguyen, Schrieber, E. Bailey, D. L. Bailey, Roach, Laurent.

REFERENCES 1. Hall S, Persellin S, Lie JT, O’Brien PC, Kurland LT, Hunder GG. The

therapeutic impact of temporal artery biopsy. Lancet 1983;2:1217–20.

2. Klein RG, Campbell RJ, Hunder GG, Carney JA. Skip lesions in temporal arteritis. Mayo Clin Proc 1976;51:504–10.

3. Gonzalez­Gay MA, Garcia­Porrua C, Llorca J, Gonzalez­Louzao C, Rodriguez­Ledo P. Biopsy­ negative giant cell arteritis: clinical spectrum and predictive factors for positive temporal artery biopsy. Semin Arthritis Rheum 2001;30:249–56.

4. Allsop CJ, Gallagher PJ. Temporal artery biopsy in giant­ cell arteritis: a reappraisal. Am J Surg Pathol 1981;5:317–23.

5. Luqmani R, Lee E, Singh S, Gillett M, Schmidt WA, Bradburn M, et al. The role of ultrasound compared to biopsy of temporal arteries in the diagnosis and treatment of giant cell arteritis (TABUL): a diagnostic accuracy and cost­ effectiveness study. Health Technol Assess 2016;20:1–238.

6. Muratore F, Kermani TA, Crowson CS, Green AB, Salvarani C, Matteson EL, et al. Large­ vessel giant cell arteritis: a cohort study. Rheumatology (Oxford) 2015;54:463–70.

7. Prieto­González S, Depetris M, García­Martínez A, Espígol­Frigolé G, Tavera­Bahillo I, Corbera­Bellata M, et al. Positron emission tomography assessment of large vessel inflammation in patients with newly diagnosed, biopsy­ proven giant cell arteritis: a prospective, case­ control study. Ann Rheum Dis 2014;73:1388–92.

8. Blockmans D, Stroobants S, Maes A, Mortelmans L. Positron emission tomography in giant cell arteritis and polymyalgia rheumatica: evidence for inflammation of the aortic arch. Am J Med 2000;108:246–9.

9. Soussan M, Nicolas P, Schramm C, Katsahian S, Pop G, Fain O, et al. Management of large­ vessel vasculitis with FDG­ PET: a systematic literature review and meta­ analysis. Medicine (Baltimore) 2015;94:e622.

10. Besson FL, Parienti JJ, Bienvenu B, Prior JO, Costo S, Bouvard G, et al. Diagnostic performance of 18F­ fluorodeoxyglucose positron emission tomography in giant cell arteritis: a systematic review and meta­ analysis. Eur J Nucl Med Mol Imaging 2011;38:1764–72.

11. Nielsen BD, Gormsen LC, Hansen IT, Keller KK, Therkildsen P, Hauge EM. Three days of high­ dose glucocorticoid treatment attenuates large­ vessel 18F­ FDG uptake in large­ vessel giant cell arteritis but with a limited impact on diagnostic accuracy. Eur J Nucl Med Mol Imaging 2018;45:1119–28.

12. Sammel AM, Hsiao E, Nguyen K, Schembri G, Laurent R. Maxillary artery 18F­ FDG uptake as a new finding on PET/CT scan in a cohort of 41 patients suspected of having giant cell arteritis. Int J Rheum Dis 2018;21:560–2.

13. Sammel AM, Hsiao E, Schrieber L, Janssen B, Youssef P, Fraser CL, et al. Fluorine­ 18 fluoro­ 2­ deoxyglucose positron emission tomography uptake in the superficial temporal and vertebral arteries in biopsy positive giant cell arteritis. J Clin Rheumatol 2017;23:443.

14. Nielsen BD, Hansen IT, Kramer S, Haraldsen A, Hjorthaug K, Bogsrud TV, et al. Simple dichotomous assessment of cranial artery inflammation by conventional 18F­ FDG PET/CT shows high accuracy for the diagnosis of giant cell arteritis: a case­ control study. Eur J Nucl Med Mol Imaging 2019;46:184–93.

SAMMEL ET AL 1328       |

15. Hunder GG, Bloch DA, Michel BA, Stevens MB, Arend WP, Calabrese LH, et al. The American College of Rheumatology 1990 criteria for the classification of giant cell arteritis. Arthritis Rheum 1990;33:1122–8.

16. Besson FL, de Boysson H, Parienti JJ, Bouvard G, Bienvenu B, Agostini D. Towards an optimal semiquantitative approach in giant cell arteritis: an (18)F­ FDG PET/CT case­ control study. Eur J Nucl Med Mol Imaging 2014;41:155–66.

17. Blockmans D, de Ceuninck L, Vanderschueren S, Knockaert D, Mortelmans L, Bobbaers H. Repetitive 18F­ fluorodeoxyglucose positron emission tomography in giant cell arteritis: a prospective study of 35 patients. Arthritis Rheum 2006;55:131–7.

18. Rhéaume M, Rebello R, Pagnoux C, Carette S, Clements­Baker M, Cohen­Hallaleh V, et al. High­ resolution magnetic resonance imaging of scalp arteries for the diagnosis of giant cell arteritis: results of a prospective cohort study. Arthritis Rheum 2017;69:161–8.

19. Cavazza A, Muratore F, Boiardi L, Restuccia G, Pipitone N, Pazzola G, et al. Inflamed temporal artery: histologic findings in 354 biopsies, with clinical correlations. Am J Surg Pathol 2014;38:1360–70.

20. Buderer NM. Statistical methodology: I. Incorporating the prevalence of disease into the sample size calculation for sensitivity and specificity. Acad Emerg Med 1996;3:895–900.

21. Ponge T, Barrier JH, Grolleau JY, Ponge A, Vlasak AM, Cottin S. The efficacy of selective unilateral temporal artery biopsy versus bilateral biopsies for diagnosis of giant cell arteritis. J Rheumatol 1988;15:997–1000.

22. Zenone T, Puget M. Sensitivity of clinically abnormal temporal artery in giant cell arteritis. Int J Rheum Dis 2013;16:771–3.

23. Dasgupta B, Borg FA, Hassan N, Alexander L, Barraclough K, Bourke B, et al. BSR and BHPR guidelines for the management of giant cell arteritis. Rheumatology (Oxford) 2010;49:1594–7.

24. Dejaco C, Ramiro S, Duftner C, Besson FL, Bley TA, Blockmans D, et al. EULAR recommendations for the use of imaging in large vessel vasculitis in clinical practice. Ann Rheum Dis 2018;77:636–43.

25. Schmidt WA, Kraft HE, Vorpahl K, Volker L, Gromnica­Ihle EJ. Color duplex ultrasonography in the diagnosis of temporal arteritis. N Engl J Med 1997;337:1336–42.

26. Ball EL, Walsh SR, Tang TY, Gohil R, Clarke JM. Role of ultrasonography in the diagnosis of temporal arteritis. Br J Surg 2010;97:1765–71.

27. Blockmans D, Coudyzer W, Vanderschueren S, Stroobants S, Loeckx D, Heye S, et al. Relationship between fluorodeoxyglucose uptake in the large vessels and late aortic diameter in giant cell arteritis. Rheumatology (Oxford) 2008;47:1179–84.

28. Grayson PC, Alehashemi S, Bagheri AA, Civelek AC, Cupps TR, Kaplan MJ, et al. 18F­ fluorodeoxyglucose–positron emission tomography as an imaging biomarker in a prospective, longitudinal cohort of patients with large vessel vasculitis. Arthritis Rheumatol 2018;70:439–49.

29. Klink T, Geiger J, Both M, Ness T, Heinzelmann S, Reinhard M, et al. Giant cell arteritis: diagnostic accuracy of MR imaging of superficial cranial arteries in initial diagnosis­ results from a multicenter trial. Radiology 2014;273:844–52.

1329

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1329–1338DOI 10.1002/art.40876 © 2019 The Authors. Arthritis & Rheumatology published by Wiley Periodicals, Inc. on behalf of American College of Rheumatology. This is an open access article under the terms of the Creative Commons Attribution- NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

Glucocorticoid Dosages and Acute- Phase Reactant Levels at Giant Cell Arteritis Flare in a Randomized Trial of TocilizumabJohn H. Stone,1 Katie Tuckwell,2 Sophie Dimonaco,3 Micki Klearman,2 Martin Aringer,4 Daniel Blockmans,5 Elisabeth Brouwer,6 Maria C. Cid,7 Bhaskar Dasgupta,8 Juergen Rech,9 Carlo Salvarani,10 Hendrik Schulze-Koops,11 Georg Schett,9 Robert Spiera,12 Sebastian H. Unizony,1 and Neil Collinson3

Objective. This study was undertaken to evaluate glucocorticoid dosages and serologic findings in patients with giant cell arteritis (GCA) flares.

Methods. Patients with GCA were randomly assigned to receive double- blind dosing with either subcutane-ous tocilizumab (TCZ) 162 mg weekly plus 26- week prednisone taper (TCZ- QW + Pred- 26), every- other- week TCZ plus 26- week prednisone taper (TCZ- Q2W + Pred- 26), placebo plus 26- week prednisone taper (PBO + Pred- 26), or placebo plus 52- week prednisone taper (PBO + Pred- 52). Outcome measures were prednisone dosage, C- reactive protein (CRP) level, and erythrocyte sedimentation rate (ESR) at the time of flare.

Results. One hundred patients received TCZ- QW + Pred- 26, 49 received TCZ- Q2W + Pred- 26, 50 received PBO + Pred- 26, and 51 received PBO + Pred- 52. Of the 149 TCZ- treated patients, 36 (24%) experienced flare, 23 (64%) of whom were still receiving prednisone (median dosage 2.0 mg/day). Among 101 PBO + Pred–treated patients, 59 (58%) experienced flare, 45 (76%) of whom were receiving prednisone (median dosage 5.0 mg/day). Many flares occurred while patients were taking >10 mg/day prednisone: 9 (25%) in the TCZ groups and 13 (22%) in the placebo groups. Thirty- three flares (92%) in TCZ- treated groups and 20 (34%) in PBO + Pred–treated groups occurred with normal CRP levels. More than half of the PBO + Pred–treated patients had elevated CRP levels without flares. Bene-fits of the TCZ and prednisone combination over prednisone alone for remission induction were apparent by 8 weeks.

Conclusion. Most GCA flares occurred while patients were still receiving prednisone. Acute- phase reactant levels were not reliable indicators of flare in patients treated with TCZ plus prednisone or with prednisone alone. The addi-tion of TCZ to prednisone facilitates earlier GCA control.

INTRODUCTION

Giant cell arteritis (GCA) is a vasculitis of large- and medium- sized arteries that affects people ≥50 years old (1). Upon being diagnosed as having GCA, patients are treated immediately with high dosages of glucocorticoids to reduce the risk of vision loss

and large vessel complications. Long- term glucocorticoid treat-ment has traditionally been required to control symptoms and pre-vent relapse in GCA patients (2), but flares frequently occur (3–5). Although GCA is the most common primary form of systemic vas-culitis in Western countries, there are few data from randomized clinical trials regarding prednisone dosages at disease flares,

ClinicalTrials.gov identifier: NCT01791153.Supported by F. Hoffmann-La Roche Limited.1John H. Stone, MD, MPH, Sebastian H. Unizony, MD: Massachusetts

General Hospital and Harvard Medical School, Boston, Massachusetts; 2Katie Tuckwell, PhD, Micki Klearman, MD, PhD: Genentech, San Francisco, California; 3Sophie Dimonaco, MSc, Neil Collinson, PhD: Roche Products Ltd., Welwyn Garden City, UK; 4Martin Aringer, MD: University Medical Center and Technische Universität Dresden, Dresden, Germany; 5Daniel Blockmans, MD: University Hospitals Gasthuisberg, Leuven, Belgium; 6Elisabeth Brouwer, MD, PhD: University of Groningen and University Medical Center Groningen, Groningen, The Netherlands; 7Maria C. Cid, MD: University Hospital Clínic de Barcelona and University of Barcelona, Barcelona, Spain; 8Bhaskar Dasgupta, MBBS, MD, FRCP: Southend University Hospital, NHS Foundation Trust, Southend, UK; 9Juergen Rech, MD, George Schett, MD: Universitätsklinikum Erlangen, Erlangen, Germany; 10Carlo Salvarani, MD: Azienda USL-IRCCS di Reggio Emilia and Università di Modena and Reggio

Emilia, Reggio Emilia, Italy; 11Hendrik Schulze-Koops, MD, PhD: University of Munich, Munich, Germany; 12Robert Spiera, MD: Hospital for Special Surgery, New York, New York.

Dr. Stone has received consulting fees from Chugai (less than $10,000) and from Roche/Genentech (more than $10,000) and has received research support from Chugai. Dr. Tuckwell owns stock or stock options in Roche/Genentech. Ms Dimonaco owns stock or stock options in Roche. Dr. Klearman has received consulting fees from Roche/Genentech (less than $10,000) and owns stock or stock options in Roche/Genentech. Dr. Aringer has received consulting fees, speaking fees, and/or honoraria from Roche and Chugai (less than $10,000 each). Dr. Blockmans has received consulting fees from Roche (less than $10,000). Dr. Brouwer has received consulting fees, speaking fees, and/or honoraria from Roche (less than $10,000) on behalf of the University Medical Center Groningen, The Netherlands. Dr. Cid has received consulting fees, speaking fees, and/or honoraria from Roche, Novartis, Boehringer-Ingelheim, GlaxoSmithKline, Vifor, and AbbVie (less

STONE ET AL 1330       |

particularly for patients treated with prednisone for 1 year—a course that approximates the standard of care for many clinicians. Additionally, the usefulness of acute- phase reactants (APRs) in the clinical assessment of GCA flares has been poorly studied in patients treated with prednisone alone or with tocilizumab (TCZ). Moreover, no randomized clinical trials have been conducted in which clinicians and patients were blinded with regard to pred-nisone dosages and APR levels. TCZ, a humanized monoclonal antibody against the interleukin- 6 (IL- 6) receptor α, inhibits IL- 6–mediated signaling and inflammatory pathways (6,7).

In the Giant Cell Arteritis Clinical Research Study (GiACTA), a randomized, double- blind, placebo (PBO)–controlled phase III study of patients with GCA, TCZ was superior to PBO in the achievement of sustained remission at 1 year (8). TCZ was approved for the treatment of patients with GCA in 2017. Blocking IL- 6 signaling with TCZ reduces levels of APRs such as C- reactive protein (CRP) and decreases the erythrocyte sedimentation rate (ESR) (7). Consequently, measuring APR levels to quantify sys-temic inflammation is believed to have limited value in the clinical assessment of disease flares in patients with GCA treated with TCZ (9). GiACTA was the first randomized clinical trial in any dis-ease (to our knowledge) to include a blinded, variable- dosage prednisone taper. Once patients reduced their daily prednisone dosage, according to protocol, to <20 mg/day, patients and physician- investigators were blinded with regard to glucocorticoid dosages unless a flare occurred. Disease flares were assessed largely on a clinical basis, irrespective of APR levels, because investigators were blinded with regard to CRP levels, and initially only the laboratory assessor was aware of ESR results.

The design of the GiACTA trial permits a unique opportunity to study prednisone dosages and laboratory features associated with disease flares in GCA patients treated with prednisone alone and those treated with TCZ plus prednisone. These trial data therefore provide guidance for clinical decision- making within the new and traditional treatment landscapes for GCA.

PATIENTS AND METHODS

Ethics board approval and informed consent. This trial was approved by institutional review boards and/or ethics

committees at the appropriate institutions and was conducted in accordance with the Guidelines for Good Clinical Practice and the Declaration of Helsinki. All patients provided written informed consent.

Patients and study design. The patient eligibility criteria and study design for the GiACTA trial (ClinicalTrials.gov identifier: NCT01791153) have previously been published (10). Patients were randomly assigned 2:1:1:1 to 4 groups to receive treatment with weekly subcutaneous TCZ 162 mg plus a 26- week prednisone taper (TCZ- QW + Pred- 26), every- other- week subcutaneous TCZ 162 mg plus a 26- week prednisone taper (TCZ- Q2W + Pred- 26), subcutaneous placebo plus a 26- week prednisone taper (PBO + Pred- 26), or subcutaneous placebo plus a 52- week prednisone taper (PBO + Pred- 52) (8). Randomization was stratified by base-line prednisone dosage (≤30 mg/day or >30 mg/day). During the study, prednisone was tapered in a double- blind, protocol- defined manner (11). Prednisone was initially administered on an open- label basis at dosages of >20 mg/day and tapered according to weekly, protocol- defined decrements. Patients and investigators were blinded with regard to prednisone dosages of ≤20 mg/day.

To maintain rigorous blinding of investigators given the antic-ipated normalization of APR levels with IL- 6 receptor blockade, a separate laboratory assessor was assigned to monitor laboratory parameters (including ESR) independently of the blinded inves-tigator/efficacy assessor, who assessed the patient’s GCA and managed the prednisone taper. All investigators were blinded with regard to CRP level. The laboratory assessor was informed of ESR values, but the efficacy assessor was informed only whether the ESR was higher or lower than 30 mm/hour under strict protocol guidelines to ensure patient safety while preserving the blind (8).

Definition of GCA flares. GCA flare was determined by the investigator and defined as the recurrence of signs or symptoms of GCA or an ESR of ≥30 mm/hour attributed by the investigator to GCA even in the absence of any other overt clinical manifestations of active disease. The definition of disease flare included the need for an increase in the prednisone dosage at the time of the clinical event. The symptoms and signs of flare were categorized as fol-lows: GCA signs and symptoms (new- onset localized headache,

than $10,000 each) and research support from Kiniksa. Dr. Dasgupta has received consulting fees from Roche, GlaxoSmithKline, and Sanofi Aventis (less than $10,000 each). Dr. Rech has received consulting fees, speaking fees, and/or honoraria from Bristol-Myers Squibb, Celgene, Chugai, GlaxoSmithKline, Janssen, Eli Lilly, Novartis, Roche, Sanofi Aventis, and UCB (more than $10,000 each) and research support from Bristol-Myers Squibb and Celgene. Dr. Salvarani has received consulting fees from Roche (less than $10,000) and research support from Roche. Dr. Schulze-Koops has received consulting fees, speaking fees, and/or honoraria from Roche and Chugai (more than $10,000 each). Dr. Schett has received consulting fees from AbbVie, Bristol Myers-Squibb, Celgene, Chugai, GlaxoSmithKline, Eli Lilly, Novartis, Roche, Sanofi Aventis, and UCB (less than $10,000 each). Dr. Spiera has received consulting fees from Roche/Genentech, GlaxoSmithKline, CSL Behring, and Sanofi Aventis (less than $10,000 each) and research support from Roche/Genentech, GlaxoSmithKline, Bristol- Myers Squibb, Boehringer Ingelheim, Cytori, Chemocentryx, and Corbus. Dr. Unizony has received

research support from Roche/Genentech. Dr. Collinson owns stock or stock options in Roche. No other disclosures relevant to this article were reported.

Qualified researchers may request access to data through the clinical study data request platform (www.clini calst udyda tareq uest.com). Further details on Roche’s criteria for eligible studies are available (https ://clini calst udyda tareq uest.com/Study-Spons ors/Study-Spons ors-Roche.aspx). For further details on Roche’s Global Policy on the Sharing of Clinical Information and how to request access to related clinical study documents, see https ://www.roche.com/resea rch_and_devel opmen t/who_we_are_how_we_work/clini cal_trial s/our_commi tment_to_data_shari ng.htm.

Address correspondence to John H. Stone, MD, MPH, Harvard Medical School, Massachusetts General Hospital, Boston, MA 02114. E-mail: jhstone@mgh.harvard.edu.

Submitted for publication November 8, 2018; accepted in revised form February 28, 2019.

CHARACTERISTICS OF GIANT CELL ARTERITIS FLARE |      1331

scalp tenderness, temporal artery tenderness or decreased pul-sation, ischemia- related vision loss, and otherwise unexplained mouth or jaw pain on mastication) only, polymyalgia rheumatica (PMR) symptoms only, fever (≥38°C) only, visual symptoms (unilat-eral and bilateral blindness, ischemic optic neuropathy, amaurosis fugax, blurred vision, and diplopia) only, elevated ESR (≥30 mm/hour) attributed to active GCA only, and multiple symptoms (e.g., the occurrence of GCA and PMR symptoms together).

Study analysis. Assessments were performed at each study visit to determine whether a patient’s disease was fully con-trolled and the patient could safely continue the prednisone taper. Remission or flare was determined by the investigator at each study visit. CRP normalization was not included in the definition of remission for this analysis, and, as noted, the investigators were blinded with regard to CRP level. Prednisone dosages and APR values at disease flare were evaluated at the time of first flare. CRP levels and ESR values were examined before flare, at the time of flare, and in the absence of clinical signs or symptoms of flare. If the ESR at the time of flare was unavailable, the last ESR value preceding the flare was used. Remission before baseline was determined by the investigator at each site. Many patients had achieved disease control before the baseline visit because of glucocorticoid use during the 6- week screening period. The proportions of patients achieving sustained remission, time to first flare, and cumulative prednisone duration and exposure were analyzed according to the patients’ baseline prednisone dosages (≤30 mg/day or >30 mg/day).

Statistical analysis. TCZ and PBO treatment groups were compared using the Cochran- Mantel- Haenszel test with adjust-ment for the stratification factor of starting prednisone dosage (≤30 mg/day or >30 mg/day) for analysis of the proportions of patients in sustained remission and using a Cox proportional haz-ards model with adjustment for the starting prednisone dosage for time to first flare. The secondary end point of remission rates over time and subgroup analyses by starting prednisone dosage and disease onset were prespecified. All other analyses were per-formed post hoc and are exploratory, with no adjustment for Type I error control; therefore, we performed a limited number of statisti-cal comparisons for these analyses. Hazard ratios (HRs) and 99% confidence intervals (99% CIs) were calculated.

RESULTS

Baseline characteristics of patients in the GiACTA trial have been reported (8). Briefly, 251 patients were randomly assigned to receive TCZ- QW + Pred- 26 (n = 100), TCZ- Q2W + Pred- 26 (n = 50), PBO + Pred- 26 (n = 50), or PBO + Pred- 52 (n = 51). The intent- to- treat and safety populations included 250 patients, because 1 patient assigned to the TCZ- Q2W + Pred- 26 group did not receive the study drug.

Flares. Among the 250 patients included in this analy-sis, 95 (38%) experienced disease flares following a period of remission during the first 52 weeks of the study. Of these 95 flares, 13 (13.7%) were characterized by symptoms consid-ered typical of GCA only and 13 (13.7%) by PMR symptoms only. No flares following remission were characterized by visual symptoms alone, and none were based on the presence of fever alone. Clinical features at disease flare were not char-acterized in additional detail. Nine flares (9.5%) were based only on increased ESR attributed to GCA in the absence of an alternative explanation. Only 1 patient in the TCZ- QW + Pred- 26 group was reported to have a flare after remission without symptoms of flare or an ESR of ≥30 mm/hour (Table  1). All

flares responded to increased glucocorticoid dosages.

Prednisone dosages at flare. The median prednisone dosage at the time of disease flare was 2.0 mg/day (range 0.0–25.0) for the combined TCZ groups and 5.0 mg/day (0.0–30.0) for the combined PBO groups. Of the 149 patients in the combined TCZ groups, 36 (24%) experienced GCA flares (Table 1); 23 of these 36 flares (63.9%) occurred while patients were still receiving prednisone (8 [22.2%] at a dosage of 1–5 mg/day), and 13 (36.1%) occurred after discontinuation of the prednisone taper. Of the 101 patients in the combined PBO groups, 59 (58%) experienced GCA flares; 45 of these (76.3%) occurred while patients were still receiving prednisone, and 14 (23.7%) occurred after discontinuation of the prednisone taper. Of the 50 patients who received PBO + Pred- 26, 9 of the 34 reported flares (26.4%) occurred at prednisone dosages of 1–5 mg/day. In contrast, two- thirds of the flares (17 of 25; 68%) in the 51 patients in the PBO + Pred- 52 group occurred at prednisone dosages of >5–20 mg/day. Substantial numbers of disease flares occurred while patients were receiving >10 mg/day prednisone, accounting for 9 (25%) of the flares in the TCZ groups and 13 (22.0%) in the PBO groups. No patients experienced flares while receiving prednisone dosages of >30 mg/day, and only 3 patients experienced flares while receiving dosages of >20–30 mg/day.

Among patients randomly assigned to receive PBO + Pred, 13 (38.2%) of the disease flares in the PBO + Pred- 26 group became manifest only after patients had tapered to a prednisone dosage of 0 mg/day (Table 1). In contrast, only 1 (4.0%) of the disease flares in the PBO + Pred- 52 group was diagnosed in a patient receiving a prednisone dosage of 0 mg/day; the rest of the flares in this group became clinically manifest before the patients tapered to 0 mg/day prednisone.

Relationships between APR level, disease activity, and flare. Flare with normal APR levels. Median CRP and ESR levels preceding GCA flare were lower in the TCZ groups than in the PBO groups (Table 1). This was expected based on the biology of IL- 6 receptor blockade and its downstream effects on

STONE ET AL 1332       |

Table 1. Prednisone dosages and acute- phase reactant levels at time of GCA flare*

Assessment at time of GCA flare

PBO + Pred- 26 (n = 50)

PBO + Pred- 52 (n = 51)

TCZ- QW + Pred- 26 (n = 100)

TCZ- Q2W + Pred- 26 (n = 49)

Flare experienced after remission 34 (68.0) 25 (49.0) 23 (23.0) 13 (26.5)Signs and symptoms experienced

at time of flare†GCA signs and symptoms only‡ 2 (5.9) 1 (4.0) 5 (21.7) 5 (38.5)PMR symptoms only 2 (5.9) 2 (8.0) 6 (26.1) 3 (23.1)Fever (≥38°C) only 0 0 0 0Visual symptoms only§ 0 0 0 0Elevated ESR (≥30 mm/hour) only 6 (17.6) 2 (8.0) 1 (4.3) 0Multiple symptoms 24 (70.6) 20 (80.0) 10 (43.5) 5 (38.5)No symptoms of flare 0 0 1 (4.3) 0

Receiving steroids at time of first flare†¶

21 (61.8) 24 (96) 17 (73.9) 6 (46.2)

Prednisone dosage at flare, median (range) mg/day

2.5 (0.0–30.0) 8.0 (0.0–20.0) 7.0 (0.0–25.0) 0.0 (0.0–12.5)

First flare experienced while receiving prednisone, by dosage in mg/day†

0 13 (38.2) 1 (4.0) 6 (26.1) 7 (53.8)1–5 9 (26.5) 7 (28.0) 5 (21.7) 3 (23.1)>5–10 5 (14.7) 11 (44.0) 4 (17.4) 2 (15.4)>10–20 6 (17.6) 6 (24.0) 6 (26.1) 1 (7.7)>20–30 1 (2.9) 0 2 (8.7) 0>30–40 0 0 0 0>50–60 0 0 0 0>60 0 0 0 0

CRP level preceding flare, median (range) mg/liter#

23.1 (1.4–119.0) 17.3 (0.2–122.0) 0.4 (0.2–93.2) 1.0 (0.2–18.1)

Presence of elevated CRP at time of flare†#

22 (65) 17 (68) 1 (4) 2 (15)

Presence of elevated CRP without flare#

26 (52.0) 31 (60.8) 5 (5.0) 3 (6.1)

ESR preceding flare, median (range) mm/hour**

51.0 (8.0–140.0) 39.0 (4.0–138.0) 5.0 (0.0–80.0) 5.0 (1.0–43.0)

Presence of elevated ESR at time of flare†**

27 (79) 14 (56) 1 (4) 3 (23)

Presence of elevated ESR without flare 31 (62.0) 28 (54.9) 2 (2.0) 3 (6.1)Presence of elevated CRP

and ESR without flare20 (40.0) 20 (39.2) 2 (2.0) 1 (2.0)

* Except where indicated otherwise, values are the number (%) of patients. GCA = giant cell arteritis; PBO + Pred- 26 = placebo plus 26- week prednisone taper; PBO + Pred- 52 = placebo plus 52- week prednisone taper; TCZ- QW + Pred- 26 = tocilizumab once weekly plus 26- week prednisone taper; TCZ- Q2W + Pred- 26 = tocilizumab once every 2 weeks plus 26- week prednisone taper; PMR = polymyalgia rheumatica; ESR = erythrocyte sedimentation rate; CRP = C-reactive protein. † Values are the number (%) of patients in the corresponding treatment group who experienced flare following remission. ‡ Includes new- onset localized headache, scalp tenderness, temporal artery tenderness or decreased pulsation, ischemia- related vision loss, and otherwise unexplained mouth or jaw pain on mastication. § Includes unilateral and bilateral blindness, ischemic optic neuropathy, amaurosis fugax, blurred vision, and diplopia. ¶ Includes prednisone and all other steroids. # Normal ≤10 mg/liter. ** Normal <30 mm/hour.

CHARACTERISTICS OF GIANT CELL ARTERITIS FLARE |      1333

Figure 1. Disease control during the first 52 weeks of treatment in all patients in the intent- to- treat population (A), and in patients with newly diagnosed (B) and relapsing (C) giant cell arteritis. Prespecified exploratory analysis of remission rates over time is shown. Remission was defined as absence of flare and did not include C- reactive protein level in the definition. Patients who withdrew from the study or received escape therapy were excluded from that point. Patients with missing information on remission status were considered not in remission for that time point only. Responders (patients in remission) were analyzed; therefore, values at week 52 are slightly higher than the highest values for sustained remission, which accounts for patients not adhering to the protocol- defined tapering regimen as nonresponders. PBO + Pred- 26 = placebo plus 26- week prednisone taper; PBO + Pred- 52 = placebo plus 52- week prednisone taper; TCZ- QW + Pred- 26 = tocilizumab once weekly plus 26- week prednisone taper; TCZ- Q2W + Pred- 26 = tocilizumab once every 2 weeks plus 26- week prednisone taper.

100

90

80

70

60

50

40

30

20

10

0

Patie

nts

in re

mis

sion

, %

C

Time, weeks12840 16 20 24 28 32 36 40 44 48 52

PBO+Pred-26 (n=27) PBO+Pred-52 (n=28) TCZ-QW+Pred-26 (n=53) TCZ-Q2W+Pred-26 (n=23)

100

90

80

70

60

50

40

30

20

10

0

Patie

nts

in re

mis

sion

, %

B

Time, weeks12840 16 20 24 28 32 36 40 44 48 52

PBO+Pred-26 (n=23) PBO+Pred-52 (n=23) TCZ-QW+Pred-26 (n=47) TCZ-Q2W+Pred-26 (n=26)

100

90

80

70

60

50

40

30

20

10

0

Patie

nts

in re

mis

sion

, %

A

Time, weeks12840 16 20 24 28 32 36 40 44 48 52

PBO+Pred-26 (n=50) PBO+Pred-52 (n=51) TCZ-QW+Pred-26 (n=100) TCZ-Q2W+Pred-26 (n=49)

STONE ET AL 1334       |

APR levels. The overwhelming majority of TCZ- treated patients had low APR levels during disease flare. In the TCZ groups, 33 of 36 first disease flares (91.7%) occurred with normal CRP levels (≤10 mg/liter) and 32 of 36 flares (88.9%) with normal ESR (<30 mm/hour). In the PBO groups, APR levels remained normal at the time of first disease flares for approximately one- third of patients; of the 59 first disease flares, 20 (33.9%) were associated with nor-mal CRP levels and 18 (30.5%) with normal ESRs.

Elevated APR levels without flare. More than half of the pa-tients in the PBO groups had CRP elevations without disease flare during 1 year of follow- up (Table 1). Fifty- seven of the 101 patients (56.4%) in the combined PBO groups had CRP level elevations (>10 mg/liter) during the 52- week follow- up period without disease flare (defined as CRP level or ESR elevated at 2 consecutive visits between weeks 12 and 52 in the absence of clinical manifestations—beyond laboratory markers—of a dis-ease flare). As anticipated, the CRP level was elevated in the ab-sence of flare in only a minority of TCZ- treated patients: 5 (5.0%) in TCZ- QW + Pred- 26 and 3 (6.1%) in TCZ- Q2W + Pred- 26. Few patients in the TCZ groups had elevations of ESR that were not associated with disease flare: 2 patients (2.0%) in the TCZ- QW + Pred- 26 group and 3 patients (6.1%) in the TCZ- Q2W + Pred- 26 group. In contrast, more than half the patients in the PBO groups had elevations of ESR that were not reported as disease flares. Elevated CRP level and ESR in the absence of flare was observed in 40 patients (39.6%) in the PBO groups and 3 patients (2.0%) in the TCZ groups (Table 1).

Remission. More than half of all patients (142 of 250; 56.8%) were in remission by the baseline visit because of glucocorticoids received during the 6- week screening period. Numerically higher proportions of patients in the TCZ groups than in the PBO groups achieved remission between baseline and week 12 (Figure  1A); this was observed for subgroups with newly diagnosed disease

(Figure  1B) and relapsing disease (Figure  1C). Proportions of patients achieving remission increased from baseline to week 12 by 28% in the TCZ- QW + Pred- 26 group and 23% in the TCZ- Q2W + Pred- 26 group, compared with 2% in the PBO + Pred- 26 group and 14% in the PBO + Pred- 52 group. At week 12, the proportions of patients in remission were 66.0% (n = 33) in the PBO + Pred- 26 group and 64.7% (n = 33) in the PBO + Pred- 52 group, compared with 83.0% (n = 83) in the TCZ- QW + Pred- 26 group (P = 0.10, ver-sus PBO + Pred- 26; P = 0.03, versus PBO + Pred- 52). The propor-tion of patients in remission in the TCZ- Q2W + Pred- 26 group was 81.6% (n = 40), which was not significantly different from the PBO + Pred- 26 group (P = 0.57) or the PBO + Pred- 52 group (P = 0.30).

Time to first flare according to starting prednisone dosage. Patients in the TCZ groups were more likely than those in the PBO groups to achieve sustained remission at each of the baseline prednisone dosages (Figure  2). Among patients who started at prednisone dosages of >30 mg/day, those in each of the TCZ groups experienced longer times to disease flares than those in either of the PBO + Pred groups (Figure  3A). Among patients who started at prednisone dosages of ≤30 mg/day, a divergence of the PBO + Pred- 26 group from the other 3 groups was evident from week 12, with a shorter time to flare in the PBO + Pred- 26 group than the other groups (Figure 3B). In patients who started at prednisone dosages of ≤30 mg/day, the risk for flare was significantly lower among TCZ- treated patients than among PBO + Pred- 26–treated patients (HR 0.21 [99% CI 0.08–0.54], P < 0.0001 for TCZ- QW + Pred- 26 and HR 0.28 [99% CI 0.09–0.86], P = 0.0035 for TCZ- Q2W + Pred- 26). Among patients who started at prednisone dosages of ≤30 mg/day, the risk for flare did not differ between either of the TCZ groups and the PBO + Pred- 52 group (HR 0.59 [99% CI 0.20–1.73], P = 0.2039 for TCZ- QW + Pred- 26 and HR 0.76 [99% CI 0.21–2.72], P = 0.5866 for TCZ- Q2W + Pred- 26).

Figure 2. Sustained remission through week 52 according to baseline prednisone dosage. Percentages are based on number of patients receiving baseline dosage (*) and total number of patients in each treatment group at baseline (†). See Figure 1 for definitions. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40876/abstract.

8

23

46 44

25

17

60

50

40

53 50

9

20

67

0

1114

85

70

25

0

33

17

4540

36

67

0

12

24

n

%†

13

26

26

26

9

18

4

8

6

12

5

5

6

12

5

10

19

19

8

16

11

22

5

10

6

6

1

2

9

10

7

14

20

20

10

20

4

8

7

14

6

12

6

12

11

11

5

10

11

11

9

18

4

8

100

90

80

70

60

50

40

30

20

10

0

Patients receiving baseline dose

Patie

nts

achi

evin

g su

stai

ned

rem

issi

on, %

*

Baseline prednisone dose, mg/day

PBO+Pred-26 (n=50)

PBO+Pred-52 (n=51)

TCZ-QW+Pred-26 (n=100)

TCZ-Q2W+Pred-26 (n=49)

20 25 30 35 40 50 60

CHARACTERISTICS OF GIANT CELL ARTERITIS FLARE |      1335

Methotrexate and sustained remission. Only 35 of 250 patients (14%) received concomitant methotrexate (MTX) therapy during the study: 18 in the PBO groups and 17 in the TCZ groups (Table 2). One of the 18 patients (5.6%) receiving MTX in the combined PBO + Pred groups achieved sustained remission, compared with 15 of 83 patients (18.1%) who were not receiving MTX. Seven of 17 patients (41.2%) in the combined TCZ groups achieved sustained remission while receiving MTX, compared with

75 of 132 patients (56.8%) who did not receive MTX.

DISCUSSION

The treatment of GCA has long been defined by glucocor-ticoid treatment. Our analyses of glucocorticoid dosages and APR levels at disease flare provide new information about the natural history of GCA patients treated according to protocol with up to 1 year of prednisone. Given the importance of gluco-

corticoids in the management of GCA for the past 70 years and the fact that prolonged courses of glucocorticoid treatment have been the only treatment clearly known to be effective, it may seem remarkable that such information was not available earlier. Results from other trials with shorter prednisone tapers (12–14) and longitudinal studies using data from observational cohorts (15–17) have implied that the failure rate of glucocorticoid ther-apy for GCA is high. However, GiACTA is the first randomized clinical trial to use a prednisone- tapering regimen in GCA for as long as 1 year, and the first trial in any disease to use a variable- dosage prednisone taper in which patients and investigators were blinded with regard to prednisone dosages. It therefore provides the most rigorous assessment to date of how well—or how poorly—glucocorticoids work in patients with GCA.

Equally important are the insights that these trial data provide into the contemporary management of GCA in the era of TCZ treatment. Three findings of major importance have resulted from

Figure 3. Kaplan- Meier plot of time to first giant cell arteritis (GCA) flare according to starting prednisone dosage of >30 mg/day (A) and ≤30 mg/day (B) in the intent- to- treat population. Patients never in remission were censored at day 1, and patients who withdrew from the study before week 52 were censored from the time of withdrawal. See Figure 1 for definitions. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40876/abstract.

A

PBO+Pred-26

PBO+Pred-52

TCZ-QW+Pred-26

TCZQ2W+Pred-26

100

90

80

70

60

50

40

30

20

10

0

Patients, n

Prop

ortio

n of

pat

ient

s w

ithou

t GC

A fl

are,

%

Time, weeks4 8 12 16 20 24 28 32 40 44 48 52360

21

24

45

25

20

22

43

23

18

19

40

20

18

17

40

20

16

17

40

20

12

15

38

19

10

13

36

17

9

12

35

17

8

9

34

17

7

7

34

16

7

5

33

16

2

–

3

2

8

11

35

17

23

25

48

25

PBO+Pred-26 (N=23)

PBO+Pred-52 (N=25)

TCZ-QW+Pred-26 (N=48)

TCZ-Q2W+Pred-26 (N=25)

Censored+

+

++++++++++

+++

+++ + +

+ +++++++++++

++

+ + +++++++

B

PBO+Pred-26

PBO+Pred-52

TCZ-QW+Pred-26

TCZQ2W+Pred-26

100

90

80

70

60

50

40

30

20

10

0

Patients, n

Prop

ortio

n of

pat

ient

s w

ithou

t GC

A fl

are,

%

Time, weeks4 8 12 16 20 24 28 32 40 44 48 52360

23

24

48

22

20

22

45

22

18

22

45

20

16

21

45

20

13

18

41

19

11

17

39

16

9

17

38

15

9

16

36

13

6

13

33

12

6

10

30

10

6

10

30

8

1

–

2

–

8

14

34

13

27

26

52

24

PBO+Pred-26 (N=27)

PBO+Pred-52 (N=26)

TCZ-QW+Pred-26 (N=52)

TCZ-Q2W+Pred-26 (N=24)

Censored+

+++++

++++++++++

++

++++ + +++

+ + +++++++++

+

+ + +++ ++++

STONE ET AL 1336       |

this analysis, all of which have the potential to impact current treat-ment strategies.

First, most disease flares observed in the GiACTA trial occurred while patients were still receiving prednisone and, in many cases, at dosages incompatible with long- term use. More than one- fifth of the 95 first disease flares that occurred in the GiACTA trial were observed in patients receiving >10 mg/day prednisone, and 72% occurred while patients were still receiving some dosage of prednisone. This finding is consistent with the fact that GCA patients have endured long courses of prednisone and experienced adverse effects of long- term glucocorticoid use as a nearly universal feature of their disease management, and with the concept of a lag time between loss of immunosuppressive control of GCA and clinical expression of disease flare. Although it was not formally addressed in this trial, MTX did not show benefits; the benefit of adjunctive MTX in GCA has previously been demon-strated in a meta- analysis of 3 trials (18), but not with an effect size comparable to that of TCZ.

In the PBO + Pred groups, the slower (52- week) taper was associated with disease flare at higher dosages of prednisone in patients in the PBO + Pred- 52 group than in patients in the PBO + Pred- 26 group, who were more likely to reach a prednisone dos-age of 0 mg/day before experiencing flare. These findings, which may appear counterintuitive, suggest that a lag exists between the reemergence of disease activity and the appearance of clinical manifestations. The data support the concept that subclinical dis-ease activity begins in many patients as they taper to lower daily prednisone dosages, but that they do not experience symptoms until they have discontinued prednisone completely if the pred-nisone taper is rapid (e.g., 26 weeks versus 52 weeks). Patients undergoing shorter prednisone tapers are therefore more likely

to discontinue their prednisone entirely before the disease recur-rence manifests as a clinical flare. In contrast, patients undergoing a slower prednisone taper may be more likely to experience clini-cal disease flare while still receiving prednisone.

Second, APR levels are of little value in monitoring longitudi-nal disease activity in patients treated with TCZ. CRP levels and ESRs remained low in nearly all TCZ- treated patients who experi-enced disease flares. This is not surprising considering the antic-ipated effects of IL- 6 inhibition on CRP concentrations and ESR. Our data also confirm that APR measurements have shortcom-ings and the potential to mislead clinicians with regard to patients treated with prednisone alone. Although this point has been made by others (16,19), it remains underappreciated by many clinicians who manage patients with GCA. Approximately one- third of all disease flares in the PBO groups occurred while patients had normal CRP levels and ESRs (34% and 31%, respectively). Fur-thermore, more than half of PBO- treated patients had elevations of either CRP level or ESR, and more than one- third had eleva-tions of both APR levels without subsequent clinical disease flares (56.4%, 58.4%, and 39.6%, respectively).

These findings underscore the importance of the clinical assessment—particularly the importance of clinical experience and astute history- taking—in gauging whether GCA activity might be present. They also highlight the need for advances in the devel-opment of clinically useful biomarkers and rigorous correlation between imaging study results and disease activity in longitudi-nal disease assessments. It is possible that APR level elevations observed in the PBO- Pred groups that were not followed by disease flares during 52 weeks of follow- up indicate low- grade disease activity that remained subclinical (imaging studies were not performed on these patients). A discord between GCA clin-

Table 2. Methotrexate (MTX) use and sustained remission*

PBO + Pred- 26 (n = 50)

PBO + Pred- 52 (n = 51)

TCZ- QW + Pred- 26 (n = 100)

TCZ- Q2W + Pred- 26 (n = 49)

Received concomitant MTX 8 (16) 10 (20) 11 (11) 6 (12)Did not receive concomitant

MTX42 (84) 41 (80) 89 (89) 43 (88)

MTX dosage, mean ± SD mg/week

17.0 ± 6.0 15.4 ± 4.2 13.7 ± 3.1 13.1 ± 5.0

Total MTX dose during 52 weeks, mean ± SD mg

663.1 ± 434.3 635.3 ± 414.7 577.1 ± 250.2 491.7 ± 376.4

Patients receiving MTX who achieved sustained remission†

0 (0) 1 (10) 4 (36) 3 (50)

Patients not receiving MTX who achieved sustained remission‡

7 (17) 8 (20) 52 (58) 23 (54)

* Except where indicated otherwise, values are the number (%) of patients in the corresponding treatment group. See Table 1 for definitions. † Percentage based on number of patients receiving MTX. ‡ Percentage based on number of patients not receiving MTX.

CHARACTERISTICS OF GIANT CELL ARTERITIS FLARE |      1337

ical symptoms and vascular changes detected by imaging has previously been described (20,21). The optimal imaging protocol in large vessel vasculitis in clinical practice is unclear, and the inter-pretation of large vessel imaging studies in GCA is frequently not straightforward (22–24).

Third, these analyses demonstrate how swiftly the use of TCZ exerts a beneficial effect in GCA. The primary outcome measure of the trial was maintenance of remission at 52 weeks. However, our findings demonstrate that the effects of TCZ on induction of remission are rapid, which was most obvious in patients who experienced relapse (Figure  1C). In addition, even after 12 weeks, benefits of the combination of TCZ and prednisone over prednisone alone were apparent for the whole patient population (Figure  1A). These findings provide impor-tant information about the timing of TCZ use. For clinicians to establish disease control as quickly as possible, avoid acute and chronic complications of poorly controlled disease, and prevent excessive side effects of glucocorticoid use, a treatment strategy that emphasizes initiation of TCZ treatment as early as possible, accompanied by the institution of aggressive prednisone tapers (perhaps even shorter than 6 months) may be appropriate. Longer- term follow- up of patients in the GiACTA trial will deter-mine whether such a strategy has a durable effect in a substan-tial number of patients, even after discontinuation of TCZ at 1 year. The appropriateness of prednisone tapers more aggressive than those used in GiACTA cannot be routinely recommended until carefully conducted longitudinal studies and clinical trials are performed.

The addition of TCZ to glucocorticoid therapy enables faster glucocorticoid tapering than is possible for patients treated with glucocorticoids alone. Nevertheless, treatment with TCZ did not prevent disease flare in all patients. Nearly one- quarter of patients (24%) randomly assigned to either of the TCZ groups experienced ≥1 disease flare while receiving treatment. One potential contributor to the risk of disease flare in a subset of TCZ- treated patients is that Th1 cells, which are not directly affected by IL- 6 receptor blockade, may play an important role in some patients with GCA (25,26). Other explanations for dis-ease recurrence despite ongoing IL- 6 receptor blockade must be examined in other studies. Nevertheless, the fact that GCA flares are possible in patients still taking TCZ underscores the argument that glucocorticoid tapering during GCA remission should continue to be carefully monitored to ensure patient safety and that clinicians must remain alert to the possibility of disease flare. Nevertheless, most patients treated with TCZ were able to discontinue glucocorticoids entirely, which is a highly desirable goal.

The data included in this analysis were from the largest randomized controlled trial of treatment for GCA conducted to date, which highlights the robustness of these findings. However, there are some limitations in these primarily post hoc exploratory analyses, including limited statistical testing.

Flares were determined by investigators blinded with regard to CRP and informed only whether the ESR was higher or lower than 30 mm/hour. Although clinical descriptions were consis-tent with typical symptoms of GCA flare, this approach in the setting of the randomized, double- blind, PBO- controlled trial might have led to a lower threshold for the diagnosis of some disease flares, out of concern for patient safety. The defini-tion of remission without CRP that was used for the current analysis was not predefined per protocol except as part of a sensitivity analysis, but it allows for evaluation of remission outside the effect of TCZ on APR levels. Another limitation is the fact that clinical investigators were informed of clinically rel-evant elevations in ESR in accordance with the dual- assessor approach previously described (8). However, only 9 patients had flares judged on the basis of ESR alone, and excluding these patients would only reduce the proportion of patients who experienced flare in the PBO + Pred- 26 group from 68% (34 of 50) to 64% (28 of 44). Finally, granular details of the clinical aspects of the disease flares were not collected, and future studies should aim to collect more detailed information.

In conclusion, these data provide important information about the efficacy of glucocorticoid treatment alone and about the occurrence of disease flares in the era of IL- 6 receptor blockade. The addition of TCZ to prednisone in the treatment of GCA allows more patients to achieve remission within the first few months of initiating treatment. This may be impor-tant to prevent short- and long- term complications and reduce complications of long- term glucocorticoid use. Disease flares occur commonly in GCA, even when patients are receiving substantial dosages of prednisone and in patients treated with TCZ. Finally, APR levels have substantial limitations as indi-cators of disease activity, both in patients treated with TCZ plus glucocorticoids and in those treated with glucocorticoids alone. Astute clinical assessment of patient symptoms and signs remains crucial in the longitudinal management of GCA. Greater emphasis on the identification of useful biomarkers and application of large vessel imaging studies to understand disease activity are important components of the research agenda for GCA.

ACKNOWLEDGMENTS

We thank the teams of trial investigators and subinvestiga-tors, the Roche trial team, and the patients who participated in the trial.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti-cally for important intellectual content, and all authors approved the final version to be published. Dr. Stone had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

STONE ET AL 1338       |

Study conception and design. Stone, Tuckwell, Dimonaco, Klearman, Aringer, Blockmans, Brouwer, Cid, Dasgupta, Rech, Salvarani, Schulze- Koops, Schett, Spiera, Unizony, Collinson.Acquisition of data. Stone, Tuckwell, Dimonaco, Klearman, Aringer, Blockmans, Brouwer, Cid, Dasgupta, Rech, Salvarani, Schulze- Koops, Schett, Spiera, Unizony, Collinson.Analysis and interpretation of data. Stone, Tuckwell, Dimonaco, Klear-man, Aringer, Blockmans, Brouwer, Cid, Dasgupta, Rech, Salvarani, Schulze- Koops, Schett, Spiera, Unizony, Collinson.

ROLE OF THE STUDY SPONSOR

The study was funded by F. Hoffmann- La Roche Limited, which was involved in the design, conduct, reporting of the study, and interpretation of the results. The first draft of the manuscript was written by Dr. Stone. Third- party medical writing assistance was provided by Sara Duggan, PhD, and was funded by F. Hoff-mann- La Roche Limited. Publication of this article was contingent upon approval by F. Hoffmann-La Roche Limited.

REFERENCES 1. Salvarani C, Pipitone N, Versari A, Hunder GG. Clinical features of

polymyalgia rheumatica and giant cell arteritis. Nat Rev Rheumatol 2012;8:509–21.

2. Gonzalez-Gay MA, Martinez-Dubois C, Agudo M, Pompei O, Blanco R, Llorca J. Giant cell arteritis: epidemiology, diagnosis, and man-agement. Curr Rheumatol Rep 2010;12:436–42.

3. Hoffman GS, Cid MC, Hellmann DB, Guillevin L, Stone JH, Schousboe J, et al. A multicenter, randomized, double- blind, placebo- controlled trial of adjuvant methotrexate treatment for giant cell arteritis. Arthritis Rheum 2002;46:1309–18.

4. Hoffman GS, Cid MC, Rendt-Zagar KE, Merkel PA, Weyand CM, Stone JH, et al. Infliximab for maintenance of glucocorticosteroid- induced remission of giant cell arteritis: a randomized trial. Ann Intern Med 2007;146:621–30.

5. Seror R, Baron G, Hachulla E, Debandt M, Larroche C, Puéchal X, et al. Adalimumab for steroid sparing in patients with giant- cell arteritis: results of a multicentre randomised controlled trial. Ann Rheum Dis 2014;73:2074–81.

6. Mihara M, Kasutani K, Okazaki M, Nakamura A, Kawai S, Sugimoto M, et al. Tocilizumab inhibits signal transduction mediated by both mIL- 6R and sIL- 6R, but not by the receptors of other members of IL- 6 cytokine family. Int Immunopharmacol 2005;5:1731–40.

7. Nishimoto N, Terao K, Mima T, Nakahara H, Takagi N, Kakehi T. Mechanisms and pathologic significances in increase in serum interleukin- 6 (IL- 6) and soluble IL- 6 receptor after administration of an anti- IL- 6 receptor antibody, tocilizumab, in patients with rheuma-toid arthritis and Castleman disease. Blood 2008;112:3959–64.

8. Stone JH, Tuckwell K, Dimonaco S, Klearman M, Aringer M, Blockmans D, et al. Trial of tocilizumab in giant- cell arteritis. N Engl J Med 2017;377:317–28.

9. Villiger PM, Adler S, Kuchen S, Wermelinger F, Dan D, Fiege V, et al. Tocilizumab for induction and maintenance of remission in giant cell arteritis: a phase 2, randomised, double- blind, placebo- controlled trial. Lancet 2016;387:1921–7.

10. Unizony SH, Dasgupta B, Fisheleva E, Rowell L, Schett G, Spiera R, et al. Design of the tocilizumab in giant cell arteritis trial. Int J Rheu-matol 2013;2013:912562.

11. Collinson N, Tuckwell K, Habeck F, Chapman M, Klearman M, Stone JH. Development and implementation of a double- blind

corticosteroid- tapering regimen for a clinical trial. Int J Rheumatol 2015;2015:589841.

12. Spiera RF, Mitnick HJ, Kupersmith M, Richmond M, Spiera H, Peterson MG, et al. A prospective, double- blind, randomized, pla-cebo controlled trial of methotrexate in the treatment of giant cell arteritis (GCA). Clin Exp Rheumatol 2001;19:495–501.

13. Jover JA, Hernández-García C, Morado IC, Vargas E, Bañares A, Fernández-Gutiérrez B. Combined treatment of giant- cell arteritis with methotrexate and prednisone: a randomized, double- blind, placebo- controlled trial. Ann Intern Med 2001;134:106–14.

14. Martínez-Taboada VM, Rodríguez-Valverde V, Carreño L, López-Lon-go J, Figueroa M, Belzunegui J, et al. A double- blind placebo con-trolled trial of etanercept in patients with giant cell arteritis and corticosteroid side effects. Ann Rheum Dis 2008;67:625–30.

15. Alba MA, García-Martínez A, Prieto-González S, Tavera-Bahillo I, Corbera-Bellalta M, Planas-Rigol E, et al. Relapses in patients with giant cell arteritis: prevalence, characteristics, and associated clinical findings in a longitudinally followed cohort of 106 patients. Medicine 2014;93:194–201.

16. Kermani TA, Warrington KJ, Cuthbertson D, Carette S, Hoffman GS, Khalidi NA, et al. Disease relapses among patients with giant cell arteritis: a prospective, longitudinal cohort study. J Rheumatol 2015;42:1213–7.

17. Restuccia G, Boiardi L, Cavazza A, Catanoso M, Macchioni P, Muratore F, et al. Flares in biopsy- proven giant cell arteritis in north-ern Italy: characteristics and predictors in a long- term follow- up study. Medicine 2016;95:e3524.

18. Mahr AD, Jover JA, Spiera RF, Hernández-García C, Fernández-Gutiérrez B, LaValley MP, et al. Adjunctive methotrexate for treatment of giant cell arteritis: an individual patient data meta- analysis. Arthritis Rheum 2007;56:2789–97.

19. Salvarani C, Cantini F, Boiardi L, Hunder GG. Laboratory investiga-tions useful in giant cell arteritis and Takayasu’s arteritis. Clin Exp Rheumatol 2003;21 Suppl 32:S23–8.

20. Maleszewski JJ, Younge BR, Fritzlen JT, Hunder GG, Goronzy JJ, Warrington KJ, et al. Clinical and pathological evolution of giant cell arteritis: a prospective study of follow- up temporal artery biopsies in 40 treated patients. Mod Pathol 2017;30:788–96.

21. De Boysson H, Aide N, Liozon E, Lambert M, Parienti JJ, Monteil J, et al. Repetitive 18F- FDG- PET/CT in patients with large- vessel giant- cell arteritis and controlled disease. Eur J Intern Med 2017;46:66–70.

22. Blockmans D, de Ceuninck L, Vanderschueren S, Knockaert D, Mortelmans L, Bobbaers H. Repetitive 18F- fluorodeoxyglucose pos-itron emission tomography in giant cell arteritis: a prospective study of 35 patients. Arthritis Rheum 2006;55:131–7.

23. Prieto-González S, García-Martínez A, Tavera-Bahillo I, Hernández-Rodríguez J, Gutiérrez-Chacoff J, Alba MA, et al. Effect of glucocorticoid treatment on computed tomography angiography detected large- vessel inflammation in giant- cell arteritis: a prospec-tive, longitudinal study. Medicine (Baltimore) 2015;94:e486.

24. Grayson PC, Alehashemi S, Bagheri AA, Civelek AC, Cupps TR, Kaplan MJ, et al. F- fluorodeoxyglucose–positron emission to-mography as an imaging biomarker in a prospective, longitudinal cohort of patients with large vessel vasculitis. Arthritis Rheumatol 2018;70:439–49.

25. Watanabe R, Hosgur E, Zhang H, Wen Z, Berry G, Goronzy JJ, et al. Pro- inflammatory and anti- inflammatory T cells in giant cell arteritis. Joint Bone Spine 2017;84:421–6.

26. Visvanathan S, Rahman MU, Hoffman GS, Xu S, García-Martínez A, Segarra M, et al. Tissue and serum markers of inflammation during the follow- up of patients with giant- cell arteritis: a prospective longi-tudinal study. Rheumatology (Oxford) 2011;50:2061–70.

1339

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1339–1349DOI 10.1002/art.40862 © 2019, American College of Rheumatology

Prevalence, Treatment, and Outcomes of Coexistent Pulmonary Hypertension and Interstitial Lung Disease in Systemic SclerosisAmber Young,1 Dharshan Vummidi,1 Scott Visovatti,1 Kate Homer,1 Holly Wilhalme,2 Eric S. White,1 Kevin Flaherty,1 Vallerie McLaughlin,1 and Dinesh Khanna3

Objective. Systemic sclerosis (SSc) is associated with interstitial lung disease (ILD) and pulmonary hypertension (PH). This study was undertaken to determine the prevalence, characteristics, treatment, and outcomes of PH in a cohort of patients with SSc- associated ILD.

Methods. Patients with SSc- associated ILD on high- resolution computed tomography (HRCT) were included in a prospective observational cohort. Patients were screened for PH based on a standardized screening algorithm and underwent right- sided heart catheterization (RHC) if indicated. PH classification was based on hemodynamic findings and the extent of ILD on HRCT. Summary statistics and survival using the Kaplan- Meier method were calculated.

Results. Of the 93 patients with SSc- associated ILD included in the study, 76% were women and 65.6% had diffuse cutaneous SSc. The mean age was 54.9 years, and the mean SSc disease duration was 8 years. Twenty- nine patients (31.2%) had RHC- proven PH; of those 29 patients, 24.1% had PAH, 55.2% had World Health Organization (WHO) Group III PH, 34.5% had WHO Group III PH with pulmonary vascular resistance >3.0 Wood units, 48.3% had a PH diagnosis within 7 years of SSc onset, 82.8% received therapy for ILD, and 82.8% received therapy for PAH. The survival rate 3 years after SSc- associated ILD diagnosis for all patients was 97%. The survival rate 3 years after PH diagnosis for those with SSc- associated ILD and PH was 91%.

Conclusion. In a large cohort of patients with SSc- associated ILD, a significant proportion of patients had coexisting PH, which often occurs early after SSc diagnosis. Most patients were treated with ILD and PAH therapies, and survival was good. Patients with SSc- associated ILD should be evaluated for coexisting PH.

INTRODUCTION

Systemic sclerosis (SSc) can be a devastating multiorgan system autoimmune disease. It can affect the skin, peripheral vas-culature, muscles, joints, tendons, kidneys, gastrointestinal tract, lungs, and heart through fibrosis, vascular damage, and immune dysregulation. Interstitial lung disease (ILD) and pulmonary arterial hypertension (PAH) are the leading causes of mortality in SSc (1,2).

Up to 90% of patients with SSc have ILD on high- resolution computed tomography (HRCT), and ~40% of patients with SSc have clinically significant ILD (3,4). Pulmonary hypertension (PH) is also common in SSc, and patients with SSc can have various types of PH. Three common types of PH in patients with SSc include World Health Organization (WHO) Group I PH (PAH), WHO Group II PH (PH due to left- sided heart disease), and WHO Group III PH (PH due to ILD). However, in observational cohorts of SSc, which

Dr. Young’s work was supported by grant T32-AR-007080-38 from the NIH. Dr. Flaherty’s work was supported by grant K24-HL-111316 from the National Heart, Lung, and Blood Institute, NIH. Dr. Khanna’s work was supported by grants K24-AR-063120 and R01-AR-07047 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH.

1Amber Young, MD, Dharshan Vummidi, MD, Scott Visovatti, MD, Kate Homer, MM, Eric S. White, MD, Kevin Flaherty, MD, Vallerie McLaughlin, MD: University of Michigan, Ann Arbor, 2Holly Wilhalme: University of California, Los Angeles; 3Dinesh Khanna: University of Michigan, Ann Arbor and CiviBioPharma, Chevy Chase, Maryland.

Dr. Vummidi has received consulting fees, speaking fees, and/or honoraria from Boehringer Ingelheim (less than $10,000) and royalties from Amirsys for authorship in ExpertDDx. Dr. White has received consulting fees, speaking fees, and/or honoraria from Boehringer Ingelheim (less than $10,000). Dr. McLaughlin has received consulting fees from Arena, Bayer, Medtronic, Merck,

St. Jude Medical, SteadyMed, and United Therapeutics Corporation (less than $10,000 each and from Actelion Pharmaceuticals (more than $10,000) and research support for the University of Michigan from Actelion Pharmaceuticals, Arena, Eiger, Sonovie, and Bayer. Dr. Khanna has received consulting fees, speaking fees, and/or honoraria from Actelion, AstraZeneca, Bristol-Myers Squibb, ChemomAb, GlaxoSmithKline, Medac, Sanofi-Aventis/Genzyme, and UCB Pharma (less than $10,000 each) and from Bayer, Boehringer Ingelheim, Corbus, Cytori, Eicos, EMD Serono, and Genentech/Roche (more than $10,000 each) and owns stock or stock options in Eicos Sciences, Inc. No other conflicts of interest relevant to this article were reported.

Address correspondence to Dinesh Khanna, MD, MSc, University of Michigan, Suite 7C27, 300 North Ingalls Street, SPC 5422, Ann Arbor, MI 48109. E-mail: khannad@med.umich.edu.

Submitted for publication September 30, 2017; accepted in revised form February 12, 2019.

YOUNG ET AL 1340       |

are enriched for patients at risk of or with early PH, the majority of patients are diagnosed as having PAH and a smaller proportion are classified as having WHO Group II PH or WHO Group III PH (5,6). For example, in the Pulmonary Hypertension Assessment and Recognition of Outcomes in Scleroderma (PHAROS) study, ~69% of the patients with PH were classified as having PAH, 10% of the patients were classified as having WHO Group II PH, and 21% of the patients were classified as having WHO Group III PH (5). In the DETECT study, which recruited patients at high risk of PH, ~60% of the patients with PH were classified as having PAH, 21% were classified as having WHO Group II PH, and 19% were classified as having WHO Group III PH (6).

Although previous studies have analyzed patients with con-comitant SSc- associated ILD and PH, to our knowledge there is a lack of in- depth review of the clinical characteristics and man-agement of PH in patients with established SSc- associated ILD. This is an essential clinical question, since worsening dyspnea in a patient with underlying ILD may represent progressive ILD, new- onset PH, or a combination of both. In addition, many ongoing clinical trials are specifically recruiting patients with SSc- associated ILD, particularly for the evaluation of new and existing pharmaco-logic therapies for the treatment of ILD. Hence, it is imperative to recognize the prevalence of PH with concomitant ILD in SSc and its impact on the clinical course, outcome measures, such as dyspnea and quality of life, and survival. This need prompted us to investigate the prevalence of PH in a well- characterized cohort of patients with SSc- associated ILD, and to explore the clinical char-acteristics, pharmacologic therapies, and outcomes of patients with PH in that cohort.

PATIENTS AND METHODS

Study design and patients. Patients evaluated in this study were participants in a prospective observational cohort study of SSc- associated ILD (Figure 1). Patients were recruited from the University of Michigan Scleroderma and Connective Tissue Disease (CTD)–ILD clinics starting January 8, 2014, and data were extracted on October 1, 2016. Patients who were at least 18 years of age, met the 2013 American College of Rheu-matology/European League Against Rheumatism classification criteria for SSc (7), had ILD on HRCT, and could provide informed consent were included in the study, which was approved by the University of Michigan Institutional Review Board. All reported experiments performed by the authors have been previously published and complied with all applicable ethical standards (including the Declaration of Helsinki and its amendments, insti-tutional/national research committee standards, and interna-tional/national/institutional guidelines).

All patients in this study had a baseline HRCT confirming the presence of ILD, defined as the presence of bilateral, subpleural, lower lobe predominant distribution of either 1) reticular and/or ground- glass opacity with or without traction bronchiectasis or

2) honeycombing with the absence of a pattern that is predomi-nantly nodular, cystic, peribronchovascular/central or upper lung predominant, mosaic attenuation, or consolidation. Pulmonary function tests (PFTs) were performed at baseline for each patient. Patients with a forced expiratory volume in 1 second/forced vital capacity (FVC) ratio of <0.7 were excluded to rule out patients with concomitant obstructive pulmonary disease. Demographic characteristics and additional clinical variables were obtained for each patient.

All patients in the University of Michigan Scleroderma and CTD- ILD clinics undergo screening for PH based on the 2013 recommendations for screening and detection of CTD- associated PAH by Khanna et al, which also include the DETECT algorithm (8). Although the DETECT algorithm has been vali-dated in patients with a disease duration of >3 years and diffus-ing capacity for carbon monoxide (DLco) <60%, we use it for all SSc patients with uncorrected DLco ≤80% (6). We performed a chart review for all patients in this SSc- associated ILD cohort to determine which patients fulfilled criteria indicating possible PH based on the 2013 recommendations and had undergone right- sided heart catheterization (RHC) during their SSc dis-ease course. Within this SSc- associated ILD cohort, the 2013 recommendations were prospectively applied to 81 patients during routine clinical care (Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40862/ abstract) and retrospectively applied to 12 patients who had undergone RHC prior to 2013 (Supplementary Table 2, available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40862/ abstract) (8).

A total of 53 patients in the cohort had undergone RHC. Patients who had PH on RHC, which was defined as a mean pul-monary artery pressure (PAP) of ≥25 mm Hg, are referred to herein as patients with “SSc- associated ILD and PH,” and patients who did not meet the 2013 recommendations or had a mean PAP of <25 mm Hg on RHC are referred to herein as patients with “SSc- associated ILD without PH.” For those with SSc- associated ILD and PH, the HRCT scan was reviewed by a chest radiolo-gist (DV) to determine ILD severity. ILD severity was based on Goh’s criteria, in which <20% extent of ILD on HRCT was consid-ered minimal and >20% extent of ILD on HRCT was considered moderate-to-severe (9,10). The HRCT scans chosen for review were the scans obtained closest to the time of diagnosis of RHC. (One patient’s HRCT images were not available for review, so a CT angiogram was reviewed, and another patient did not have any images available at our institution for review, so ILD type and severity were based on an external radiologist’s initial report.)

Classification of PH. Patients classified as having WHO Group I PH/PAH had a mean PAP of ≥25 mm Hg, pulmonary capillary wedge pressure (PCWP) ≤15 mm Hg, and <20% extent of ILD on HRCT (10–12). Patients classified as having WHO Group

COEXISTENT PH AND ILD IN SSc |      1341

II PH had a mean PAP of ≥25 mm Hg and PCWP >15 mm Hg. Patients classified as having WHO Group III PH had a mean PAP of ≥25 mm Hg, PCWP ≤15 mm Hg, and >20% extent of ILD on HRCT (10–12). Of the patients with WHO Group III PH, those with a mean PAP of ≥35 mm Hg were further classified as having severe PH based on the recommendations of Seeger et al (13) (Figure 1).

Statistical analysis. Descriptive statistics for the overall SSc- associated ILD cohort, patients with SSc- associated ILD without PH, and patients with SSc- associated ILD and PH were calculated for demographic and clinical characteristics using the mean ± SD for continuous variables and percentage for categor-ical variables. The differences between the group without PH and

the group with PH were compared using Student’s t- test for con-tinuous variables and the chi- square test for categorical variables.

The Kaplan- Meier method was used to evaluate time until patient death in the overall SSc- associated ILD cohort and in the subset of those with SSc- associated ILD and PH, or censored at October 1, 2016. The log rank test was used to determine if there was a statistically significant difference in survival time in patients with SSc- associated ILD and PH versus those with SSc- associated ILD without PH. Cox proportional hazards regression was used to determine if race (non- Hispanic white versus non-white) and age predicted survival in both subsets.

To evaluate whether there was a significant trend over time in FVC and DLco for the PFT findings obtained in the prospective observational cohort, a linear mixed- effects model

Figure 1. Study design and characterization of pulmonary hypertension (PH) in a cohort of patients with systemic sclerosis (SSc)–associated interstitial lung disease (ILD). * Of the 14 patients who did not undergo right- sided heart catheterization (RHC), 7 were referred to cardiology but did not undergo RHC since there was a low likelihood of PH based on evidence, 2 refused RHC, 1 was lost to follow- up, 2 had negative findings on RHC after data analysis, 1 had normailization of transthoracic echocardiography (TTE) findings, and 1 had stable lower diffusing capacity for carbon monoxide (DLco), normal N- terminal pro–brain natriuretic peptide levels, and no pulmonary arterial hypertension (PAH) findings on TTE. † Of the 3 patients without signs/symptoms for RHC who underwent RHC, 2 underwent RHC due to severe symptoms and had negative RHC findings, and 1 underwent RHC due to a decline in DLco and had World Health Organization (WHO) Group III PH; the data needed to calculate DETECT scores were not available for any of these patients. ‡ Of the 3 patients with WHO Group II PH with >20% extent of ILD, 1 also had combined postcapillary and precapillary PH according to the guidelines by Vachiery et al (16) (pulmonary capillary wedge pressure [PCWP] >15 mm Hg and diastolic pulmonary artery pressure [PAP] − PCWP ≥7 mm Hg) and had features of chronic thromboembolic disease based on pulmonary artery angiography. § Patients with severe PH had SSc- associated ILD with PH due to ILD with a mean PAP (mPAP) of ≥35 mm Hg on RHC according to the criteria by Seeger et al (13). CTD = connective tissue disease; PVR = pulmonary vascular resistance; WU = Wood units.

YOUNG ET AL 1342       |

Table 1. Baseline characteristics of the patients with SSc- associated ILD*

All patients (n = 93)

Patients with SSc- associated ILD without PH

(n = 64)

Patients with SSc- associated

ILD and PH (n = 29) P

Age at initial non- RP sign/symptom, years 46.9 ± 13.3 45.2 ± 13.2 50.9 ± 12.8 0.054Age at ILD diagnosis, years 51.6 ± 12.2 49.7 ± 11.8 55.8 ± 12.1 0.02Age at study enrollment, years 54.9 ± 11.5 52.9 ± 11.4 59.3 ± 10.6 0.01Sex, no. (%) women 71 (76.3) 47 (73.4) 24 (82.8) 0.33Race, no. (%) 0.02

White 79 (84.9) 57 (89.1) 22 (75.9)African American 8 (8.6) 2 (3.1) 6 (20.7)Asian/Asian American 3 (3.2) 3 (4.7) 0Native American/Alaska Native 1 (1.1) 0 1 (3.4)Other 2 (2.2) 2 (3.1) 0

Ethnicity, no. (%) 0.51Hispanic 9 (9.7) 7 (10.9) 2 (6.9)Non- Hispanic 82 (88.2) 55 (85.9) 27 (93.1)Other 2 (2.2) 2 (3.1) 0

SSc subtype, no. (%) 0.34dcSSc 61 (65.6) 44 (68.8) 17 (58.6)lcSSc 32 (34.4) 20 (31.3) 12 (41.4)

Disease durationTime from initial non- RP sign/symptom to study

enrollment, years7.9 ± 7.2 7.7 ± 7.6 8.5 ± 6.1 0.65

Time from initial non- RP sign/symptom to ILD diagnosis, years

4.7 ± 6.4 4.6 ± 6.9 4.9 ± 5.4 0.80

Time from ILD diagnosis to study enrollment, years

3.2 ± 3.6 3.2 ± 3.2 3.5 ± 4.3 0.65

ILD duration, years 4.7 ± 3.6 4.6 ± 3.2 4.8 ± 4.3 0.74MRSS (at enrollment) (n = 89) 9.4 ± 9.5 9.9 ± 9.8 8.4 ± 8.9 0.51Autoantibodies, no. (%)

ANA positivity (n = 86) 79 (91.9) 57 (95) 22 (84.6) 0.19ANA pattern (n = 76) 0.10

Nucleolar 11 (14.5) 9 (16.4) 2 (9.5)Centromere 6 (7.9) 2 (3.6) 4 (19)Other† 59 (77.6) 44 (80) 15 (71.4)Scl- 70 positivity (n = 85) 24 (28.2) 20 (33.3) 4 (16) 0.12RNA polymerase III positivity (n = 51) 11 (21.6) 9 (25.7) 2 (12.5) 0.47PM/Scl positivity (n = 35) 2 (5.7) 1 (4.2) 1 (9.1) 0.54

ILD (on HRCT), no. (%) 0.13Nonspecific interstitial pneumonia 84 (90.3) 60 (93.8) 24 (82.8)Usual interstitial pneumonia 9 (9.7) 4 (6.3) 5 (17.2)

PFTs (obtained closest to the ILD diagnosis date)FVC, % predicted (n = 93) 76.2 ± 15.7 77.2 ± 14.3 73.9 ± 18.5 0.35TLC, % predicted (n = 66) 83 ± 16.3 85.0 ± 14.8 78.7 ± 18.8 0.14DLco, % predicted (n = 85) 58.3 ± 20.3 63.4 ± 20.1 46.8 ± 15.6 <0.001FVC, % predicted/DLco, % predicted (n = 85) 1.5 ± 0.8 1.3 ± 0.5 1.9 ± 1.2 0.007

(Continued)

COEXISTENT PH AND ILD IN SSc |      1343

with a fixed effect for time (continuous, in months) and a ran-dom effect for the patient was used to predict change from baseline for both FVC and DLco. All analyses were performed using SAS software version 9.4. P values less than 0.05 were considered significant.

RESULTS

Baseline characteristics of the total cohort. A total of 93 patients with SSc- associated ILD were evaluated in this study. The mean ± SD time from ILD diagnosis to study enrollment was 3.2 ± 3.6 years. Most patients were white (84.9%), female (76.3%), non- Hispanic (88.2%), and had diffuse cutaneous SSc (dcSSc) (65.6%). The mean ± SD overall SSc disease duration from initial non- Raynaud’s phenomenon (RP) sign/symptom for the entire cohort was 7.9 ± 7.2 years, and the mean ± SD time to diagnosis of ILD after initial non- RP sign/symptom was 4.7 ± 6.4 years. The mean ± SD modified Rodnan skin thickness score was 9.4 ± 9.5. The most common ILD pattern on HRCT was nonspecific interstitial pneumonia (NSIP; 90.3%). PFTs at ILD diagnosis revealed an FVC % predicted of 76.2 ± 15.7, total lung capacity (TLC) % predicted of 83.0 ± 16.3 (n = 66), DLco % predicted of 58.3 ± 20.3 (n = 85) (4 patients had DLco assessment attempted without success, 2 patients were ill, and 2 patients did not have DLco ordered with PFTs at the time of ILD diagnosis), and a ratio of FVC % predicted to DLco % predicted of 1.5 ± 0.8 (n = 85) (Table 1).

The mean ± SD study duration for patients within this SSc- associated ILD cohort was 16.6 ± 4.3 months. At the time of data analysis, 12 patients were considered lost to fol-low- up since they had not followed up in clinic for >12 months,

2 patients had withdrawn consent, and 3 patients had died. There was a significant trend over time for both FVC and DLco after ILD diagnosis in all 93 patients with SSc- associated ILD in this cohort. Each year after ILD diagnosis, FVC was reduced by a mean ± SEM of 1.23 ± 0.14% and DLco was reduced by a mean ± SEM of 1.22 ± 0.18%. However, there was no signif-icant change in FVC or DLco from study enrollment for any of the 93 patients in the cohort.

Prevalence of SSc- associated ILD and PH and baseline characteristics of the patients. Twenty- nine patients (31.2%) in this SSc- associated ILD cohort had RHC- proven PH, referred to herein as the subgroup with “SSc- associated ILD and PH.” The mean ± SD time from initial non- RP signs/symptoms to PH diagnosis was 7.0 ± 5.5 years, and 75.9% of the patients with PH (22 of 29) had PH diagnosed prior to enrollment in the ILD cohort. Thirty- one percent of the patients with SSc- associated ILD and PH (9 of 29) were diag-nosed as having PH prior to the development of the 2013 rec-ommendations for screening and detection of CTD- associated PAH, 65.5% of the patients with SSc- associated ILD and PH (19 of 29) were at risk for PH according to the 2013 recommen-dations, and 1 patient with SSc- associated ILD and PH did not have signs/symptoms indicating a need for RHC based on the 2013 recommendations but underwent RHC due to progressive symptoms and declining DLco.

The remaining 64 patients with SSc- associated ILD are referred to herein as patients with “SSc- associated ILD without PH.” Twenty- four patients had negative findings on RHC. The remaining 40 patients did not have signs/symptoms to proceed for an evaluation for RHC based on the 2013 PAH screening rec-

All patients (n = 93)

Patients with SSc- associated ILD without PH

(n = 64)

Patients with SSc- associated

ILD and PH (n = 29) P

TTE‡RV function, no. (%) (n = 92) <0.001

Normal 84 (91.3) 62 (98.4) 22 (75.9)Abnormal 8 (8.7) 1 (1.6) 7 (24.1)

RV enlargement, no. (%) 0.02No 80 (86.0) 59 (92.2) 21 (72.4)Yes 13 (14.0) 5 (7.8) 8 (27.6)

RVSP, mm Hg (n = 63)§ 37.8 ± 19.6 30.9 ± 8.2 49 ± 26.7 <0.001

* Except where indicated otherwise, values are the mean ± SD. SSc = systemic sclerosis; ILD = interstitial lung disease; PH = pulmonary hypertension; RP = Raynaud’s phenomenon; dcSSc = diffuse cutaneous SSc; lcSSc = limited cutaneous SSc; MRSS = modified Rodnan skin thickness score; HRCT = high- resolution computed tomography; PFTs = pulmonary function tests; FVC = forced vital capacity; TLC = total lung capacity; DLco = diffusing capacity for carbon monoxide; RV = right ventricular; RVSP = right ventricular systolic pressure. † Antinuclear antibody (ANA) positivity indicated by any immunofluorescence pattern other than nucleolar or centromere. ‡ Transthoracic echocardiography (TTE) data were captured after enrollment in the cohort. § No tricuspid regurgitation jet was observed in 30 patients.

Table 1. (Cont’d)

YOUNG ET AL 1344       |

ommendations by Khanna et al (n = 26), were believed by a PH expert to have a low likelihood of PH based on available evidence (n = 7), had a pending RHC at the time of data analysis (n = 2), refused RHC (n = 2), were lost to cardiology follow- up (n = 1), had normalization of previously abnormal transthoracic echocardiog-raphy (TTE) findings (n = 1), or had stability of lower DLco in the setting of normal N- terminal pro–brain natriuretic peptide levels and normal TTE findings (n = 1) (Figure 1). When compared to patients with SSc- associated ILD without PH, patients with SSc- associated ILD and PH were older at ILD diagnosis (mean age 55.8 years versus 49.7 years; P = 0.02), were more likely to be Afri-can American (20.7% versus 3.1%, P = 0.02), had lower DLco % predicted (46.8% versus 63.4%; P < 0.001) and had a higher ratio of FVC % predicted to DLco % predicted (1.9 versus 1.3; P = 0.007) at ILD diagnosis (Table 1).

Analysis of the time to diagnosis of PH after the initial non- RP sign/symptom showed that 37.9%, 44.8%, and 48.3% of the patients had a diagnosis of PH within 3, 5, and 7 years, respec-tively. We also performed analyses to evaluate baseline character-istics, cardiopulmonary characteristics, treatments, and outcomes for those with PH diagnosed <7 years (n = 14) compared to ≥7 years (n = 15) after the initial non- RP sign/symptom, which was based on the inclusion of patients with a disease duration of <7 years from initial non- RP sign/symptom in recent SSc- associated ILD clinical trials (14,15). There were no significant differences between patients with early diagnosis of PH (<7 years from ini-tial non- RP symptom) and patients with late diagnosis of PH (≥7 years from initial non- RP symptom) except age at initial non- RP sign/symptom (mean ± SD 57.2 ± 12.6 versus 45 ± 10.2 years; P = 0.008), Scl- 70 positivity (0 versus 4 [33.3%]; P = 0.04) (n = 25); ILD pattern on HRCT (14 NSIP [100%] versus 10 NSIP [66.7%] and 5 usual interstitial pneumonia [33.3%]; P = 0.04), and cardiac output measured by thermodilution (mean ± SD 5.9 ± 1.8 versus 4.8 ± 1.1; P = 0.047) (data not shown).

Cardiopulmonary characteristics of the patients with SSc- associated ILD and PH. All CT scans assessed for ILD severity were obtained a median of 2.2 months (inter quartile range 0.3–6.3 months) after the diagnosis of PH based on diag-nostic RHC. Seven patients with SSc- associated ILD and PH (24.1%) were classified as having PAH. Six patients (20.7%) were classified as having WHO Group II PH; 3 of those patients had <20% extent of ILD on HRCT, and 3 patients had >20% extent of ILD on HRCT. One of the patients with WHO Group II PH with >20% extent of ILD on HRCT also had features of combined postcapillary and precapillary PH based on PCWP >15 mm Hg and diastolic PAP − PCWP ≥7 mm Hg according to the clas-sification by Vachiery et al (16). That patient also had features of chronic thromboembolic disease based on pulmonary artery angiogram (16). Sixteen patients (55.2%) were classified as hav-ing WHO Group III PH. Ten patients (34.5%) with WHO Group III PH had pulmonary vascular resistance (PVR) >3.0 Wood units,

and 4 of those patients (13.8%) were classified as having severe PH based on a mean PAP of ≥35 mm Hg (13) (Figure 1). Car-diopulmonary characteristics are summarized for patients with

SSc- associated ILD and PH in Table 2.We also applied a previously published definition of clinically

significant ILD, which classifies patients as having clinically signif-icant ILD if they have >30% disease extent on HRCT, or 10–30% disease extent on HRCT and an FVC of <70% (17,18). Of the 29 patients with PH in the present study, 15 (51.7%) had clinically significant ILD.

Treatment and outcomes of SSc- associated ILD and PH. Twenty- four patients (82.8%) with SSc- associated ILD and PH underwent ILD treatment during their SSc- associated ILD disease course. Eleven patients were treated only with mycophenolate mofetil monotherapy, 6 patients received mycophenolate mofetil after treatment with cyclo-phosphamide, 2 patients participated in clinical trials and transitioned to mycophenolate mofetil, and other patients received mycophenolate mofetil and pirfenidone (1 patient), cyclophosphamide followed by rituximab (1 patient), rituximab followed by mycophenolate mofetil and tocilizumab (1 patient), rituximab only (1 patient), and autologous hematopoietic stem

cell transplantation (1 patient) (Table 3).During their SSc- associated PH disease course, 24

patients with SSc- associated ILD and PH (82.8%) were treated with PAH- specific therapies. Nine patients (31%) were treated with dual PAH- targeted therapy, 1 of whom was treated with inhaled prostacyclins and 1 of whom was treated with intrave-nous (IV) prostacyclins. Four patients (13.8%) were treated with triple PAH- targeted therapy, 2 of whom were treated with inhaled prostacyclins and 2 of whom were treated with IV prostacyc-lins. Five patients (17.2%) with SSc- associated ILD and PH, 3 of whom were characterized as having WHO Group II and 2 of whom were characterized as having WHO Group III, were not prescribed PAH- specific therapies. The majority (79.2%) of the patients treated with PAH- specific therapies were started on phosphodiesterase 5 (PDE5) inhibitors alone, and the majority (45.8%) of the patients treated with PAH- specific therapies dur-ing their SSc- associated PH disease course were treated with single- agent therapy only (Table 3).

Of the 24 patients who were treated with PAH- specific therapies, 7 patients had PAH, 10 patients had WHO Group III PH with PVR >3.0 Wood units, 1 patient had WHO Group II PH with >20% extent of ILD on HRCT and PVR >3.0 Wood units, and 1 patient had combined postcapillary and precapil-lary PH. The remaining 5 patients had WHO Group II or Group III PH with PVR <3.0 and were treated with PAH therapy due to unexplained decline in DLco, worsening symptoms, and/or severe WHO functional class. Twenty patients (69%) were receiving ILD therapies and PAH- specific therapies simulta-neously. Adverse events due to PAH- specific therapies were

COEXISTENT PH AND ILD IN SSc |      1345

known side effects of the PAH- specific therapies, and no case of worsening ventilation/perfusion mismatch was observed based on chart review.

At PH diagnosis and at the time of data analysis, most patients had a WHO functional class of II or III. Five patients with SSc- associated ILD and PH, who were receiving ILD ther-apies and PAH- specific therapies, improved by at least 1 WHO functional class from PH diagnosis to the time of data analy-sis. One patient with WHO Group III PH, with hemodynamic

Table  2. Cardiopulmonary characteristics of the patients with SSc- associated ILD and PH (n = 29)*

WHO PH group, no. (%)Group 1 7 (24.1)Group 2 6 (20.7)Group 3 16 (55.2)

ILD involvement on HRCT, no. (%)>20% 19 (65.5)<20% 10 (34.5)

PFTs (at PH diagnosis)FVC, % predicted 70.3 ± 18.1TLC, % predicted 84.7 ± 16.5DLco, % predicted 43.1 ± 15.8FVC, % predicted/DLco, % predicted 2 ± 1.5

TTE (at PH diagnosis)RVSP, mm Hg (n = 27)† 44.9 ± 21.6RAP, mm Hg (n = 27)† 7.9 ± 3.2RA dilation, no. (%) 13 (44.8)RV dilation, no. (%) 11 (37.9)Abnormal RV function, no. (%) (n = 28) 6 (21.4)

RHCMean PAP, mm Hg 33.4 ± 7.2Mean PCWP, mm Hg 13.2 ± 3.2Mean RAP, mm Hg 9.9 ± 3.3Cardiac output (by the Fick method) 5.3 ± 1.5Cardiac output (by thermodilution) 5.3 ± 1.5PVR (Wood units) 4.3 ± 3.3Mean PAP on RHC, no. (%)

25–35 mm Hg 20 (69)35–45 mm Hg 8 (27.6)>45 mm Hg 1 (3.4)

PVR on RHC, no. (%)0–6 Wood units 21 (72.4)6–12 Wood units 7 (24.1)>12 Wood units 1 (3.4)

* Except where indicated otherwise, values are the mean ± SD. WHO = World Health Organization; RAP = right atrial pressure; RA = right atrial; RHC = right- sided heart catheterization; PAP = pulmo-nary arterial pressure; PCWP = pulmonary capillary wedge pres-sure; PVR = pulmonary vascular resistance (see Table 1 for other definitions). † No tricuspid regurgitation jet was observed in 2 patients.

Table  3. Treatment of and outcomes in patients with SSc- associated ILD and PH (n = 29)*

History of treatment with PAH- targeted therapies 24 (82.8)Initial treatment with PAH- targeted therapies

None 5 (17.2)PDE5 inhibitor 19 (65.5)PDE5 inhibitor and ERA 4 (13.8)IV prostacyclin 1 (3.4)

Most recent PAH- targeted therapy regimenNone 6 (20.7)PDE5 inhibitor 13 (44.8)ERA 1 (3.4)PDE5 inhibitor and ERA 3 (10.3)PDE5 inhibitor and inhaled prostacyclin 1 (3.4)ERA and IV prostacyclin 1 (3.4)PDE5 inhibitor, ERA, and inhaled prostacyclin 2 (6.9)PDE5 inhibitor, ERA, and IV prostacyclin 1 (3.4)PDE5 inhibitor, ERA, and clinical trial 1 (3.4)

Use of single- agent PAH- targeted therapy 11 (37.9)Use of dual- agent PAH- targeted therapy 9 (31)Use of triple- agent PAH- targeted therapy 4 (13.8)Requirement of prostacyclin during PH therapy 6 (20.7)History of ILD treatment 24 (82.8)History of ILD treatment with mycophenolate mofetil 21 (72.4)History of ILD treatment with IV pulse

cyclophosphamide8 (27.6)

Most recent ILD treatmentNone 9 (31)Mycophenolate mofetil 15 (51.7)Rituximab 2 (6.9)Tocilizumab and mycophenolate mofetil 1 (3.4)Pirfenidone and mycophenolate mofetil 1 (3.4)Cyclophosphamide 1 (3.4)

History of supplemental oxygen use 9 (31)History of transplantation† 1 (3.4)Alive or deceased

Alive 27 (93.1)Deceased 2 (6.9)

WHO functional class (prior to PH diagnosis)Class I 0 (0)Class II 12 (41.4)Class III 16 (55.2)Class IV 1 (3.4)

WHO functional class (most recent)Class I 4 (13.8)Class II 8 (27.6)Class III 17 (58.6)Class IV 0 (0)

* Values are the number (%). PAH = pulmonary arterial hyperten-sion; PDE5 = phosphodiesterase 5; ERA = endothelin receptor an-tagonist; IV = intravenous; WHO = World Health Organization (see Table 1 for other definitions). † Autologous hematopoietic stem cell transplantation (HSCT).

YOUNG ET AL 1346       |

features of PAH with PVR >3.0, who had a PDE5 inhibitor and endothelin receptor antagonist (ERA) added to prior long- term mycophenolate mofetil therapy, improved by 2 functional classes from functional class III to functional class I. Sixteen patients with SSc- associated ILD and PH receiving PAH- specific therapies, 4 of whom were never treated with ILD ther-apy, had a stable WHO functional class during their disease course (Figure 2).

Nine patients with SSc- associated ILD and PH (31%) received supplemental oxygen during their PH disease course. No patients with SSc- associated ILD and PH underwent heart or lung trans-plantation during their ILD and/or PH disease course; however, 1 patient had undergone autologous hematopoietic stem cell transplantation 1 year after ILD diagnosis. Two patients with SSc- associated ILD and PH (6.9%) died during the study. One patient whose cause of death was PH had WHO Group III PH and had been treated with cyclophosphamide followed by mycophenolate mofetil and was not treated with PAH- specific therapy. The other patient had PAH, was treated with an ERA and IV prostacyclin, and died of a respiratory infection.

Survival analysis. Overall, 3 patients in the entire SSc- associated ILD cohort died (1 of respiratory infection, 1 of PH, and 1 of an unknown cause due to the patient being lost to follow- up). The survival rate 3 years after diagnosis of SSc- associated ILD for

all patients in the cohort was 97% (survival standard error 0.03). Survival analysis for those with SSc- associated ILD and PH indi-cated that after diagnosis of PH 2 patients died (1 of respiratory infection and the other of PH), resulting in a survival rate at 3 years after PH diagnosis of 91% (survival standard error 0.06); one of the patients with SSc- associated ILD and PH who died had clin-ically meaningful ILD. For the entire SSc- associated ILD cohort and those with SSc- associated ILD and PH, Cox proportional hazards regression of time to death was conducted with adjust-ment for current age and race (non- Hispanic white versus non-white) to account for differences between the cohorts; however, neither of those variables were significant in either survival model.

DISCUSSION

In this study, we examined the prevalence and explored the clinical characteristics, pharmacologic therapies, and outcomes in SSc- associated ILD. Our results demonstrate that a large propor-tion of patients with SSc- associated ILD (31.2%) have coexisting PH, and 37.9% and 48.3% of PH diagnoses occurred within 3 years and 7 years, respectively, of the onset of SSc. Most patients with SSc- associated ILD and PH had WHO Group III PH based on >20% extent of ILD on HRCT (55.2%), and the majority (10 of 16 [63%]) of those patients with WHO Group III PH had hemody-namic features of PAH with PVR >3.0 Wood units. The group with

Figure 2. World Health Organization functional class (FC) prior to diagnosis of pulmonary hypertension (PH) in patients with systemic sclerosis–associated interstitial lung disease and PH and change in functional class at the time of data analysis.

COEXISTENT PH AND ILD IN SSc |      1347

SSc- associated ILD and PH also included a significant proportion of African Americans, consistent with prior studies of SSc that have shown that African Americans have a higher incidence of PH, and a recent multicenter African American cohort in which a similar proportion of patients (18%) had RHC- proven PH (19–21). The majority of patients who had SSc- associated ILD and PH had received immunosuppressive therapy for their ILD (82.8%), and 82.8% received PAH- specific therapy. The survival rate for those with SSc- associated ILD and PH 3 years after PH diagnosis was 91%, irrespective of ILD- and/or PAH- specific therapies.

Compared to prior studies that evaluated SSc patients with ILD and PH, our study is unique in that it focused on a cohort of patients with SSc- associated ILD and assessed the prevalence and the clinical course of PH in a prospective fashion, which indicated that the coexistence of PH with SSc- associated ILD is not uncommon. Thus, patients with SSc- associated ILD may not have cardiopulmonary symptoms solely due to ILD, and may in fact have PH contributing to their abnormal cardiopulmonary symptoms and physiology. Most studies report PH prevalence only for PAH in patients without significant lung fibrosis, so direct comparison of our results with other study populations is difficult. Also, the high prevalence of PH in our SSc- associated ILD cohort may reflect some referral bias, since our institution is well known for its PH expertise and our adherence to the 2013 recommen-dations for screening and detection of CTD- associated PAH in our population (8). Our review of the literature identified 7 origi-nal studies of RHC- proven SSc- associated PH with and without ILD. Of those, 4 studies evaluated SSc- associated PH related to ILD (SSc- associated PH- ILD) (17,18,22,23), 2 studies evaluated coexisting ILD in SSc patients who were initially evaluated for RHC- proven PAH (24,25), and an additional study by Trad et al (26) evaluated a cohort of patients with dcSSc, 17% of whom had ILD and PAH, although some of those patients had only TTE- proven PAH. Thus, future studies need to focus on assessing the prevalence of coexisting PH and ILD in a systematic fashion.

Although the development of PAH is largely thought to be a late complication in SSc, our study indicates that a large proportion of patients with SSc- associated ILD and PH have a diagnosis of PH within 5 years of SSc onset (27). Similar results were reported in a study by Mathai et  al (23), which demonstrated a median SSc disease duration of 4 years for all patients with PH, 3 years for patients with SSc- associated PH- ILD, and 4.5 years for those with SSc- associated PAH (23). Also, in a retrospective multicenter study by Hachulla et al, the mean ± SD SSc disease duration at PAH diagnosis was 6.3 ± 6.6 years, 55.1% of the patients were classified as having early- onset PAH based on a PAH diagnosis occurring within 5 years of an initial non- RP symptom, and >50% of the population studied had evidence of pulmonary fibrosis on CT as well (28). One possible explanation for the earlier diagnosis of PH may be the recognition and incorporation of screening algo-rithms for PH in SSc (6,8,12). This evidence of early development of PH in SSc patients sheds light on a potential problem that may

exist in the current design of SSc- associated ILD clinical trials, which typically recruit patients within 5–7 years of the initial SSc non- RP sign/symptom and do not require a diagnostic RHC to rule out PH as a cause of cardiopulmonary signs and symptoms.

Clinical trials evaluating PAH therapies typically exclude patients with clinically significant ILD and/or PH due to SSc- related PH- ILD; however, according to our experience and pre-vious studies by Launay et al (22) and Mathai et al (23), patients with clinically significant ILD and PH receive both ILD- specific and PAH- specific therapies. The 5th World Symposium on Pulmonary Hypertension (12) has endorsed management of WHO Group III PH by expert centers. Their recommendations acknowledge the lack of evidence for the use of PAH- specific therapies for patients with WHO Group III PH, citing the potential for worsening gas exchange in ILD patients due to ventilation/perfusion mismatch. Despite this, our cohort and other cohorts highlight the use of PAH- specific therapies in these patients. However, we acknowl-edge that differentiating WHO Group I PH from WHO Group III PH in this population remains a challenge, and should not be based upon arbitrary cutoffs involving FVC and HRCT findings (13).

Patients with SSc- associated ILD and PH may also have a combination of WHO Group I PH and WHO Group III PH, which have different disease mechanisms. PAH is a vasculopathy char-acterized by vascular remodeling with inflammation, fibrosis, and thrombosis, whereas WHO Group III PH is due to vascular destruc-tion from lung fibrosis, vasoconstriction due to chronic hypoxia, and/or a vasculopathy similar to that seen in PAH but “dispropor-tionate” to what is seen in PH due to chronic lung disease (13,29). “Out of proportion” PH in some forms of chronic lung diseases has recently been defined by Seeger et al (13) as severe PH due to chronic lung disease with hemodynamic findings of a mean PAP of ≥35 mm Hg or a mean PAP of ≥25 mm Hg and a low cardiac index (<2.0 liters/minute/m2). As evidenced in our cohort, a large proportion of patients with >20% extent of ILD on HRCT actually have features of both PAH and WHO Group III PH, and individuals within this group were treated with both PAH- specific therapies and immunosuppression for ILD on a case- by- case basis. The majority of patients in our study tolerated PAH- specific therapy regardless of simultaneous or prior ILD treatment and had stability in WHO functional class and/or 6- minute walk test.

Survival for patients with coexisting PH in the setting of SSc- associated ILD has varied across cohorts, but despite the exis-tence of various PAH therapies, survival overall remains poor for both SSc- associated PAH and SSc- associated PH- ILD, but tends to be worse if PH is due to ILD. However, our 91% survival rate 3 years after PH diagnosis is higher than survival rates reported in studies by Launay et al (22), Le Pavec et al (17), Mathai et al (23), Michelfelder et al (25), and Volkmann et al (18). Based on the average mean PAP and PVR on RHC and use of prosta cyclins, patients in our SSc- associated ILD and PH cohort appear to have less severe PH than those in the previously mentioned studies, which is likely due to our aggressive PH screening, creating lead

YOUNG ET AL 1348       |

time bias and improved survival rates. Our survival rates may also be overestimated due to our small cohort size and infrequent events. We also evaluated our population of patients for clinically significant ILD (>30% disease extent on HRCT, or 10–30% dis-ease extent on HRCT and an FVC <70%), and found that 51.7% of the patients with PH (15 of 29) had clinically significant ILD. In comparison, the cohorts in the studies by Le Pavec et al (17) and Volkmann et al (18) included SSc patients with PH and clinically significant ILD with a total of 70 patients with a 3- year survival estimate of 21% and 71 patients with a 3- year survival estimate of 50%, respectively (17,18). The 3- year survival rate of 91% in our SSc- associated ILD and PH cohort, with >50% of patients with clinically significant ILD, is likely related to inclusion of HRCT and implementation of the 2013 recommendations in all patients seen at the University of Michigan and milder PH based on hemody-namic findings and the use of prostacyclins in our cohort (8).

Our cohort study also highlights an ongoing dilemma in the classification of SSc- associated PH, as there is lack of a standard definition of what constitutes a significant degree of ILD based on pulmonary physiology and/or radiographic severity to classify patients as having WHO Group I PH/PAH or WHO Group III PH. Recent data from a single- center cohort highlight the lack of spec-ificity of FVC % predicted for the assessment of the presence and severity of ILD in SSc (30). Additional differences in the definition of clinically significant ILD are evident when evaluating 2 recent large clinical trials of PAH- targeted therapies. In the PAH trial evaluating combination therapy with ambrisentan and tadalafil and the PAH trial evaluating the use of selexipag, patients with moderate-to- severe restrictive lung disease defined as a TLC of <60–70% were excluded (31,32). In the cohort evaluated in the present study, moderate- to- severe ILD causing WHO Group III PH was defined using Goh’s criteria (10), with all patients with features of WHO Group III PH having >20% extent of ILD on HRCT. Fischer et al (24), Volkmann et al (18), and Le Pavec et al (17) defined SSc- associated PH- ILD similar to or based on Goh’s criteria. Volkman et al and Le Pavec et al specifically defined significant ILD as an extent of fibrosis of >30% of lung involvement on HRCT, or an extent of fibrosis of 10–30% and an FVC of <70%. Launay et al (22) and Mathai et  al (23) based their SSc- associated PH- ILD diagnosis on PFTs and HRCT findings. A consensus is urgently needed for the definition of significant ILD to determine whether a patient has SSc- associated PAH or SSc- associated PH- ILD.

Our study has many strengths. First, we studied a well- characterized prospective SSc- associated ILD cohort recruited at a single center. Second, all patients underwent screening for PH based on the 2013 recommendations by Khanna et al, and those who met the criteria underwent RHC (8). However, this study is not without limitations. Our results may be skewed since patients were recruited at a tertiary care center with highly spe-cialized scleroderma, ILD, and PAH clinics. Like other cohorts, the management of PH and ILD was not standardized, which may have impacted the outcomes. Last, although we instituted

a standardized algorithm for PH screening, we may have missed some patients with mild PH.

Patients with SSc- associated ILD can also develop PH early on in their SSc disease course. These patients with ILD and PH often have dcSSc and often have features of WHO Group III PH due to their ILD but also have hemodynamic features of PAH, which may warrant the use of both immunosuppressive ther-apies and PAH- specific therapies. The presence of PH early on in patients with SSc- associated ILD is a key factor we must recognize when designing clinical trials for SSc- associated ILD, since PH may be confounding patient- reported outcomes and cardiopulmonary physiology in these patients, which may affect the outcome of clinical trials. Future prospective studies in SSc- associated ILD should confirm our findings and also explore the impact of the new hemodynamic definition of PH, which was recently proposed at the 6th World Symposium on Pulmonary Hypertension (33,34).

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Khanna had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Young, Visovatti, Flaherty, McLaughlin, Khanna.Acquisition of data. Young, Vummidi, Visovatti, Flaherty, McLaughlin, Khanna.Analysis and interpretation of data. Young, Vummidi, Visovatti, Homer, Wilhalme, White, Flaherty, McLaughlin, Khanna.

ADDITIONAL DISCLOSURES

Author Khanna is an employee of CiviBioPharma.

REFERENCES 1. Steen VD, Medsger TA. Changes in causes of death in systemic

sclerosis, 1972- 2002. Ann Rheum Dis 2007;66:940–4.

2. Denton CP, Khanna D. Systemic sclerosis. Lancet 2017;390:1685–99.

3. Steen VD, Conte C, Owens GR, Medsger TA Jr. Severe restrictive lung disease in systemic sclerosis. Arthritis Rheum 1994;37:1283–9.

4. Schurawitzki H, Stiglbauer R, Graninger W, Herold C, Polzleitner D, Burghuber OC, et al. Interstitial lung disease in progressive systemic sclerosis: high- resolution CT versus radiography. Radiology 1990;176:755–9.

5. Hinchcliff M, Fischer A, Schiopu E, Steen VD. for the PHAROS Investigators. Pulmonary Hypertension Assessment and Recognition of Outcomes in Scleroderma (PHAROS): baseline characteristics and description of study population. J Rheumatol 2011;38:2172–9.

6. Coghlan JG, Denton CP, Grunig E, Bonderman D, Distler O, Khanna D, et al. Evidence- based detection of pulmonary arterial hypertension in systemic sclerosis: the DETECT study. Ann Rheum Dis 2014;73:1340–9.

7. Van den Hoogen F, Khanna D, Fransen J, Johnson SR, Baron M, Tyndall A, et al. 2013 classification criteria for systemic sclerosis: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum 2013;65:2737–47.

COEXISTENT PH AND ILD IN SSc |      1349

8. Khanna D, Gladue H, Channick R, Chung L, Distler O, Furst DE, et al. Recommendations for screening and detection of connective tissue disease–associated pulmonary arterial hypertension. Arthritis Rheum 2013;65:3194–201.

9. Nihtyanova SI, Schreiber BE, Ong VH, Rosenberg D, Moinzadeh P, Coghlan JG, et al. Prediction of pulmonary complications and long- term survival in systemic sclerosis. Arthritis Rheumatol 2014;66:1625–35.

10. Goh NS, Desai SR, Veeraraghavan S, Hansell DM, Copley SJ, Maher TM, et al. Interstitial lung disease in systemic sclerosis: a simple staging system. Am J Respir Crit Care Med 2008;177:1248–54.

11. Galie N, Humbert M, Vachiery JL, Gibbs S, Lang I, Torbicki A, et  al. 2015 ESC/ERS guidelines for the diagnosis and treatment of pulmonary hypertension: The Joint Task Force for the Diagnosis and Treatment of Pulmonary Hypertension of the European Society of Cardiology (ESC) and the European Respiratory Society (ERS): endorsed by: Association for European Paediatric and Congenital Cardiology (AEPC), International Society for Heart and Lung Transplantation (ISHLT). Eur Respir J 2015;46:903–75.

12. Galie N, Simonneau G. The Fifth World Symposium on Pulmonary Hypertension. J Am Coll Cardiol 2013;62 Suppl:D1–3.

13. Seeger W, Adir Y, Barbera JA, Champion H, Coghlan JG, Cottin V, et al. Pulmonary hypertension in chronic lung diseases. J Am Coll Cardiol 2013;62 Suppl:D109–16.

14. Tashkin DP, Elashoff R, Clements PJ, Goldin J, Roth MD, Furst DE, et  al. Cyclophosphamide versus placebo in scleroderma lung disease. N Engl J Med 2006;35:2655–66.

15. Tashkin DP, Roth MD, Clements PJ, Furst DE, Khanna D, Kleerup EC, et al. Mycophenolate mofetil versus oral cyclophosphamide in scleroderma- related interstitial lung disease (SLS II): a randomised controlled, double- blind, parallel group trial. Lancet Respir Med 2016;4:708–19.

16. Vachiery JL, Adir Y, Barbera JA, Champion H, Coghlan JG, Cottin V, et al. Pulmonary hypertension due to left heart diseases. J Am Coll Cardiol 2013;62 Suppl:D100–8.

17. Le Pavec J, Girgis RE, Lechtzin N, Mathai SC, Launay D, Hummers  LK,  et al. Systemic sclerosis–related pulmonary hypertension associated with interstitial lung disease: impact of pulmonary arterial hypertension therapies. Arthritis Rheum 2011;63:2456–64.

18. Volkmann ER, Saggar R, Khanna D, Torres B, Flora A, Yoder L, et al. Improved transplant- free survival in patients with systemic sclerosis–associated pulmonary hypertension and interstitial lung disease. Arthritis Rheumatol 2014;66:1900–8.

19. Blanco I, Mathai S, Shafiq M, Boyce D, Kolb TM, Chami H, et al. Severity of systemic sclerosis- associated pulmonary arterial hypertension in African Americans. Medicine (Baltimore) 2014;93:177–85.

20. Reveille JD, Fischbach M, McNearney T, Friedman AW, Aguilar MB, Lisse J, et al. Systemic sclerosis in 3 US ethnic groups: a comparison of clinical, sociodemographic, serologic, and immunogenetic determinants. Semin Arthritis Rheum 2001;30:332–46.

21. Morgan ND, Shah AA, Mayes MD, Domsic RT, Medsger TA Jr, Steen VD, et al. Clinical and serological features of systemic sclerosis in a multicenter African American cohort: analysis of the genome research in African American Scleroderma Patients clinical database. Medicine (Baltimore) 2017;96:e8980.

22. Launay D, Humbert M, Berezne A, Cottin V, Allanore Y, Couderc LJ, et al. Clinical characteristics and survival in systemic sclerosis- related pulmonary hypertension associated with interstitial lung disease. Chest 2011;140:1016–24.

23. Mathai SC, Hummers LK, Champion HC, Wigley FM, Zaiman A, Hassoun PM, et al. Survival in pulmonary hypertension associated with the scleroderma spectrum of diseases: impact of interstitial lung disease. Arthritis Rheum 2009;60:569–77.

24. Fischer A, Swigris JJ, Bolster MB, Chung L, Csuka ME, Domsic R, et al. Pulmonary hypertension and interstitial lung disease within PHAROS: impact of extent of fibrosis and pulmonary physiology on cardiac haemodynamic parameters. Clin Exp Rheumatol 2014; 32 Suppl 86:S109–14.

25. Michelfelder M, Becker M, Riedlinger A, Siegert E, Dromann D, Yu X, et al. Interstitial lung disease increases mortality in systemic sclerosis patients with pulmonary arterial hypertension without affecting hemodynamics and exercise capacity. Clin Rheumatol 2017;36:381–90.

26. Trad S, Amoura Z, Beigelman C, Haroche J, Costedoat N, Le Boutin TH, et al. Pulmonary arterial hypertension is a major mortality factor in diffuse systemic sclerosis, independent of interstitial lung disease. Arthritis Rheum 2006;54:184–91.

27. Medsger TA Jr. Natural history of systemic sclerosis and the assessment of disease activity, severity, functional status, and psychologic well- being. Rheum Dis Clin North Am 2003;29:255– 73.

28. Hachulla E, Launay D, Mouthon L, Sitbon O, Berezne A, Guillevin L, et al. Is pulmonary arterial hypertension really a late complication of systemic sclerosis? Chest 2009;136:1211–9.

29. Girgis RE, Mathai SC. Pulmonary hypertension associated with chronic respiratory disease. Clin Chest Med 2007;28:219–32.

30. Suliman YA, Dobrota R, Huscher D, Nguyen-Kim TD, Maurer B, Jordan S, et al. Pulmonary function tests: high rate of false- negative results in the early detection and screening of scleroderma- related interstitial lung disease. Arthritis Rheumatol 2015;67:3256–61.

31. Sitbon O, Channick R, Chin KM, Frey A, Gaine S, Galie N, et al. Selexipag for the treatment of pulmonary arterial hypertension. N Engl J Med 2015;373:2522–33.

32. Galie N, Barbera JA, Frost AE, Ghofrani HA, Hoeper MM, McLaughlin  VV, et al. Initial use of ambrisentan plus tadalafil in pulmonary arterial hypertension. N Engl J Med 2015;373:834–44.

33. Frost A, Badesch D, Gibbs JS, Gopalan D, Khanna D, Manes A, et al. Diagnosis of pulmonary hypertension. Eur Resp J 2019;53.

34. Simonneau G, Montani D, Celermajer DS, Denton CP, Gatzoulis MA, Krowka M, et al. Haemodynamic definitions and updated clinical classification of pulmonary hypertension. Eur Respir J 2018;53.

1350

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1350–1359DOI 10.1002/art.40890 © 2019, American College of Rheumatology

Identification of Cysteine- Rich Angiogenic Inducer 61 as a Potential Antifibrotic and Proangiogenic Mediator in SclerodermaPei-Suen Tsou, Dinesh Khanna, and Amr H. Sawalha

Objective. We previously identified CYR61 as a histone deacetylase 5 (HDAC- 5)–repressed gene in systemic sclerosis (SSc; scleroderma) endothelial cells (ECs). When overexpressed, cysteine- rich angiogenic inducer 61 (CYR- 61) promoted angiogenesis in SSc ECs. This study was undertaken to examine the role of CYR- 61 in fibrosis and determine the mechanisms involved in CYR- 61–mediated angiogenesis in SSc.

Methods. Dermal ECs and fibroblasts were isolated from biopsy specimens from healthy subjects and patients with SSc. CYR- 61 level was determined by quantitative polymerase chain reaction, Western blotting, and enzyme- linked immunosorbent assay. CYR- 61 was overexpressed using a CYR61 vector or knocked down using small inter-fering RNA, and functional and mechanistic studies were then conducted in fibroblasts and ECs.

Results. Lower CYR 61 messenger RNA levels were observed in dermal fibroblasts and ECs from SSc patients than in those from healthy controls. In SSc fibroblasts, overexpression of CYR- 61 led to significant reduction in the expression of profibrotic genes, including COL1A1 (P = 0.002) and ACTA2 (P = 0.04), and an increase in the expression of matrix- degrading genes, including MMP1 (P = 0.002) and MMP3 (P =0.004), and proangiogenic VEGF (P = 0.03). The antifibrotic effect of CYR- 61 was further demonstrated by delay in wound healing, inhibition of gel contraction, inactivation of the transforming growth factor β pathway, and early superoxide production associated with senescence in SSc fibroblasts. In SSc ECs, overexpression of CYR- 61 led to increased production of vascular endothelial cell growth factor. The proangiogenic effects of CYR- 61 were mediated by signaling through αvβ3 recep-tors and downstream activation of AMP- activated protein kinase, AKT, and the endothelial cell nitric oxide synthase/nitric oxide pathway system.

Conclusion. CYR- 61, which is epigenetically regulated by HDAC- 5, is a potent antifibrotic and proangiogenic mediator in SSc. Therapeutic intervention to promote CYR- 61 activity or increase CYR- 61 levels might be of benefit in SSc.

INTRODUCTION

Systemic sclerosis (SSc; scleroderma) is a connective tissue disease that is characterized by immune activation, vascular abnormalities, and progressive fibrosis in the skin and internal organs. It is a rare disease with a prevalence of 150–300 cases per million in Europe and the US (1). This dis-ease is associated with significant morbidity and can lead to life- threatening complications, including pulmonary arterial hypertension, interstitial lung disease, and scleroderma renal

crisis. Although the etiology of SSc is still unclear, activation of immune cells and abnormalities of fibroblasts and endothelial cells (ECs) contribute to the pathogenesis of the disease (2). In addition, dysregulation of epigenetic mechanisms in SSc has been implicated in various cell types, and the use of epigenetic modifying drugs has been shown to be beneficial in cells as well as in animal models of SSc (2).

We recently showed that overexpression of the antiangiogenic histone deacetylase 5 (HDAC- 5) in SSc ECs contributes to impaired angiogenesis by repressing proangio-

Dr. Tsou’s work was supported by the Arthritis National Research Foundation and the Scleroderma Foundation. Dr. Khanna’s work was supported by NIH grants K-24-AR-063120 and UM1-AI-110557 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) and the National Institute of Allergy and Infectious Diseases (NIAID), respectively. Dr. Sawalha’s work was supported by the NIH (NIAID grants R01-AI-097134 and U19-AI-110502 and NIAMS grant R01-AR-070148).

Pei-Suen Tsou, PhD, Dinesh Khanna, MD, Amr H. Sawalha, MD: University of Michigan, Ann Arbor.

No potential conflicts of interest relevant to this article were reported.Address correspondence to Amr H. Sawalha, MD, Division of

Rheumatology, University of Michigan, 1150 West Medical Center Drive, 5520 MSRB1, SPC 5680, Ann Arbor, MI 48109. E-mail: asawalha@umich.edu.

Submitted for publication October 5, 2018; accepted in revised form March 12, 2019.

CYR- 61 IS ANTIFIBROTIC AND PROANGIOGENIC IN SSc |      1351

genic factors (3). We took an unbiased approach to examine genome- wide changes in chromatin accessibility after HDAC- 5 knockdown in ECs using an assay for transposase- accessible chromatin with sequencing. HDAC- 5 knockdown led to increased chromatin accessibility, and through bioinformatics analyses and functional assays we identified novel HDAC- 5 tar-get genes associated with impaired angiogenesis in SSc ECs, including CYR61.

Cysteine- rich angiogenic inducer 61 (CYR- 61) is a mem-ber of the CCN protein family (CYR- 61, connective tissue growth factor [CTGF], nephroblastoma- overexpressed gene, and Wnt- 1–inducible signaling pathway proteins 1–3), which plays important roles in development, inflammation, tissue repair, and a broad range of pathologic processes including fibrosis and cancer (4). CYR- 61 assumes its diverse functions by its ability to bind different combinations of co- receptors. For example, CYR- 61 promotes EC proliferation and sur-vival through binding to αvβ3 (5,6), but enhances fibroblast adhesion and senescence through binding to α6β1 and hep-aran sulfate proteoglycans (HSPGs) (7,8). Since this matri-cellular protein supports angiogenesis (6,9) and possesses antifibrotic properties (8,10), we hypothesize that CYR- 61 promotes angiogenesis and inhibits fibrosis in SSc ECs and fibroblasts, respectively. In this study, we characterized the expression and function of CYR- 61 in diffuse SSc dermal fibro-blasts, and further dissected the mechanisms involved in the proangiogenic properties of CYR- 61 in this disease.

PATIENTS AND METHODS

Patients and controls. Two 4- mm punch biopsy spec-imens from the distal forearm of healthy controls and patients with SSc were obtained for fibroblast and EC isolation. Plasma samples from the study participants were also collected to assess plasma CYR- 61 protein levels. All patients met the Amer-ican College of Rheumatology/European League Against Rheu-matism criteria for the classification of SSc (11). This study was approved by the Institutional Review Board of the University of Michigan and all study participants signed a written informed consent. The demographic and clinical characteristics of the

study participants are summarized in Table 1.

Cell culture. Dermal fibroblasts and ECs were isolated and characterized as previously described (12–14). After digestion, ECs were purified using a CD31 MicroBead Kit (Miltenyi Biotec) and grown in EBM- 2 media with growth factors (Lonza). Fibro-blasts were maintained in RPMI supplemented with 10% fetal bovine serum (FBS). Cells between passage 3 and 6 were used in all experiments.

CYR- 61 overexpression. Overexpression of CYR- 61 in ECs was induced as previously described (3). ECs were trans-

fected with 1.65 μg of CYR61 (control vector pCMV6- XL4; Ori-Gene) and Lipofectamine 2000 (3.3 μl; Invitrogen) for 24 hours in T12.5 flasks. Five hours after transfection, culture media were changed to allow cells to grow in endothelial growth medium sup-plemented with bovine brain extract (Lonza). To overexpress CYR- 61 in dermal fibroblasts, 1 μg of CYR61 and 1 μl of Lipofectamine 2000 were used to transfect the cells in a 12- well plate for a total of 48 hours.

CYR- 61 knockdown. CYR- 61 expression was knocked down using CYR61 small interfering RNA (siRNA) (On- Target Plus siRNA; Dharmacon), while a nontargeting siRNA (Dharma-con) was used as a control. Fibroblasts isolated from healthy controls were transfected with 350 nM of siRNA using TransIT‐TKO transfection reagent (Mirus Bio) for 48 hours before mes-senger RNA (mRNA) was collected.

Western blotting. After obtaining cell lysates from cul-tured cells, proteins were separated by sodium dodecyl sul-fate–polyacrylamide gel electrophoresis and electroblotted onto nitrocellulose membranes. Antibodies used included phospho–transforming growth factor β receptor type II (phospho- TGFβRII; Thermo Scientific), CYR- 61, TGFβRII, p21, and p16 (all from Abcam), phospho–retinoblastoma protein (phospho-pRB), pRB, TGFβ, phospho- Smad2, phospho- Smad3, Smad2/3, p53, phospho- p38, p38, phospho–endothelial cell nitric oxide syn-thase (phospho- eNOS), eNOS, phospho- AKT, AKT, phospho–AMP- activated protein kinase (phospho- AMPK), and AMPK (all from Cell Signaling Technology). β- actin was used as a loading

Table 1. Characteristics of the SSc patients and healthy controls*

SSc patients (n = 62)

Age, mean ± SD years 55.8 ± 1.7Sex, female/male 50/12dcSSc 41Disease duration, mean ± SD years 8.9 ± 1.4Modified Rodnan skin thickness score,

mean ± SD 12.2 ± 1.3

Raynaud’s phenomenon 56Early disease (<5 years) 37Digital ulcers 12Telangiectasias 27Gastrointestinal disease 47Interstitial lung disease 29Pulmonary arterial hypertension 16Renal involvement 3

* In the healthy control group (n = 40), there were 31 women and 9 men, and the mean ± SD age was 49.5 ± 2.5 years. Except where indicated otherwise, values are the number of subjects. SSc = systemic sclerosis; dcSSc = diffuse cutaneous SSc.

TSOU ET AL 1352       |

control (Sigma- Aldrich). Band quantification was performed using ImageJ software (15).

Messenger RNA extraction and quantitative reverse transcriptase–polymerase chain reaction (PCR). Total RNA from ECs and fibroblasts was isolated using a Direct- zol RNA MiniPrep kit (Zymo Research). A Verso complementary DNA (cDNA) synthesis kit was used to prepare cDNA (Thermo Sci-entific). Primers for human CYR61, COL1A1, ACTA2, PPARG, MMP1, MMP3, TGFB, TGFBR2, VEGF, FGF2, CDK1A, and ACTB along with Power SYBR Green PCR Master Mix (Applied Biosystems), were used in quantitative PCR, which was run on a ViiA 7 Real- Time PCR System. Primers used were KiCqStart SYBR Green primers from Sigma-Aldrich or QuantiTect Primer Assays from Qiagen.

Enzyme- linked immunosorbent assay (ELISA). After transfection with CYR61 or control vectors, cell culture media were changed to RPMI (for fibroblasts) or EBM- 2 (for ECs) with 0.1% FBS and cultured overnight. The levels of CYR- 61, matrix metalloproteinase 1 (MMP- 1), MMP- 3, vascular endothelial growth factor (VEGF), and basic fibroblast growth factor in cell culture supernatants were measured using ELISA kits from R&D Systems. Absorbance at 450 nm in each well was read using a microplate reader.

β- galactosidase measurement. To examine the effect of CYR- 61 on cell senescence, we measured β- galactosidase using a Senescence β- Galactosidase Staining Kit from Cell Signaling Technology.

Gel contraction and cell migration assays. The gel con-traction assay was conducted as previously described (16). Forty- eight hours after CYR- 61 overexpression or knockdown, dermal fibroblasts were suspended in culture media at 2 × 106 cells/ml. A Cell Contraction Assay kit (Cell Biolabs) was used for gel contrac-tion. The area of the gel was analyzed using ImageJ software (15). To evaluate the effect of CYR- 61 on cell migration, we performed cell migration assays using SSc fibroblasts transfected with con-trol or CYR61 vectors. Transfected cells were grown to confluence and a wound gap was created. Culture media were replaced with RPMI with 0.1% FBS, and pictures were taken using an EVOS XL Core Cell Imaging System (Life Technologies) at 0 hours and 48 hours after wounding. The gap difference was quantified using ImageJ software (15).

Immunofluorescence staining. Cells grown in 8- well chambers were fixed in 4% formalin and blocked. They were then probed with anti–Ki- 67 antibodies (Abcam) or anti- human p21 antibodies (Abcam). Alexa Fluor antibodies (Invitrogen) were subsequently used. The nuclei were stained using DAPI (Invitrogen). Ki- 67–positive and p21- positive cells were counted

using ImageJ software (15). To measure superoxide levels, dihydroethidium (DHE; Invitrogen) was used. After fixation, DHE (10 μM) was added to CYR61- transfected or control- transfected fibroblasts for 15 minutes, and then stained with DAPI. Fluores-cence was detected using an Olympus FV500 confocal micro-scope and photographs were taken at 400×.

Nitrite/nitrate measurement. To examine whether increased nitric oxide (NO) was present after CYR- 61 overexpres-sion, nitrite and nitrate levels in cell culture supernatant, both sta-ble metabolites for NO, were measured using an OxiSelect In Vitro Nitric Oxide (Nitrite/Nitrate) Assay Kit from Cell Biolabs.

Matrigel tube formation assay. In our previous study, we showed that CYR- 61 overexpression in SSc ECs resulted in increased tube formation (3). To examine whether this effect is medi-ated through αvβ3 and its downstream pathways, we pretreated CYR- 61–overexpressing SSc ECs with neutralizing antibodies to αvβ3 (10 μg/ml) or inhibitors for AMPK (compound c; 10 μM) or AKT (LY297002; 20 μM) pathways before performing Matrigel tube formation assays. Treated SSc ECs were suspended in EBM- 2 with 0.1% FBS and plated in 8- well Lab- Tek chambers coated with growth factor–reduced Matrigel (BD Biosciences). Cells were fixed and stained after 6 hours of incubation. Quantitation of the tubes formed by ECs was performed using the Angiogenesis Analyzer function in ImageJ software (15). Pictures of each well were taken using an EVOS XL Core Cell Imaging System (Life Technologies).

Statistical analysis. Results are expressed as the mean ± SD. To determine the differences between the groups, Student’s t- test, paired t- test, or one- way analysis of variance was performed using GraphPad Prism software, version 6. P values less than 0.05 were considered significant.

RESULTS

CYR- 61 expression in SSc. Since CYR- 61 is secreted and detected in blood, we first measured CYR- 61 levels in plasma collected from healthy controls and SSc patients. We observed no differences in CYR- 61 levels between healthy controls and SSc patients, including after stratifying patients into the cate-gories diffuse cutaneous SSc (dcSSc) or limited cutaneous SSc (lcSSc) (Figure 1A). In dermal fibroblasts, there was significant down- regulation of CYR61 mRNA levels in both dcSSc and lcSSc fibroblasts compared to normal fibroblasts (Figure  1B). Protein levels of CYR- 61 were measured in dcSSc fibroblasts and were found to be similar to those in control fibroblasts, as indicated by levels of secreted CYR- 61 in supernatants or CYR- 61 protein levels in cell lysates (Figures 1C and D).

Antifibrotic properties of CYR- 61 in SSc. To eval-uate the effect of CYR- 61 in dermal fibroblasts, we first

CYR- 61 IS ANTIFIBROTIC AND PROANGIOGENIC IN SSc |      1353

overexpressed CYR- 61 in normal fibroblasts. Overexpres-sion of CYR- 61 led to a significant decrease in fibrotic mark-ers, including COL1A1 and ACTA2, suggesting that CYR- 61 is indeed antifibrotic in these cells (Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40890/ abstract). We then overexpressed CYR- 61 in dcSSc dermal fibro-blasts. Similar to what was observed in normal fibroblasts, increased CYR- 61 led to a significant reduction in COL1A1 and ACTA2 expression (Table  2). This was accompanied by a significant increase in the expression of antifibrotic PPARG as well as matrix- degrading MMP1 and MMP3. In addition to its effects on fibrosis- related genes, CYR- 61

overexpression led to a significant increase in angiogenic factors, such as VEGF and FGF2, in dcSSc fibroblasts

(Table 2).At the protein level, overexpression of CYR- 61 in dcSSc

fibroblasts resulted in the release of significant amounts of CYR- 61, MMP- 1, MMP- 3, and VEGF into the culture media (Figure 1E), suggesting that CYR- 61 can act in both an auto-crine and a paracrine manner. To further confirm the role of CYR- 61 in dermal fibroblasts, we knocked down CYR- 61 in normal fibroblasts and showed that down- regulation of CYR- 61 led to a profibrotic phenotype, as indicated by elevated levels of COL1A1 and ACTA2 and decreased levels of MMP1 and MMP3 (Supplementary Table 2, available on the Arthritis & Rheumatology

Figure 1. Antifibrotic role of cysteine- rich angiogenic inducer 61 (CYR- 61) in dermal fibroblasts from patients with diffuse cutaneous systemic sclerosis (dcSSc). A, Expression of CYR- 61 in plasma from healthy controls (normal), all SSc patients, patients with limited cutaneous SSc (lcSSc), and patients with dcSSc. There were no differences among the groups. B, Levels of CYR61 mRNA in healthy controls, all patients with SSc, patients with lcSSc, and patients with dcSSc. CYR61 mRNA expression was lower in both dcSSc and lcSSc dermal fibroblasts than in normal fibroblasts. C, CYR- 61 levels in control and dcSSc fibroblast culture media. There was no difference between the 2 groups. D, Western blot (WB) (left) and Western blot band quantification (right) showing similar CYR- 61 protein levels in normal and dcSSc dermal fibroblasts. E, Significant increases in CYR- 61, matrix metalloproteinase 1 (MMP- 1), MMP- 3, and vascular endothelial growth factor (VEGF) in the culture media after overexpression of CYR- 61 protein in dcSSc dermal fibroblasts. F, Illustration (left) and quantification (right) of wound area (in pixels) in SSc dermal fibroblasts after overexpression of CYR- 61 using a CYR61 vector. At 48 hours the wound area was significantly wider in cells transfected with CYR61 vector than in cells transfected with control vector. G, Photograph (left) and quantification (right) of the gel area after overexpression of CYR- 61 in dcSSc fibroblasts in a gel contraction assay. The gel area was increased in cells transfected with CYR61 vector. In A–D, symbols represent individual subjects; horizontal lines and error bars show the mean ± SD. In G, bars show the mean ± SD.

TSOU ET AL 1354       |

web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40890/ abstract).

To examine the antifibrotic effect of CYR- 61 in SSc fibro-blasts, we performed 2 functional assays: a scratch wound assay and a gel contraction assay. Overexpression of CYR- 61 resulted in slower migration of dcSSc fibroblasts (Figure  1F) and reduced gel contraction (Figure  1G). In addition, CYR- 61 knockdown in normal dermal fibroblasts resulted in increased gel contraction (Supplementary Figure 1, available on the Arthri-tis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40890/ abstract), providing further evidence of the antifibrotic properties of CYR- 61 in SSc.

CYR- 61 induces senescence in SSc fibroblasts. After confirming the antifibrotic functional role of CYR- 61 in dcSSc fibroblasts, we examined the possible mechanism involved. We hypothesized that CYR- 61–overexpressing SSc fibroblasts are converted from extracellular matrix (ECM)–producing myofi-broblasts into ECM- degrading senescent cells, since CYR- 61 has been shown to trigger senescence by increasing reactive oxygen species (ROS) and activating the p38 pathway through α6β1 integrin and HSPGs (8,10,17). We first examined whether CYR- 61 affects cell proliferation, using the proliferation marker Ki- 67. As shown in Figure  2A, dcSSc fibroblasts transfected with CYR61 showed a significant reduction in Ki- 67 staining, indicating decreased proliferation. We then measured super-oxide production after CYR- 61 overexpression. We found that superoxide levels increased after 24 hours of CYR- 61 over-expression, reaching a peak after 48 hours (Figure 2B). At 72 hours after transfection there was no difference in superoxide levels between control- transfected and CYR- 61–transfected cells. These results suggest that superoxide production is an early event in the CYR- 61–triggered senescence pathway. We then measured p21 expression, which is a marker for cell senescence. Both p21 mRNA and protein levels were signif-

icantly elevated in CYR- 61–overexpressing dcSSc fibroblasts (Figures 2C–E).

To further dissect the mechanism of CYR- 61–mediated fibro-blast senescence, we examined whether different senescence pathways were involved. Since we observed an increase in ROS production (Figure  2B), and ROS can activate stress- activated kinases such as MAPKs and downstream p38 MAPK (18), we first determined whether p38 MAPK was activated after CYR- 61 overexpression in dcSSc dermal fibroblasts. Increased levels of phospho–p38 MAPK were indeed observed in CYR- 61–overex-pressing cells (Figure 2E). Because p38 MAPK activation results in increased transcriptional activity of p53 as well as up- regulation of p21 and p16, we further examined their expression in CYR- 61–overexpressing cells. Similar to phospho–p38 MAPK, increased accumulation of p53, p21, and p16 was also observed in fibro-blasts transfected with the CYR61 vector compared to those transfected with control vector (Figure 2E). Since all triggers and signaling cascades involved in cell senescence converge on the hypophosphorylated form of pRB, we showed that overexpres-sion of CYR- 61 in dcSSc dermal fibroblasts led to a decrease in pRB phosphorylation (Figure 2E). Cell senescence after CYR- 61 overexpression was also confirmed by β- galactosidase staining (Supplementary Figure 2, available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40890/ abstract). These results indicate that the p38 MAPK, p53, and p16/phospho- pRB pathways contribute to CYR- 61–induced fibroblast senescence.

CYR- 61 affects the TGFβ pathway in SSc fibroblasts. Because TGFβ is the most prominent growth factor in driving fibrosis and collagen production in SSc, we next investigated the effects of CYR- 61 on TGFβ signaling in control- transfected and CYR61- transfected dcSSc dermal fibroblasts. We found that cells overexpressing CYR- 61 showed attenuated phosphorylation of TGFβRII and Smad2/3 and reduced levels of TGFβ. The expres-sion levels of total TGFβRII and Smad2/3 did not change after CYR- 61 overexpression (Figure 2E).

Characterization of proangiogenic mechanisms of CYR- 61 in SSc ECs. We previously showed that CYR61 was overexpressed after HDAC- 5 knockdown in ECs, and that it mediated the antiangiogenic effects of HDAC- 5 in dcSSc ECs (3). To further confirm its role in SSc angiogenesis, we overexpressed CYR- 61 in dcSSc ECs and showed that this led to increased angiogenic ability of SSc ECs (3). In the pres-ent study we further dissected the mechanisms involved in the proangiogenic properties of CYR- 61 in SSc. We first deter-mined whether CYR61 expression was altered in dcSSc ECs compared to normal ECs and found that CYR61 expression was lower in dcSSc ECs (Figure  3A). We hypothesized that overexpression of CYR- 61 in SSc ECs leads to increased secretion of CYR- 61 and other proangiogenic factors. We

Table 2. Expression levels of mRNA for selected target genes after CYR- 61 overexpression in dcSSc dermal fibroblasts*

GeneFold change versus control-

transfected cells P

CYR61 213 ± 233 0.002COL1A1 0.73 ± 0.12 0.002ACTA2 0.75 ± 0.34 0.04PPARG 1.34 ± 0.49 0.01MMP1 3.70 ± 5.32 0.002MMP3 2.40 ± 1.62 0.004FGF2 1.42 ± 0.53 0.01VEGF 1.45 ± 0.98 0.03

* Values are the mean ± SD fold change in expression in cells trans-fected with cysteine- rich angiogenic inducer 61 (CYR- 61) versus control- transfected cells after transfection for 48 hours (n = 10 pa-tients with diffuse cutaneous systemic sclerosis [dcSSc]).

CYR- 61 IS ANTIFIBROTIC AND PROANGIOGENIC IN SSc |      1355

focused on VEGF since it is reported to be a CYR- 61–regu-lated gene (19). At the mRNA level, CYR- 61 overexpression resulted in significant up- regulation of VEGF while FGF2, another proangiogenic mediator, remained unchanged (Fig-ure 3B). We also measured secreted CYR- 61 and VEGF in the culture media in control and CYR- 61–overexpressing dcSSc ECs, and showed that overexpression of CYR- 61 in dcSSc ECs increased both CYR- 61 and VEGF secretion (Figures 3C and D).

To determine whether the receptor for CYR- 61, αvβ3, is pres-ent on ECs, we measured the mRNA levels in both normal and dcSSc ECs. There was a significant increase in ITGAV and ITGB3 expression in dcSSc ECs compared to normal cells (Figures 3E and F). In contrast, ITGAV and ITGB3 levels did not change in CYR- 61–overexpressing ECs (Supplementary Figure 3).

Since we established that the receptor for CYR- 61, αvβ3, is present on SSc ECs, and that overexpression of CYR- 61 leads to increased CYR- 61 and VEGF secretion, we then delineated the cellular events that were involved. CYR- 61 was previously shown to promote angiogenesis in ECs by activat-ing the AMPK pathway and the AKT/eNOS/NO pathway, both via αvβ3 (20,21). Therefore, we first measured mRNA levels of

NOS3 (encoding eNOS) and found that it was up- regulated in CYR- 61–overexpressing dcSSc ECs (Figure  3B). At the protein level the expression of eNOS was variable; however, when we measured phospho- eNOS, it was clear that CYR- 61 overexpression led to eNOS activation (Figure 3G). To exam-ine whether the eNOS–NO cascade is involved in CYR- 61–mediated angiogenesis, metabolites of NO were measured in culture media after control and CYR- 61 overexpression. As expected, activation of eNOS led to a significant increase in NO metabolites in CYR- 61–overexpressing cells (Figure 3H), suggesting that this pathway is involved in CYR- 61–medi-ated angiogenesis in SSc ECs. We then examined whether the AMPK and AKT pathways were activated. As shown in Figure  3G, enhanced phosphorylation of AMPK and AKT were observed in CYR- 61–overexpressing dcSSc ECs. Taken together, these results suggest that CYR- 61 overexpression in dcSSc ECs leads to increased excretion of CYR- 61, which, through binding to αvβ3, activates the AMPK/AKT/eNOS path-ways to promote angiogenesis.

To further confirm the involvement of αvβ3 in CYR- 61–medi-ated angiogenesis in SSc ECs, we performed Matrigel tube for-mation assays using CYR- 61–overexpressing dcSSc ECs in

Figure 2. CYR- 61–induced senescence in dcSSc dermal fibroblasts. A–D, Decreased cell proliferation after overexpression of CYR- 61 in dcSSc dermal fibroblasts, as shown by Ki- 67 staining (A), an increase in superoxide production that peaked at 48 hours (B), increased p21 (CDKN1A) mRNA levels (C), and increased p21 protein levels (D). In the left panels of A and D and in B, scale bars = 10 μm. In the right panels of A and D and in C, bars show the mean ± SD. E, Western blot showing activation of senescence pathways and impairment of the transforming growth factor β (TGFβ) pathway in CYR- 61–transfected dermal fibroblasts from 4 patients with dcSSc. The senescence pathways that were activated by CYR- 61 in dcSSc dermal fibroblasts included p53, p38, p21, and p16, ultimately leading to hypophosphorylation of retinoblastoma protein (pRB). In addition, the antifibrotic effect of CYR- 61 was mediated by inactivation of the TGFβ pathway, as indicated by reduced levels of phosphorylated TGFβ type II receptor and Smad2/3. See Figure 1 for other definitions.

TSOU ET AL 1356       |

the presence or absence of blocking antibodies for αvβ3. The proangiogenic property of CYR- 61 was blocked by the pres-ence of the αvβ3 antibodies compared to the IgG isotype con-trol (Figure 4A). In addition, inhibitors of AMPK (compound c) or AKT (LY294002) pathways blocked the proangiogenic activities of CYR- 61 (Figure 4B).

DISCUSSION

In this study, we examined the effect of CYR- 61 on both SSc fibroblasts and ECs. As illustrated in Figure 4C, overexpressing CYR- 61 in SSc fibroblasts attenuates fibrogenesis by inducing cellular senescence and impairing the TGFβ pathway. We showed that overexpressing CYR- 61 converted SSc fibroblasts from ECM- producing myofibroblasts into ECM- degrading senescent cells. This finding is supported by a reduction in fibrotic markers and an increase in matrix- degrading MMPs after overexpression of CYR- 61 in SSc fibroblasts. In addition, the myofibroblast phe-notype was inhibited, as shown by functional assays. We also demonstrated that markers for cell proliferation were reduced,

key mediators for cell senescence were elevated, and the activa-tion of TGFβRII and downstream Smad pathways were inhibited. We believe this study provides sufficient evidence to support the notion that CYR- 61 is indeed antifibrotic in SSc fibroblasts.

CYR- 61 has been studied extensively in the skin, since it is a negative regulator of collagen homeostasis in fibroblasts, and is substantially elevated in aged and senescent human skin (22,23). CYR- 61 has been shown to trigger senescence by increas-ing ROS and activating the p38 pathway through α6β1 integrin and HSPGs (8,10,17). In addition, it down- regulates TGFβRII and thereby impairs TGFβ responsiveness in fibroblasts (17,24). Although CYR- 61 blockade leads to improvement of fibrosis in lung and kidney fibrosis models (25,26), CYR- 61 has been shown to possess antifibrotic properties in animal models of cutaneous wound healing as well as liver fibrosis (8,10,17). We were able to recapitulate the involvement of the p38 MAPK pathway in cell senescence in CYR- 61–overexpressing SSc fibroblasts. However, instead of down- regulation of TGFβRII, we showed decreased levels of TGFβ, possibly leading to inactivation of TGFβRII and its downstream Smad2/3 pathway in these cells.

Figure 3. CYR- 61 promotes angiogenesis in endothelial cells (ECs) from patients with dcSSc through αvβ3 integrin, endothelial cell nitric oxide synthase (eNOS), AMP- activated protein kinase (AMPK), and AKT. A, Significantly lower CYR61 mRNA levels in dcSSc ECs than in normal ECs. B, Significant increases in CYR61, VEGF, and NOS3 after overexpression of CYR- 61 in dcSSc ECs. C and D, Significant elevation of CYR- 61 (C) and VEGF (D) levels in culture media from CYR- 61–overexpressing dcSSc ECs. E and F, Elevated ITGAV (E) and ITGB3 (F) mRNA expression levels in dcSSc ECs compared to normal ECs. G, Western blot showing significant elevation of phosphorylated eNOS, AKT, and AMPK in CYR61-transfected dcSSc ECs. H, Increased levels of nitrates and nitrites (metabolites of nitric oxide) in culture media collected from CYR61-transfected dcSSc ECs. In A, B, E, F, and H, bars show the mean ± SD. See Figure 1 for other definitions.

CYR- 61 IS ANTIFIBROTIC AND PROANGIOGENIC IN SSc |      1357

Interestingly, another member of the CCN family, CCN2/CTGF, is also involved in fibrosis. Although CYR- 61 and CTGF share similar structures, receptors, and downstream signaling pathways, the subtle differences in their sequences alter the surface area and charge, thereby affecting their interactions with binding partners making them functionally unique (27). It is suggested that in wound healing, CYR- 61 and CTGF act as yin and yang; CTGF promotes myofibroblast proliferation and matrix deposition to provide tissue integrity and promote wound repair when wounding occurs, while CYR- 61 promotes wound resolution by inducing myofibroblast senescence and matrix degradation (8). CTGF, which promotes fibrosis, is elevated in

SSc patients and plays key roles in myofibroblast transforma-tion through interaction with the TGFβ pathway (28–30). The imbalance between CTGF and CYR- 61 could play a major role in promoting fibrosis in SSc.

It has been suggested that CYR- 61 activates VEGF pro-duction in fibroblasts through the α6β1/HSPG/ERK axis (31). We indeed found that CYR- 61–overexpressing SSc fibroblasts released elevated levels of VEGF (Figure  1E), prompting us to speculate that CYR- 61 can promote EC angiogenesis through fibroblasts (Figure 4C). This is possible, since cross- talk between ECs and fibroblasts in SSc has been documented (32). Although VEGF has a limited effect on SSc ECs (the so- called VEGF par-

Figure 4. A and B, Inhibition of the proangiogenic property of CYR- 61 by blockade of αvβ3 (A) or by inhibition of AMP- activated protein kinase (AMPK) or AKT (B). Photomicrographs (top panels) and quantification (bottom panels) of Matrigel tube formation assays using CYR- 61–overexpressing dcSSc endothelial cells (ECs) are shown. The proangiogenic activity of CYR- 61 in promoting tube formation in dcSSc ECs was inhibited by αvβ3 blocking antibodies (ab). Pretreating the dcSSc ECs with inhibitors for AMPK (compound c [com C]) or AKT (LY294002) reduced tube formation significantly after CYR- 61 overexpression. Angiogenesis was shown by counting the number of nodes or the number of tubes, using ImageJ software. In top panels, original magnification × 4; in bottom panels, bars show the mean ± SD. C, CYR- 61 overexpression in dcSSc fibroblasts. CYR- 61 overexpression results in increased release of CYR- 61 that binds to α6β1 integrin/heparan sulfate proteoglycan (HSPG). CYR- 61 exerts antifibrotic properties by two mechanisms: 1) inducing senescence by reactive oxygen species (ROS) production, thereby activating p38/p16 or p38/p53/p21 pathways, and 2) attenuating the transforming growth factor β (TGFβ) pathway by inactivating the TGFβ/TGFβ receptor type II (TGFβRII) pathway. CYR- 61 can also increase VEGF release from fibroblasts, which may increase the angiogenic potential of ECs. Studies have suggested that CYR- 61 promotes VEGF production through activating the α6β1 integrin/HSPG–ERK pathway. D, CYR- 61 overexpression in dcSSc ECs. CYR- 61 overexpression results in increased CYR- 61 excretion, leading to improved angiogenesis by binding αvβ3 receptors and signaling through the AMPK/AKT/nitric oxide (NO) pathways. CYR- 61 can also increase the expression and excretion of proangiogenic VEGF, possibly through the AKT and AMPK pathways. The proangiogenic effect of VEGF is known to be mediated through the AKT and its downstream NO signaling pathway. eNOS = endothelial cell nitric oxide synthase (see Figure 1 for other definitions).

TSOU ET AL 1358       |

adox), we showed previously that these cells do respond to higher levels of VEGF (33), which strengthens the argument that overexpressing CYR- 61 in fibroblasts might increase the angio-genic potential of SSc ECs.

In addition to the effect of CYR- 61 on fibroblasts, its effect on SSc ECs was also examined in this study (Figure 4D). CYR- 61 overexpression in SSc ECs led to increased excretion of CYR- 61, which through binding to αvβ3, activated the AMPK/AKT pathways to promote angiogenesis. CYR- 61 also acti-vated expression of VEGF to further increase its proangiogenic potential. Both CYR- 61 and proangiogenic factors can act in an autocrine or a paracrine manner. The involvement of αvβ3 and the AKT and AMPK pathways in CYR- 61–mediated angio-genesis was supported by the findings of our experiments with blocking antibodies and inhibitors. Our results are consistent with those of previous studies suggesting the involvement of the AMPK pathway and the AKT/eNOS/NO pathway in CYR- 61–mediated angiogenesis in ECs (20,21). The ability of CYR- 61 to regulate VEGF was also documented (19). In addition, its angiogenic activity was observed in in vivo rat cornea models and ischemic rabbit hind limb models (20,21,34–38). Since CYR- 61 is also critical for EC survival (6,39) and increased apoptosis has been reported in SSc ECs (40), lower levels of CYR- 61 in SSc ECs could contribute to both the impaired angiogenesis and the increased apoptosis seen in these cells.

Inconsistent results have been reported for CYR- 61 serum levels in SSc (41,42). Similar to what we observed, Saigusa et  al (41) showed that serum CYR- 61 levels were compara-ble among healthy controls, patients with dcSSc, and patients with lcSSc (41). They also showed that CYR- 61 was down- regulated in dermal small blood vessels of SSc patients com-pared to controls, which is consistent with what we observed in this study (41).

We previously showed that CYR- 61 promotes EC tube for-mation in dcSSc ECs (3). In this study, we compiled evidence to indicate that CYR- 61 is indeed proangiogenic and antifibrotic in SSc ECs and fibroblasts, respectively. Therefore, our data sug-gest that up- regulating CYR- 61 in SSc might be of therapeutic potential in this disease. CYR- 61 attenuated two key patho-logic mechanisms in SSc: impaired angiogenesis and increased fibrosis. CYR- 61 also modulates immune responses through macrophages (43,44), and the possibility that CYR- 61 inhibits macrophage- driven fibrosis by affecting macrophage polari-zation and monocyte/macrophage migration warrants further investigation.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti-cally for important intellectual content, and all authors approved the final version to be published. Dr. Sawalha had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Tsou, Sawalha.Acquisition of data. Tsou, Khanna.Analysis and interpretation of data. Tsou, Sawalha.

REFERENCES 1. Barnes J, Mayes MD. Epidemiology of systemic sclerosis: incidence,

prevalence, survival, risk factors, malignancy, and environmental triggers. Curr Opin Rheumatol 2012;24:165–70.

2. Tsou PS, Sawalha AH. Unfolding the pathogenesis of scleroderma through genomics and epigenomics. J Autoimmun 2017;83: 73–94.

3. Tsou PS, Wren JD, Amin MA, Schiopu E, Fox DA, Khanna D, et al. Histone deacetylase 5 is overexpressed in scleroderma endothelial cells and impairs angiogenesis via repression of proangiogenic factors. Arthritis Rheumatol 2016;68:2975–85.

4. Krupska I, Bruford EA, Chaqour B. Eyeing the Cyr61/CTGF/NOV (CCN) group of genes in development and diseases: highlights of their structural likenesses and functional dissimilarities. Hum Genomics 2015;9:24.

5. Grote K, Salguero G, Ballmaier M, Dangers M, Drexler H, Schieffer B. The angiogenic factor CCN1 promotes adhesion and migration of circulating CD34+ progenitor cells: potential role in angiogenesis and endothelial regeneration. Blood 2007;110:877–85.

6. Leu SJ, Lam SC, Lau LF. Pro- angiogenic activities of CYR61 (CCN1) mediated through integrins αvβ3 and α6β1 in human umbilical vein endothelial cells. J Biol Chem 2002;277:46248–55.

7. Chen CC, Young JL, Monzon RI, Chen N, Todorovic V, Lau LF. Cytotoxicity of TNFα is regulated by integrin- mediated matrix signaling. EMBO J 2007;26:1257–67.

8. Jun JI, Lau LF. The matricellular protein CCN1 induces fibroblast senescence and restricts fibrosis in cutaneous wound healing. Nat Cell Biol 2010;12:676–85.

9. Kireeva ML, Mo FE, Yang GP, Lau LF. Cyr61, a product of a growth factor- inducible immediate- early gene, promotes cell proliferation, migration, and adhesion. Mol Cell Biol 1996;16:1326–34.

10. Kim KH, Chen CC, Monzon RI, Lau LF. Matricellular protein CCN1 promotes regression of liver fibrosis through induction of cellular senescence in hepatic myofibroblasts. Mol Cell Biol 2013;33: 2078–90.

11. Van den Hoogen F, Khanna D, Fransen J, Johnson SR, Baron M, Tyndall A, et al. 2013 classification criteria for systemic sclerosis: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum 2013;65:2737–47.

12. Tsou PS, Amin MA, Campbell P, Zakhem G, Balogh B, Edhayan G, et al. Activation of the thromboxane A2 receptor by 8- isoprostane inhibits the pro- angiogenic effect of vascular endothelial growth factor in scleroderma. J Invest Dermatol 2015;135:3153–62.

13. Tsou PS, Rabquer BJ, Ohara RA, Stinson WA, Campbell PL, Amin MA, et al. Scleroderma dermal microvascular endothelial cells exhibit defective response to pro- angiogenic chemokines. Rheumatology (Oxford) 2016;55:745–54.

14. He Y, Tsou PS, Khanna D, Sawalha AH. Methyl- CpG- binding protein 2 mediates antifibrotic effects in scleroderma fibroblasts. Ann Rheum Dis 2018;77:1208–18.

15. Schneider CA, Rasband WS, Eliceiri KW. NIH Image to ImageJ: 25 years of image analysis. Nat Methods 2012;9:671–5.

16. He Y, Tsou PS, Khanna D, Sawalha AH. Methyl- CpG- binding protein 2 mediates antifibrotic effects in scleroderma fibroblasts. Ann Rheum Dis 2018;77:1208–18.

17. Borkham-Kamphorst E, Schaffrath C, Van de Leur E, Haas U, Tihaa L, Meurer SK, et al. The anti- fibrotic effects of CCN1/CYR61 in primary portal myofibroblasts are mediated through induction

CYR- 61 IS ANTIFIBROTIC AND PROANGIOGENIC IN SSc |      1359

of reactive oxygen species resulting in cellular senescence, apoptosis and attenuated TGF- β signaling. Biochim Biophys Acta 2014;1843:902–14.

18. Muñoz-Espín D, Serrano M. Cellular senescence: from physiology to pathology. Nat Rev Mol Cell Biol 2014;15:482–96.

19. Mo FE, Muntean AG, Chen CC, Stolz DB, Watkins SC, Lau LF. CYR61 (CCN1) is essential for placental development and vascular integrity. Mol Cell Biol 2002;22:8709–20.

20. Park YS, Hwang S, Jin YM, Yu Y, Jung SA, Jung SC, et al. CCN1 secreted by tonsil- derived mesenchymal stem cells promotes endothelial cell angiogenesis via integrin αvβ3 and AMPK. J Cell Physiol 2015;230:140–9.

21. Hwang S, Lee HJ, Kim G, Won KJ, Park YS, Jo I. CCN1 acutely increases nitric oxide production via integrin αvβ3- Akt- S6K- phosphorylation of endothelial nitric oxide synthase at the serine 1177 signaling axis. Free Radic Biol Med 2015;89:229–40.

22. Quan T, Qin Z, Robichaud P, Voorhees JJ, Fisher GJ. CCN1 contributes to skin connective tissue aging by inducing age- associated secretory phenotype in human skin dermal fibroblasts. J Cell Commun Signal 2011;5:201–7.

23. Quan T, Qin Z, Voorhees JJ, Fisher GJ. Cysteine- rich protein 61 (CCN1) mediates replicative senescence- associated aberrant collagen homeostasis in human skin fibroblasts. J Cell Biochem 2012;113:3011–8.

24. Quan T, He T, Shao Y, Lin L, Kang S, Voorhees JJ, et al. Elevated cysteine- rich 61 mediates aberrant collagen homeostasis in chronologically aged and photoaged human skin. Am J Pathol 2006;169:482–90.

25. Kurundkar AR, Kurundkar D, Rangarajan S, Locy ML, Zhou Y, Liu RM, et al. The matricellular protein CCN1 enhances TGF- β1/SMAD3- dependent profibrotic signaling in fibroblasts and contributes to fibrogenic responses to lung injury. FASEB J 2016;30:2135–50.

26. Lai CF, Lin SL, Chiang WC, Chen YM, Wu VC, Young GH, et al. Blockade of cysteine- rich protein 61 attenuates renal inflammation and fibrosis after ischemic kidney injury. Am J Physiol Renal Physiol 2014;307:F581–92.

27. Holbourn KP, Acharya KR, Perbal B. The CCN family of proteins: structure–function relationships. Trends Biochem Sci 2008;33: 461–73.

28. Sato S, Nagaoka T, Hasegawa M, Tamatani T, Nakanishi T, Takigawa M, et al. Serum levels of connective tissue growth factor are elevated in patients with systemic sclerosis: association with extent of skin sclerosis and severity of pulmonary fibrosis. J Rheumatol 2000;27:149–54.

29. Holmes AM, Ponticos M, Shi-Wen X, Denton CP, Abraham DJ. Elevated CCN2 expression in scleroderma: a putative role for the TGFβ accessory receptors TGFβRIII and endoglin. J Cell Commun Signal 2011;5:173–7.

30. Abraham DJ, Shiwen X, Black CM, Sa S, Xu Y, Leask A. Tumor necrosis factor α suppresses the induction of connective tissue growth factor by transforming growth factor- β in normal and scleroderma fibroblasts. J Biol Chem 2000;275:15220–5.

31. Chen CC, Mo FE, Lau LF. The angiogenic factor Cyr61 activates a genetic program for wound healing in human skin fibroblasts. J Biol Chem 2001;276:47329–37.

32. Serratì S, Chillà A, Laurenzana A, Margheri F, Giannoni E, Magnelli L, et al. Systemic sclerosis endothelial cells recruit and activate dermal fibroblasts by induction of a connective tissue growth factor (CCN2)/transforming growth factor β–dependent mesenchymal- to- mesenchymal transition. Arthritis Rheum 2013;65:258–69.

33. Tsou PS, Rabquer BJ, Ohara RA, Stinson WA, Campbell PL, Amin MA, et al. Scleroderma dermal microvascular endothelial cells exhibit defective response to pro- angiogenic chemokines. Rheumatology (Oxford) 2016;55:745–54.

34. Babic AM, Kireeva ML, Kolesnikova TV, Lau LF. CYR61, a product of a growth factor- inducible immediate early gene, promotes angiogenesis and tumor growth. Proc Natl Acad Sci U S A 1998;95:6355–60.

35. Fataccioli V, Abergel V, Wingertsmann L, Neuville P, Spitz E, Adnot S, et al. Stimulation of angiogenesis by Cyr61 gene: a new therapeutic candidate. Hum Gene Ther 2002;13:1461–70.

36. Hinkel R, Trenkwalder T, Petersen B, Husada W, Gesenhues F, Lee S, et al. MRTF- A controls vessel growth and maturation by increasing the expression of CCN1 and CCN2. Nat Commun 2014;5:3970.

37. Rayssac A, Neveu C, Pucelle M, Van den Berghe L, Prado-Lourenco L, Arnal JF, et al. IRES- based vector coexpressing FGF2 and Cyr61 provides synergistic and safe therapeutics of lower limb ischemia. Mol Ther 2009;17:2010–9.

38. Chintala H, Krupska I, Yan L, Lau L, Grant M, Chaqour B. The matricellular protein CCN1 controls retinal angiogenesis by targeting VEGF, Src homology 2 domain phosphatase- 1 and Notch signaling. Development 2015;142:2364–74.

39. Di Y, Zhang Y, Hui L, Yang H, Yang Y, Wang A, et al. Cysteine- rich 61 RNA interference inhibits pathological angiogenesis via the phosphatidylinositol 3- kinase/Akt- vascular endothelial growth factor signaling pathway in endothelial cells. Mol Med Rep 2016;14:4321–7.

40. Wang Y, Kahaleh B. Epigenetic repression of bone morphogenetic protein receptor II expression in scleroderma. J Cell Mol Med 2013;17:1291–9.

41. Saigusa R, Asano Y, Taniguchi T, Yamashita T, Takahashi T, Ichimura Y, et al. A possible contribution of endothelial CCN1 downregulation due to Fli1 deficiency to the development of digital ulcers in systemic sclerosis. Exp Dermatol 2015;24:127–32.

42. Lin J, Li N, Chen H, Liu C, Yang B, Ou Q. Serum Cyr61 is associated with clinical disease activity and inflammation in patients with systemic lupus erythematosus. Medicine (Baltimore) 2015;94:e834.

43. Bai T, Chen CC, Lau LF. Matricellular protein CCN1 activates a proinflammatory genetic program in murine macrophages. J Immunol 2010;184:3223–32.

44. Rother M, Krohn S, Kania G, Vanhoutte D, Eisenreich A, Wang X, et al. Matricellular signaling molecule CCN1 attenuates experimental autoimmune myocarditis by acting as a novel immune cell migration modulator. Circulation 2010;122:2688–98.

1360

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1360–1370DOI 10.1002/art.40895 © 2019, American College of Rheumatology

The IgG2 Isotype of Anti–Transcription Intermediary Factor 1γ Autoantibodies Is a Biomarker of Cancer and Mortality in Adult DermatomyositisAudrey Aussy,1 Manuel Fréret,1 Laure Gallay,2 Didier Bessis,3 Thierry Vincent,4 Denis Jullien,5 Laurent Drouot,1 Fabienne Jouen,1 Pascal Joly,1 Isabelle Marie,1 Alain Meyer,6 Jean Sibilia,6 Brigitte Bader-Meunier,7 Eric Hachulla,8 Mohammed Hamidou,9 Sophie Huë,10 Jean-Luc Charuel,11 Nicole Fabien,2 Pierre-Julien Viailly,12 Yves Allenbach,13 Olivier Benveniste,13 Nadège Cordel,14 Olivier Boyer,1 and the OncoMyositis Study Group

Objective. Anti–transcription intermediary factor 1γ (anti- TIF1γ) antibodies are the main predictors of cancer in dermatomyositis (DM). Yet, a substantial proportion of anti- TIF1γ–positive DM patients do not develop cancer. This study was undertaken to identify biomarkers to better evaluate the risk of cancer and mortality in DM.

Methods. This multicenter study was conducted in adult anti- TIF1γ–positive DM patients from August 2013 to August 2017. Anti- TIF1γ autoantibody levels and IgG subclasses were identified using a newly developed quantita-tive immunoassay. Age, sex, DM signs and activity, malignancy, and creatine kinase (CK) level were recorded. Risk factors were determined by univariate and multivariate analysis according to a Cox proportional hazards regression model.

Results. Among the 51 adult patients enrolled (mean ± SD age 61 ± 17 years; ratio of men to women 0.65), 40 (78%) had cancer and 21 (41%) died, with a mean ± SD survival time of 10 ± 6 months. Detection of anti- TIF1γ IgG2 was significantly associated with mortality (P = 0.0011) and occurrence of cancer during follow- up (P < 0.0001), with a 100% positive predictive value for cancer when the mean fluorescence intensity of anti- TIF1γ IgG2 was >385. None of the patients developed cancer after 24 months of follow- up. Univariate survival analyses showed that mortality was also associated with age >60 years (P = 0.0003), active DM (P = 0.0042), cancer (P = 0.0031), male sex (P = 0.011), and CK level >1,084 units/liter (P = 0.005). Multivariate analysis revealed that age >60 years (P = 0.015) and the pres-ence of anti- TIF1γ IgG2 (P = 0.048) were independently associated with mortality.

Conclusion. Our findings indicate that anti- TIF1γ IgG2 is a potential new biomarker of cancer that should be helpful in identifying the risk of mortality in anti- TIF1γ–positive DM patients.

INTRODUCTION

Dermatomyositis (DM) is a rare form of autoimmune myop­athy that affects both children and adults (1,2). The outcome of

adult DM is variable, from benign to severe forms with lung or cardiac involvement and malignancies. Cancer has been identified as a major cause of mortality in adult DM and is generally discov­ered 1–2 years before or after the diagnosis of adult DM (2–7).

Supported by the SNFMI.1Audrey Aussy, MD, Manuel Fréret, PhD, Laurent Drouot, PhD, Fabienne

Jouen, MD, Pascal Joly, MD, PhD, Isabelle Marie, MD, PhD, Olivier Boyer, MD, PhD: Normandy University, University of Rouen, INSERM U1234, Rouen University Hospital, Rouen, France; 2Laure Gallay, MD, Nicole Fabien, MD: Édouard Herriot University Hospital, Lyon, France; 3Didier Bessis, MD, PhD: St. Eloi Hospital and Montpellier University Hospital, INSERM U1051, Montpellier, France; 4Thierry Vincent, MD, PhD: St. Eloi Hospital and Montpellier University Hospital, Montpellier, France; 5Denis Jullien, MD, PhD: Lyon University and Édouard Herriot University Hospital, Lyon, France; 6Alain Meyer, MD, PhD, Jean Sibilia, MD, PhD: Strasbourg University Hospital and Centre de Référence des Maladies Autoimmunes Rares, Strasbourg, France; 7Brigitte Bader-Meunier, MD, PhD: Necker University Hospital, AP-HP, Paris, France; 8Eric Hachulla, MD, PhD: European Reference Network on Connective Tissue and Musculoskeletal Diseases, University of Lille, Hospital Claude Huriez, Lille, France; 9Mohammed Hamidou, MD, PhD: Hôtel-Dieu and CHU de Nantes, Nantes, France; 10Sophie Huë, MD,

PhD: AP-HP, Henri Mondor Hospital, Créteil, France; 11Jean-Luc Charuel, MD, PhD: AP-HP, Pitié-Salpêtrière University Hospital, Paris, France; 12Pierre-Julien Viailly, MS: Normandy University, University of Rouen, INSERM U1245, Rouen, France; 13Yves Allenbach, MD, PhD, Olivier Benveniste, MD, PhD: AP-HP, Pitié-Salpêtrière University Hospital, Centre de Référence Maladies Neuro-Musculaires, DHU i2B, INSERM UMRS 974, Paris, France; 14Nadège Cordel, MD, PhD: Normandy University, University of Rouen, INSERM U1234, Rouen University Hospital, Rouen, France, and French West Indies University and Pointe-à-Pitre University Hospital, Pointe-à-Pitre, Guadeloupe, France.

Drs. Cordel and Boyer contributed equally to this work.No potential conflicts of interest relevant to this article were reported.Address correspondence to Olivier Boyer, MD, PhD, Laboratoire

d’immunologie, CHU de Rouen, 22 Boulevard Gambetta, F-76000 Rouen, France. E-mail: olivier.boyer@chu-rouen.fr.

Submitted for publication May 30, 2018; accepted in revised form March 14, 2019.

ANTI-­TIF1γ­IgG2­AND­CANCER­IN­DM­ |      1361

Numerous autoantibodies have been identified in DM, such as anti–Mi­ 2, anti–melanoma differentiation–associ­ated protein 5, anti–small ubiquitin­ like modifier activating enzyme, anti–transcription intermediary factor 1 (anti­ TIF1γ), and anti–nuclear matrix protein 2 (anti­ NXP2). Each of them defines a specific form of DM (8–13). Anti­ TIF1γ autoantibodies are present in 20–30% of adult DM cases and are strongly associated with cancer, with associated cancer occurring in 18–80% of patients (6,7,14,15). Anti­ TIF1γ–positive adult DM is characterized by a particularly brief time period between dis­ease onset and diagnosis of malignancy (6). Cancer types are comparable to those in the general population stratified by age and sex, but some rare cancers, i.e., gastric or thymus cancers, may also be found (6). Clinically, anti­ TIF1γ–positive adult DM is associated with severe cutaneous signs, moderate muscu­lar symptoms, and frequent dysphagia, while systemic features are uncommon (11,16). Other reported risk factors for cancer in adult DM, i.e., older age, male sex, and cutaneous necro­sis, have also been identified in anti­ TIF1γ–positive adult DM (10,16–19). Anti­ TIF1γ autoantibodies are also found in 30–40% of patients with juvenile DM, without any association with cancer (8,13,20).

Despite the strong overall association between anti­ TIF1γ autoantibodies and malignancy, a substantial proportion of anti­ TIF1γ–positive adult DM patients do not develop cancer (6,16). No biomarker has been identified to evaluate the individual risk of cancer for a patient with newly diagnosed anti­ TIF1γ–positive adult DM. Since cancer is the major mortality risk in adult DM and complicates the therapeutic management of DM, there is a need to identify tools that may help to evaluate the risk of cancer and mortality in anti­ TIF1γ–positive adult DM patients, and to decide how long and how much cancer screening should be proposed for an individual patient. The purpose of the present study was to identify such prognosis biomarkers and risk factors associated with mortality and cancer in anti­ TIF1γ–positive adult DM patients.

PATIENTS AND METHODS

Study design and population. This multicenter observa­tional study was conducted from August 2013 to August 2017. Sera obtained from patients screened for specific anti­ myositis autoantibodies at Rouen University Hospital’s immunology labo­ratory were used to identify anti­ TIF1γ–positive patients. Patients were seen at several departments of internal medicine and der­matology in university and general hospitals in France.

Inclusion criteria were age >18 years at the time of diagnosis, definite or probable DM including amyopathic DM according to European Neuromuscular Center criteria (1), anti­ TIF1γ autoan­tibody positivity as determined by addressable laser bead immu­noassay (ALBIA) for TIF1γ (>2 AU/ml), available clinical data, and a minimum follow­ up of 3 years for patients without known cancer at diagnosis.

Clinical, histologic, and standard biologic data, including age; sex; presence of malignancy within 3 years of the DM diagno­sis; delay between the DM diagnosis and occurrence of cancer; features of DM such as cutaneous lesions; muscular manifes­tations including the existence of repercussion on walking; pul­monary, cardiac, pharyngeal, articular, or digestive involvement; DM activity at last follow­ up; and creatine kinase (CK) level, were recorded on standardized forms completed by physicians. Due to retrospective inclusions, validated activity scores (21,22) were not systematically determined. Thus, DM remission was defined as the absence of DM symptoms without any treatment or with a low dosage of corticosteroids (i.e., 10 mg/day).

This study was approved by the institutional review board of Rouen University Hospital (number E2017­ 28). Members of the OncoMyositis Study Group are shown in Appendix A.

Anti-­TIF1γ­ autoantibody­ analysis­ and­ isotype­ determination. To detect and quantify anti­ TIF1γ autoanti­bodies, we developed an ALBIA (ALBIA­ TIF1γ) (see Supplementary Methods and Results, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract, for details), as previously described for other myositis­ specific autoantibodies (23). Briefly, the ALBIA­ TIF1γ consisted of coupling human recombinant TIF1γ protein (OriGene) to fluo­rescent beads (Bio­ Plex COOH­ microspheres; Bio­ Rad) using a Bio­ Plex amine coupling kit (Bio­ Rad) according to the manu­facturer’s protocol. To quantify anti­TIFγ autoantibodies in sera, TIF1γ­ coated beads were incubated with sera diluted 1:300, then incubated with anti­ IgG biotinylated secondary antibody (SouthernBiotech), and finally with streptavidin­ R–phycoerythrin (Qiagen). When indicated, isotypes of anti­ TIF1γ autoantibodies, i.e., anti­ IgG1, anti­ IgG2, anti­ IgG3, or anti­ IgG4, were deter­mined using the corresponding anti­ isotype secondary antibody. The mean fluorescence intensity (MFI) was next determined on a Bio­ Plex apparatus using Manager software version 4.0 (Bio­ Rad). When indicated, inhibition experiments were performed using another human recombinant TIF1γ protein, expressed by insect cells through a baculovirus system, that we called BV­ TIF1γ (Diarect). The diagnostic value of the assay was assessed using sera from patients with DM, patients with other autoimmune dis­eases according to established classification criteria, patients with polyclonal hypergammaglobulinemia, and healthy donors (see Supplementary Methods and Results) (1,24–26).

Anti-­TIF1γ­ autoantibody­ immunofluorescence­ analysis. Indirect immunofluorescence (IIF) was performed on HEp­ 2000 cells (Immuno Concepts). Sera were tested in phos­phate buffered saline at titration dilutions of 1:10 to 1:1,280, using a fluorescein isothiocyanate–coupled human anti­ IgG, anti­ IgG1, anti­ IgG2, anti­ IgG3, or anti­ IgG4 antibody (Southern­Biotech). Specific inhibitions of experiments were performed by adding 60 μg/ml of BV­ TIF1γ for 2 hours before performing IIF.

AUSSY ET AL 1362       |

Anti-­TIF1γ­ autoantibody­ detection­ by­ line­ blot­assay. EuroLine autoimmune inflammatory myopathies 16AG (Euroimmun) was used to qualitatively detect the presence of anti­ TIF1γ autoantibodies and other myositis autoantibodies.

Statistical­ analysis. The sensitivity and specificity of ALBIA­ TIF1γ were evaluated by computing a receiver operat­ing characteristic (ROC) curve. Quantitative data are expressed as the mean ± SD and median. Qualitative data are expressed

Figure 1. Validation of addressable laser bead immunoassay (ALBIA) for anti–transcription intermediary factor 1γ (anti­ TIF1γ) and determination of isotypes of anti­ TIF1γ autoantibodies in patients with dermatomyositis (DM). A, Indirect immunofluorescence on HEp­ 2000 cells in sera determined to be anti­ TIF1γ positive by ALBIA­ TIF1γ (see Supplementary Methods and Results and Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract, for details). Three representative anti­ TIF1γ–positive sera were not preincubated (top) and 3 representative anti­ TIF1γ–positive sera were preincubated with free human recombinant baculovirus TIFγ protein (bottom). A finely speckled antinuclear reactivity was typically seen. The specificity of this reaction was demonstrated by inhibition of nuclear fluorescence in the preincubated samples. Original magnification × 400. B, Validation of ALBIA­ TIF1γ using a commercial line blot immunoassay. Results for 5 representative anti­ TIF1γ–positive sera from patients with adult DM are shown. Sera from 1 anti–signal recognition particle (anti­ SRP)–positive patient and 1 healthy donor were used as controls. NXP­ 2 = anti–nuclear matrix protein 2; MDA­ 5 = melanoma differentiation–associated protein; SAE = small ubiquitin­ like modifier activating enzyme. C, Proportion of juvenile DM (JDM) patients and adult DM patients with each anti­ TIF1γ autoantibody isotype, determined by ALBIA­ TIF1γ using isotype­ specific secondary antibody. IgG2 was very rare in patients with juvenile DM, while IgG4 was not found in patients with adult DM. D, Anti­ TIF1γ isotype–specific fluorescence pattern on HEp­ 2000 cells revealed by anti­ IgG2 or anti­ IgG4 secondary antibodies (Ab) in sera from an anti­ TIF1γ IgG2­ negative IgG4­ positive patient with juvenile DM and an anti­ TIF1γ IgG2­ positive IgG4­ negative patient with adult DM. Anti­ TIF1γ positivity was determined by ALBIA­ TIF1γ. Original magnification × 400.

ANTI-­TIF1γ­IgG2­AND­CANCER­IN­DM­ |      1363

as percentages relative to the number of patients with available information. All recorded data and biologic characteristics of anti­ TIF1γ autoantibodies were considered in investigating risk factors for both mortality and cancer in adult patients.

To investigate risk factors for mortality, multivariate Cox and log rank tests (survival R package version 2.37.7) were used to assess differences in survival rates calculated by Kaplan­ Meier estimates. The survival period was calculated from the diagnosis of DM until death or the last follow­ up visit. Only variables with less than 10% missing data and P values less than 0.15 in univariate survival analysis were included in multivariate analysis.

To investigate factors associated with cancer, data were compared between the group of patients with cancer and the cancer­ free group in univariate analysis. Statistical differences were determined using Student’s t­ test or Fisher’s exact test when appropriate. For the subgroup of patients who developed cancer after the diagnosis of DM, a Kaplan­ Meier curve was generated to estimate the cumulative probability of acquiring associated malig­nancy during the follow­ up period.

Complete case analyses were performed, disregarding cases with partially missing data. P values less than 0.05 were consid­ered significant. Statistical analysis was performed using Graph­Pad Prism software, version 7.0 and R software, version 3.4.3.

RESULTS

Development­of­ALBIA-­TIF1γ. In the absence of an avail­able quantitative assay for anti­TIF1γ detection that could be used in routine practice, we developed ALBIA­ TIF1γ using a sera biocollection from 96 patients with DM, 99 healthy donors, 126 patients with autoimmune conditions, and 26 patients with pol­yclonal hypergammaglobulinemia. The purity of the TIF1γ protein used for this assay, and the analytical sensitivity, specificity, and reproducibility of ALBIA­ TIF1γ are described in the Supplementary Methods and Results and Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract.

The diagnostic value of ALBIA­ TIF1γ was evaluated through an ROC curve that established an optimal threshold of 2 AU/ml, yielding a high sensitivity of 96% and specificity of 99% (Supple­mentary Methods and Results and Supplementary Figure 2, avail­able on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract).

IIF on HEp­ 2000 cells was performed for 12 sera iden­tified as anti­ TIF1γ–positive by ALBIA­ TIF1γ and yielded (for 11 of the 12 samples) a nuclear, finely speckled pattern that excluded nucleoli. This reactivity was inhibited by prior incuba­tion with free BV­ TIF1γ (Figure 1A).

Forty­ three DM sera were analyzed in parallel by ALBIA­ TIF1γ and a commercial line blot assay (Figure 1B). The results of both assays were consistent (positive or negative for anti­

TIF1γ) in 39 of 43 cases; discrepancies were found for 3 samples that were positive by line blot assay and negative by ALBIA­ TIF1γ and for 1 sample that was negative by line blot assay and strongly positive by ALBIA­ TIF1γ (see Supplemen­tary Table 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract). Among the 4 sera with discrepant results, 1 was available in sufficient quantity to perform immunoprecipitation (IP) on recombinantly expressed TIF1γ from HEK 293 cells, which was negative, confirming the ALBIA result. Seven of 7 samples that were positive by both ALBIA and line blot assay and 8 of 8 samples that were negative by both ALBIA and line blot assay had consistent IP results (data not shown). Taken together, these data indicate that ALBIA­ TIF1γ has a very good analytical performance and diagnostic value for DM.

Clinical­ characteristics­of­ the­anti-­TIF1γ–positive­patients. We first identified patients who fulfilled the inclu­sion criteria from the cohort of patients who contributed sera used during the development phase of ALBIA­ TIF1γ, from August 2013 to August 2014. Then, when ALBIA­ TIF1γ was implemented in the laboratory, we analyzed the sera that we received for routine assays of myositis­ specific autoantibodies. Sera were obtained from patients seen at several depart­ments of internal medicine and dermatology in France (see Supplementary Figure 3, available on the Arthritis & Rheuma-tology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract). Hence, sera from 206 patients with adult DM were tested by ALBIA­ TIF1γ. Sixty­ five (32%) were positive for anti­ TIF1γ. Fourteen were excluded because clinical data were not available (n = 6) or the follow­ up period was too short (n = 8). Finally, 51 patients with adult DM (61% female) with a mean ± SD age at the time of diagnosis of 61 ± 17 years were included. All sera were sampled at the onset of disease or relapse, either before treatment was started or only a few days after treatment was started. The main clinical characteristics of the patients are summarized in Table 1. (Details for individual patients are provided in Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract.)

Cancer was diagnosed in 40 (78%) of the patients. Can­cer was diagnosed before DM in 11 (27.5%) of 40 cases with a median delay of −8 months (range −30, −1 months), after DM in 18 (45%) of 40 cases with a median delay of 3.5 months (range 0.5, 24 months), and concomitantly with DM in 11 (27.5%) of 40 cases. Cancer was not detected in any of the patients after 24 months of follow­ up. Cancers were solid tumors in 35 (87.5%) of 40 cases and hematologic malignancies in 5 (12.5%) of 40 cases. Twenty­ one (41%) of the patients died during follow­ up, with a mean ± SD survival time of 10 ± 6 months after diagnosis of DM (range 0.5, 28 months). All of the deaths occurred in patients with an associated cancer.

AUSSY ET AL 1364       |

Description­of­anti-­TIF1γ­autoantibodies. The mean ± SD anti­ TIF1γ IgG level in the cohort was 61 ± 70 AU/ml (median 35 AU/ml [range 5, 396]). Isotypes of anti­ TIF1γ were determined for the 51 patients with adult DM and for the 11 patients with anti­ TIF1γ–juvenile DM (Figure 1C). Notably, different distributions of isotypes were found in children and adults. The most abundant isotype in both populations was IgG1 (found in 56 of 62 patients). Most juvenile DM patients (10 of 11) harbored at least IgG1 and IgG4. In marked contrast, anti­ TIF1γ IgG4 was not found in any of the patients with adult DM, while the second most abundant isotype after IgG1 in patients with adult DM was IgG2 (found in 28 of 51 patients), alone or in association with IgG1 and/or IgG3. Results of IIF with anti­ isotype secondary antibody were consis­tent with these findings (Figure 1D). Twenty­ eight (55%) of 51 adult patients had IgG2, with or without other subclasses. Two children (18%) had IgG2 in association with other subclasses.

Identification­ of­ factors­ associated­with­ cancer­ in­anti- TIF1γ–positive­adult­DM­patients. The mean anti­ TIF1γ autoantibody level tended to be higher in the cancer group than in the cancer­ free group, but the difference was not significant (Figure 2A). Patients with an anti­ TIF1γ IgG level >44 AU/ml, which was the median level in the group with cancer, did not have a significantly higher risk of developing cancer (hazard ratio [HR] 1.9 [95% confidence interval (95% CI) 0.9–4.0]) (Figure 2B). Notably, the anti­ TIF1γ IgG2 level, expressed as the MFI, was significantly higher in the group with cancer, which had a median MFI of 385 (Figure 2C). The risk of cancer occurrence during follow­ up was higher in patients who were positive for anti­ TIF1γ IgG2 (HR 3.1 [95% CI 1.5–6.5]), particularly when the MFI of anti­ TIF1γ IgG2 was >385 (HR 3.9 [95% CI 1.7–8.8]) (Figure 2D). At the end of the follow­ up period, 90% of anti­ TIF1γ IgG2–positive patients developed cancer. This rate reached 100% in patients with anti­ TIF1γ IgG2 MFI levels >385, leading to a positive predictive value for cancer of 100% (95% CI 70–100). Interestingly, no cancer developed after 24 months in either anti­ TIF1γ IgG2–positive or anti­ TIF1γ IgG2–negative patients, suggesting that the frequency of testing for neoplasia in anti­ TIF1γ–positive adult DM patients could be decreased after 2 years of follow­ up. There was no cor­relation between the time between cancer and DM diagnoses and autoantibody titers (data not shown).

To identify other risk factors associated with cancer 3 years before or after the diagnosis of DM, we compared the cancer group (n = 40) to the cancer­ free group (n = 11). Univariate analysis showed that male sex (odds ratio [OR] 9.0 [95% CI 1.2–102.8]), the presence of anti­ TIF1γ IgG2 (OR 8.4 [95% CI 1.6–41.0]), age >60 years (OR 6.2 [95% CI 1.4–23.7]), and hypophonia (OR

infinite [95% CI 1.3–infinite) were associated with cancer (Table 2).

Identification­ of­ risk­ factors­ for­ mortality­ in­­anti-­TIF1γ–positive­ adult­ DM. To determine whether anti­ TIF1γ level impacted mortality, we analyzed survival in patients

Table 1. Baseline characteristics and disease activity of the anti­ TIF1γ–positive patients with adult DM (n = 51) at last follow­ up*

Epidemiologic and demographic characteristics

Sex, no. male/female 20/31Age, mean ± SD (median) years 61 ± 17 (63)Cancer 40/51 (78)Deaths 21/51 (41)Follow- up time, mean ± SD (median)

months45 ± 57 (24)

Autoantibody characteristicsTIF1γ IgG level, mean ± SD (median) AU/ml 61 ± 70 (35)IgG2 positive 28/51 (55)TIF1γ IgG2 level mean ± SD (median) MFI 1,241 ± 2,131

(190)Cutaneous signs 51/51 (100)

Heliotrope periorbital edema 39/51 (76)V- sign 34/50 (68)Shawl sign 15/48 (31)Gottron’s papules/sign 36/50 (72)Periungual erythema 38/49 (78)Poikiloderma 15/47 (32)Squamous rash of scalp 23/44 (52)Raynaud’s phenomenon 5/48 (10)Photosensitivity 12/49 (24)Mechanic’s hands 2/49 (4)Necrosis and/or ulcerations 15/50 (30)Calcinosis 2/50 (4)

Muscle involvement 51/51 (100)Muscle weakness 47/48 (98)Myalgia 36/44 (82)Repercussion on walking 35/44 (80)CK level, mean ± SD (median) units/liter 2,315 ± 3,130

(819)Diaphragm weakness 13/37 (35)Pharyngeal involvement 37/49 (76)Feeding tube needed† 14/37 (38)Hypophonia 13/49 (27)

Other signsAspiration pneumonia 16/41 (39)Interstitial lung disease 1/50 (2)Nonerosive arthritis 11/50 (22)Digestive signs 4/42 (10)Cardiac signs 6/48 (13)

DM activity at last follow- upPersistence of signs 31/49 (63)Complete remission‡ 18/49 (37)

* Except where indicated otherwise, values are the number of patients/number for whom data were available (%). Anti- TIF1γ = anti–transcription intermediary factor 1γ; MFI = mean fluorescence intensity; CK = creatine kinase. † Among patients with dysphagia. ‡ No signs of dermatomyositis (DM) were present, without treatment or with minimal treatment (corticosteroid dosage of <10 mg/day).

ANTI-­TIF1γ­IgG2­AND­CANCER­IN­DM­ |      1365

classified according to the above­ defined threshold of an anti­ TIF1γ IgG level of 44 AU/ml. Overall survival tended to be lower when anti­ TIF1γ IgG level was >44 AU/ml, but the difference was

not significant (HR 2.1 [95% CI 0.9–5.0]) (Figure 3A). We next eval­uated the impact of anti­ TIF1γ IgG2 positivity on mortality. The survival curve indicated a dramatically poorer overall survival rate

Figure 2. Predictive value of anti–transcription intermediary factor 1γ (anti­ TIF1γ) autoantibody level and IgG2 isotype for the occurrence of cancer in patients with adult dermatomyositis (DM). A and C, Anti­ TIF1γ IgG level (A) and anti­ TIF1γ IgG2 mean fluorescence intensity (MFI) (C) in patients with adult DM without cancer and those with cancer. Open symbols represent patients who were alive at the last follow­ up; solid symbols represent patients who died. Bars show the mean ± SD. Horizontal lines represent the median value in the cancer group (44 AU/ml in A and an MFI of 385 in C). B and D, Kaplan­ Meier curve of cancer occurrence after diagnosis of DM according to anti­ TIF1γ IgG antibody level (>44 AU/ml or <44 AU/ml) (B) and the presence or absence of anti­ TIF1γ IgG2 or anti­ TIF1γ IgG2 level (MFI >385 or MFI <385) (D). * = P < 0.05; **** = P < 0.0001, by Student’s t­ test in A and C and by log rank test in B and D. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40895/abstract.

Table 2. Parameters associated with cancer in anti­ TIF1γ–positive patients with adult DM*

Patients with cancer

(n = 40)

Patients without cancer (n = 11) OR (95% CI) P

Sex, no. male/female 19/21 1/10 9.0 (1.2–102.8) 0.034Age >60 years 28 (70) 3 (27) 6.2 (1.4–23.7) 0.015Death 21 (53) 0 (0) infinite (2.9–infinite) 0.0014Anti- TIF1γ IgG level >44 AU/ml 22 (55) 2 (18) 5.5 (1.1–27.1) 0.042Anti- TIF1γ IgG2 positivity 26 (65) 2 (18) 8.4 (1.6–41.0) 0.014Anti- TIF1γ IgG2 MFI level >385 20 (50) 0 (0) infinite (2.7–infinite) 0.0035Hypophonia† 13 (33) 0 (0) infinite (1.3–infinite) 0.045

* Except where indicated otherwise, values are the number (%). Anti- TIF1γ = anti–transcription interme-diary factor 1γ; DM = dermatomyositis; OR = odds ratio; 95% CI = 95% confidence interval; MFI = mean fluorescence intensity. † Data were available for 39 patients with cancer and 10 patients without cancer.

AUSSY ET AL 1366       |

in anti­ TIF1γ IgG2–positive patients (34%) than in anti­ TIF1γ IgG2–negative patients (86%) (HR 5.9 [95% CI 2.4, 14.1]) (Figures 3B and C).

Univariate survival analysis comparing deceased patients (n = 21) to patients alive at last follow­ up (n = 30) revealed that, in addition to anti­ TIF1γ IgG2 positivity, age >60 years (HR 8.9 [95% CI 3.8–21.1]), presence of cancer (HR 4.3 [95% CI 1.6–11.2]), muscle involvement with repercussion on walking (HR 3.8 [95% CI 1.3–11.0]), CK level >1,084 units/liter (HR 3.9 [95% CI 1.5–10.5]), male sex (HR 2.9 [95% CI 1.2–7.2]), and active DM (HR 6.4 [95% CI 2.5–16.6]) were also significantly associated with mortality (Table  3). The variables selected for multivariate analysis were age >60 years, presence of IgG2, male sex, can­cer, and anti­ TIF1γ level >44 AU/ml, which were the variables with P < 0.10 and missing data <10%. Cox proportional hazards regression analysis revealed that age >60 years (HR 6.6 [95% CI

1.4–30.3]) and anti­ TIF1γ IgG2 positivity (HR 3.7 [95% CI 1.1–

13.7]) were independently associated with mortality (Table 3).

Evolution­ of­ anti-­TIF1γ–positive­ DM. Among the cancer­ free group, 8 (73%) of 11 patients had persistent DM signs. Among patients with cancer, 18 (95%) of 19 living patients had cancer in remission, among whom 13 (72%) of 18 had com­plete remission of DM (Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40895/ abstract). Of the patients who died, 13 (87%) of 15 patients had active cancer and active DM at last follow­ up. Interestingly, for 4 patients with cancer for whom we were able to analyze a second serum sample during a period of remission of DM, sera were negative for anti­ TIF1γ autoanti­bodies by both line blot assay and ALBIA­ TIF1γ. Cancer was also in remission at that time.

Figure 3. Prognostic value of anti–transcription intermediary factor 1γ (anti­ TIF1γ) autoantibody level and IgG2 isotype for mortality in adult dermatomyositis (DM). Kaplan­ Meier curves of overall survival after diagnosis of DM according to A, the level of anti­ TIF1γ autoantibodies (>44 AU/ml or <44 AU/ml), B, the presence or absence of anti­ TIF1γ IgG2, and C, anti­ TIF1γ IgG2 level (mean fluorescence intensity [MFI] >385 or MFI <385) are shown. ** = P < 0.01, by log rank test. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.40895/abstract.

ANTI-­TIF1γ­IgG2­AND­CANCER­IN­DM­ |      1367

DISCUSSION

This study identified the IgG2 isotype of anti­ TIF1γ autoan­tibodies as a biomarker of mortality in anti­ TIF1γ–positive adult DM patients. It confirmed the now well­ documented associa­tion between anti­ TIF1γ autoantibodies and cancer in adult DM patients, and further revealed that no cancer occurred after a 2­ year period in this series (6,7,14,15,27–29). Despite the ret­rospective design of this study, selection biases were limited by the inclusion of patients from several departments of inter­nal medicine and dermatology in university and general hos­pitals in France, supporting the similarities of our series with data in the literature. Indeed, the 78% rate of cancer found in anti­ TIF1γ–positive patients in our study is consistent with the rate of 68–84% previously reported in the literature. The mean

age of the patients in our study was also consistent with that in previously published studies of anti­ TIF1γ–positive patients (6,14,27,30,31). We found an elevated rate of metastatic can­cer, consistent with the findings of Ogawa­ Momohara et al, who recently showed that among DM patients with cancer, those who were positive for anti­ TIF1γ had a high frequency of advanced cancer, higher than that in patients with other autoantibodies, such as anti–NXP­ 2 (32).

Our findings showed a high rate (41%) of death in anti­ TIF1γ–positive adult DM patients. Cancer was associated with mortality, and mortality occurred only in patients with associated cancer. Independently of anti­ TIF1γ status, cancer was pre­viously reported to be the main prognostic factor in adult DM (3,4,29). Despite this finding, half of the patients with cancer in our cohort survived during the follow­ up period. Survival analy­

Table 3. Clinical and biologic parameters associated with mortality in anti­ TIF1γ–positive patients with adult DM*

Total (n = 51)

Deceased (n = 21)

Alive at end point

(n = 30)

Univariate analysis Multivariate analysis

HR (95% CI) P HR (95% CI) P

Demographic characteristicsSex, no. male/female 20/31 13/8 7/23 2.9 (1.2–7.2) 0.011 1.3 (0.5–3.2) 0.6Age >60 years 31/51 (61) 19/21 (90) 12/30 (40) 8.9 (3.8–21.1) 0.0003 6.6 (1.4–30.3) 0.015Cancer 40/51 (78) 21/21 (100) 19/30 (63) 4.3 (1.6–11.2)† 0.0031 infinite (0–infinite) 0.9

Autoantibody characteristicsAnti- TIF1γ IgG level >44 AU/ml‡ 26/51 (51) 14/21 (67) 12/30 (40) 2.1 (0.9–5.0) 0.09 0.6 (0.2–1.7) 0.4Anti- TIF1γ IgG2 positivity 28/51 (55) 18/21 (86) 10/30 (33) 5.9 (2.4–14.1) 0.0011 3.7 (1.1–13.7) 0.048Anti- TIF1γ IgG2 MFI level >385‡ 20/51 (39) 14/21 (67) 6/30 (20) 3.9 (1.5–9.8) 0.0016 – –

Cutaneous signsNecrosis 15/50 (30) 8/21 (38) 7/29 (24) 1.5 (0.6–3.9) 0.3 – –

Muscle involvement Repercussion on walking 35/44 (80) 19/19 (100) 16/25 (64) 3.8 (1.3–11.0)† 0.015 – –CK >1,084 units/liter‡ 22/44 (50) 13/17 (76) 9/27 (33) 3.9 (1.5–10.5) 0.005 – –Diaphragm weakness 13/37 (35) 6/15 (40) 7/22 (32) 1.5 (0.5–4.4) 0.5 – –Pharyngeal involvement 37/49 (76) 18/20 (90) 19/29 (66) 3.5 (1.3–9.2) 0.08 – –Hypophonia 13/49 (27) 7/21 (33) 6/28 (21) 1.4 (0.5–3.7) 0.5 – –

Other signsAspiration pneumonia 16/41 (39) 11/18 (61) 5/23 (22) 2.3 (1.0–6.7) 0.040 – –Interstitial lung disease 1/50 (2) 1/21 (5) 0/29 (0) 3.6 (0.1–146.4) 0.2 – –Nonerosive arthritis 11/50 (22) 4/21 (19) 7/29 (24) 0.9 (0.3–2.4) 0.8 – –Digestive signs 4/42 (10) 3/19 (16) 1/23 (4) 2.3 (0.4–12.9) 0.2 – –Cardiac signs 6/48 (13) 3/20 (15) 3/28 (11) 1.3 (0.3–4.9) 0.7 – –

DM activity at last follow- upPersistence of signs 31/49 (63) 18/20 (90) 13/29 (45) 6.4 (2.5–16.6) 0.0042 – –Complete remission 18/49 (37) 2/20 (10) 16/29 (55)

* Except where indicated otherwise, values are the number of patients/number for whom data were available (%). Anti- TIF1γ = anti– transcription intermediary factor 1γ; DM = dermatomyositis; 95% CI = 95% confidence interval; MFI = mean fluorescence intensity; CK = creatine kinase. † Hazard ratios (HRs) were calculated using the Mantel- Haenszel method because the HR calculated using the log rank method was unde-fined. ‡ Cutoff values correspond to the median value observed in the cancer group.

AUSSY ET AL 1368       |

sis showed that the presence of IgG2 autoantibodies was inde­pendently associated with a higher risk of death in the first 2 years. Age >60 years, CK level, muscle involvement with reper­cussion on walking, and to a lesser extent, pharyngeal signs, were also associated with a higher risk of death, highlighting the existence of a subgroup of patients with major risk of mortality. Since a majority of anti­ TIF1γ–positive patients with adult DM have or will develop cancer, IgG2 appears to be an important biomarker to identify those patients with cancer who are at very high risk of mortality.

Interestingly, the presence of IgG2 subclass autoantibodies at the time of DM diagnosis was associated with a higher risk of cancer occurrence. An IgG2 MFI >385 was found exclusively in the group of patients who were diagnosed as having cancer during follow­ up. Age >60 years at diagnosis and male sex were also associated with the risk of cancer in our series, consistent with recent literature (7,16,29). In this context, detection of IgG2 seems to be helpful to predict cancer for patients without cancer at time of DM diagnosis and to identify patients with high risk of mortality among those with cancer.

The results of the present study were obtained using a newly developed immunoassay to quantify and characterize anti­ TIF1γ autoantibodies in accordance with our previous expertise (23,33). Immunoprecipitation is valuable for autoantibody discovery but less adapted to routine clinical biology, due to technical complex­ity and interpretation difficulties (7,14,27). Manufacturers’ line and dot strips for immunoblotting are easy to use but do not allow for quantification of autoantibodies (34,35). In this regard, the ALBIA­ TIF1γ assay is highly sensitive, specific, and suited to routine use.

Previous studies that sought to determine isotypes of autoan­tibodies in other autoimmune diseases mainly demonstrated a high prevalence of IgG1 and IgG4 (23,33,36,37). Some of them associate an IgG subclass of autoantibodies with clinical specific­ities; for example, anti–thyroid peroxidase autoantibodies IgG2/IgG4 and higher risk of hypothyroidism, anti–desmoglein 3 IgG2 and paraneoplastic pemphigus, anti–desmoglein 3 IgG1 and IgG4 and pemphigus vulgaris, or anti–ADAMTS­ 13 IgG4 and higher risk of relapsing disease (38–40). In this study, IgG1 autoantibodies were widely represented in both adult DM and juvenile DM, but IgG4 autoantibodies were found exclusively in juvenile DM. Overall IgG4 responses have traditionally been associated with chronic or repetitive immune responses, with, in some cases, a regulatory role (41). IgG4 levels may be influenced by corticosteroid therapy, as evidenced in sarcoidosis, for instance (42). In our series, juvenile DM sera were sampled at the time of diagnosis, before cortico­steroid treatment. Besides total IgG4 levels, some autoantibodies of the IgG4 isotype have been reported to be directly pathogenic, such as in myasthenia gravis and pemphigus (43,44). The specific role of anti­ TIF1γ IgG4 in juvenile DM evidenced herein may war­rant further investigation.

Juvenile DM and adult DM are associated with an interferon (IFN) signature in muscle and skin (45,46). The events triggering

this response are unknown but may be different in pediatric and adult patients. It is possible that viruses or re­ expression of endog­enous viral sequences are involved, or, in some adult patients, cancer itself. Modifications in the TRIM33 gene encoding TIF1γ in tumors have been reported in DM patients (47). Thus, one hypoth­esis is that different types of IFN triggers may elicit different types of TIF1γ­ specific immune responses (with different predominant IgG subclass) that may concur with DM pathogenesis. The duration of exposure to this trigger may also influence the type of response.

Some observations in this study suggest a paraneoplastic course of anti­ TIF1γ–positive DM. Indeed, a high proportion of patients had DM in remission after achieving cancer remission, whereas the majority of deceased patients had both active DM and active cancer at the time of death. Disappearance of both signs of DM and anti­ TIF1γ autoantibodies after the treatment of cancer in 4 patients was similar to previously reported clin­ical cases (48–50). These points may be of pathophysiologic importance. Indeed, overexpression of TIF1γ is involved in the oncogenesis of breast cancer and in a significant proportion of colorectal adenocarcinomas (51,52). The parallel evolution of anti­ TIF1γ autoantibodies and cancer corroborates the poten­tial role of tumors to promote an anti­ TIF1γ response that sec­ondarily deviates towards muscle and skin (30,53,54). As with anti­ Hu–positive autoantibody in encephalitis or anti­ POL3–pos­itive autoantibodies in systemic sclerosis (55,56), it is tempting to speculate that a tumor may harbor ectopically expressed or mutated TIF1γ antigens, leading to an antitumor response which secondarily extends to muscle and skin, by cross­ reactivity and/or epitope spreading (53). Since anti­ TIF1γ IgG2 identified patients with poorer prognosis, the presence of anti­ TIF1γ IgG2 autoantibodies may reflect a helper T cell response associated with an inefficient antitumor response, which warrants further study. On the contrary, disease evolution in adult patients with­out cancer resembles that of juvenile DM, raising the question of a different pathophysiologic process (57–59).

Taken together, the results of this study indicate that anti­ TIF1γ IgG2 autoantibody is a potential predictive marker of mor­tality and cancer, and support conducting a prospective study to justify prescribing their detection. Indeed, early detection of anti­ TIF1γ–positive IgG2 may represent a valuable marker that could be implemented in adult DM diagnosis in routine use. Moreover, in the absence of occurrence of cancer after 2 years, it is con­ceivable that the burden of cancer screening could be eased.

ACKNOWLEDGMENT

We are grateful to Nikki Sabourin­ Gibbs (Rouen University Hospital) for help in editing the manuscript.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti­cally for important intellectual content, and all authors approved the final

ANTI-­TIF1γ­IgG2­AND­CANCER­IN­DM­ |      1369

version to be published. Dr. Boyer had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Aussy, Drouot, Jouen, Viailly, Allenbach, Benveniste, Cordel, Boyer.Acquisition of data. Aussy, Fréret, Gallay, Bessis, Vincent, Jullien, Drouot, Joly, Marie, Meyer, Sibilia, Bader­ Meunier, Hachulla, Hamidou, Huë, Char­uel, Fabien, Allenbach, Benveniste, Cordel.Analysis and interpretation of data. Aussy, Fréret, Gallay, Drouot, Jouen, Viailly, Cordel, Boyer.

REFERENCES 1. Hoogendijk JE, Amato AA, Lecky BR, Choy EH, Lundberg IE, Rose

MR, et al. 119th ENMC international workshop: trial design in adult idiopathic inflammatory myopathies, with the exception of inclusion body myositis, 10–12 October 2003, Naarden, The Netherlands. Neuromuscul Disord 2004;14:337–45.

2. Iaccarino L, Ghirardello A, Bettio S, Zen M, Gatto M, Punzi L, et al. The clinical features, diagnosis and classification of dermatomyositis. J Autoimmun 2014;48–9:122–7.

3. Wakata N, Kurihara T, Saito E, Kinoshita M. Polymyositis and dermatomyositis associated with malignancy: a 30­ year retrospective study. Int J Dermatol 2002;41:729–34.

4. András C, Ponyi A, Constantin T, Csiki Z, Szekanecz E, Szodoray P, et al. Dermatomyositis and polymyositis associated with malignancy: a 21­ year retrospective study. J Rheumatol 2008;35:438–44.

5. Lundberg IE, Tjärnlund A, Bottai M, Werth VP, Pilkington C, de Visser M, et al. 2017 European League Against Rheumatism/American College of Rheumatology classification criteria for adult and juvenile idiopathic inflammatory myopathies and their major subgroups. Arthritis Rheumatol 2017;69:2271–82.

6. Hida A, Yamashita T, Hosono Y, Inoue M, Kaida K, Kadoya M, et al. Anti­ TIF1­ γ antibody and cancer­ associated myositis: a clinicohistopathologic study. Neurology 2016;87:299–308.

7. Fiorentino DF, Chung LS, Christopher­Stine L, Zaba L, Li S, Mammen AL, et al. Most patients with cancer­ associated dermatomyositis have antibodies to nuclear matrix protein NXP­ 2 or transcription intermediary factor 1γ. Arthritis Rheum 2013;65:2954–62.

8. Betteridge Z, McHugh N. Myositis­ specific autoantibodies: an important tool to support diagnosis of myositis. J Intern Med 2016;280:8–23.

9. Benveniste O, Stenzel W, Allenbach Y. Advances in serological diagnostics of inflammatory myopathies. Curr Opin Neurol 2016;29: 662–73.

10. Fujimoto M, Watanabe R, Ishitsuka Y, Okiyama N. Recent advances in dermatomyositis­ specific autoantibodies. Curr Opin Rheumatol 2016;28:636–44.

11. Best M, Jachiet M, Molinari N, Manna F, Girard C, Pallure V, et al. Distinctive cutaneous and systemic features associated with specific antimyositis antibodies in adults with dermatomyositis: a prospective multicentric study of 117 patients. J Eur Acad Dermatol Venereol 2018;32:1164–72.

12. Meyer A, Lannes B, Goetz J, Echaniz­Laguna A, Lipsker D, Arnaud L, et al. Inflammatory myopathies: a new landscape. Joint Bone Spine 2018;85:23–33.

13. Wolstencroft PW, Fiorentino DF. Dermatomyositis clinical and pathological phenotypes associated with myositis­ specific auto­antibodies. Curr Rheumatol Rep 2018;20:28.

14. Targoff IN, Mamyrova G, Trieu EP, Perurena O, Koneru B, O’Hanlon TP, et al., for the Childhood Myositis Heterogeneity and International Myositis Collaborative Study Groups. A novel autoantibody to a 155­ kd protein is associated with dermatomyositis. Arthritis Rheum 2006;54: 3682–9.

15. Trallero­Araguás E, Rodrigo­Pendás JÁ, Selva­O’Callaghan A, Martínez­Gómez X, Bosch X, Labrador­Horrillo M, et al. Usefulness of anti­ p155 autoantibody for diagnosing cancer­ associated dermatomyositis: a systematic review and meta­ analysis. Arthritis Rheum 2012;64:523–32.

16. Fiorentino DF, Kuo K, Chung L, Zaba L, Li S, Casciola­Rosen L. Distinctive cutaneous and systemic features associated with antitranscriptional intermediary factor­ 1γ antibodies in adults with dermatomyositis. J Am Acad Dermatol 2015;72:449–55.

17. Tiniakou E, Mammen AL. Idiopathic inflammatory myopathies and malignancy: a comprehensive review. Clin Rev Allergy Immunol 2017;52:20–33.

18. Burnouf M, Mahé E, Verpillat P, Descamps V, Lebrun­Vignes B, Picard­Dahan C, et al. Cutaneous necrosis is predictive of cancer in adult dermatomyositis. Ann Dermatol Venereol 2003;130:313–6.

19. Zahr ZA, Baer AN. Malignancy in myositis. Curr Rheumatol Rep 2011;13:208–15.

20. Rider LG, Nistala K. The juvenile idiopathic inflammatory myopathies: pathogenesis, clinical and autoantibody phenotypes, and outcomes. J Intern Med 2016;280:24–38.

21. Rider LG, Aggarwal R, Pistorio A, Bayat N, Erman B, Feldman BM, et al. 2016 American College of Rheumatology/European League Against Rheumatism criteria for minimal, moderate, and major clinical response in juvenile dermatomyositis: an International Myositis Assessment and Clinical Studies Group/Paediatric Rheumatology International Trials Organisation collaborative initiative. Arthritis Rheumatol 2017;69:911–23.

22. Anyanwu CO, Fiorentino DF, Chung L, Dzuong C, Wang Y, Okawa J, et al. Validation of the Cutaneous Dermatomyositis Disease Area and Severity Index: characterizing disease severity and assessing responsiveness to clinical change. Br J Dermatol 2015;173:969–74.

23. Drouot L, Allenbach Y, Jouen F, Charuel JL, Martinet J, Meyer A, et al. Exploring necrotizing autoimmune myopathies with a novel immunoassay for anti­ 3­ hydroxy­ 3­ methyl­ glutaryl­ CoA reductase autoantibodies. Arthritis Res Ther 2014;16:R39.

24. Tan EM, Cohen AS, Fries JF, Masi AT, McShane DJ, Rothfield NF, et al. The 1982 revised criteria for the classification of systemic lupus erythematosus. Arthritis Rheum 1982;25:1271–7.

25. Vitali C, Bombardieri S, Jonsson R, Moutsopoulos HM, Alexander EL, Carsons SE, et al. Classification criteria for Sjögren’s syndrome: a revised version of the European criteria proposed by the American­ European Consensus Group. Ann Rheum Dis 2002;61:554–8.

26. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24.

27. Kaji K, Fujimoto M, Hasegawa M, Kondo M, Saito Y, Komura K, et al. Identification of a novel autoantibody reactive with 155 and 140 kDa nuclear proteins in patients with dermatomyositis: an association with malignancy. Rheumatology (Oxford) 2007;46:25–8.

28. Hoshino K, Muro Y, Sugiura K, Tomita Y, Nakashima R, Mimori T. Anti­ MDA5 and anti­ TIF1­ γ antibodies have clinical significance for patients with dermatomyositis. Rheumatology (Oxford) 2010;49:1726–33.

29. Wang J, Guo G, Chen G, Wu B, Lu L, Bao L. Meta­ analysis of the association of dermatomyositis and polymyositis with cancer. Br J Dermatol 2013;169:838–47.

30. Fujimoto M, Hamaguchi Y, Kaji K, Matsushita T, Ichimura Y, Kodera M, et  al. Myositis­ specific anti­ 155/140 autoantibodies target transcription intermediary factor 1 family proteins. Arthritis Rheum 2012;64:513–22.

31. Fujimoto M, Murakami A, Kurei S, Okiyama N, Kawakami A, Mishima M, et al. Enzyme­ linked immunosorbent assays for detection of anti­

AUSSY ET AL 1370       |

transcriptional intermediary factor­ 1 γ and anti­ Mi­ 2 autoantibodies in dermatomyositis. J Dermatol Sci 2016;84:272–81.

32. Ogawa­Momohara M, Muro Y, Mitsuma T, Katayama M, Yanaba K, Nara M, et al. Strong correlation between cancer progression and anti­ transcription intermediary factor 1γ antibodies in dermatomyositis patients. Clin Exp Rheumatol 2018;36:990–5.

33. Benveniste O, Drouot L, Jouen F, Charuel JL, Bloch­Queyrat C, Behin A, et al. Correlation of anti–signal recognition particle autoantibody levels with creatine kinase activity in patients with necrotizing myopathy. Arthritis Rheum 2011;63:1961–71.

34. Vulsteke JB, De Langhe E, Claeys KG, Dillaerts D, Poesen K, Lenaerts J, et al. Detection of myositis­ specific antibodies. Ann Rheum Dis 2019;78:e7.

35. Lecouffe­Desprets M, Hémont C, Néel A, Toquet C, Masseau A, Hamidou M, et al. Clinical contribution of myositis­ related antibodies detected by immunoblot to idiopathic inflammatory myositis: a one­ year retrospective study. Autoimmunity 2018;51:89–95.

36. Zhang H, Li P, Wu D, Xu D, Hou Y, Wang Q, et al. Serum IgG subclasses in autoimmune diseases. Medicine (Baltimore) 2015;94: e387.

37. Blanco F, Kalsi J, Ravirajan CT, Speight P, Bradwell AR, Isenberg DA. IgG subclasses in systemic lupus erythematosus and other autoimmune rheumatic diseases. Lupus 1992;1:391–9.

38. Xie LD, Gao Y, Li MR, Lu GZ, Guo XH. Distribution of immunoglobulin G subclasses of anti­ thyroid peroxidase antibody in sera from patients with Hashimoto’s thyroiditis with different thyroid functional status. Clin Exp Immunol 2008;154:172–6.

39. Futei Y, Amagai M, Hashimoto T, Nishikawa T. Conformational epitope mapping and IgG subclass distribution of desmoglein 3 in paraneoplastic pemphigus. J Am Acad Dermatol 2003;49: 1023–8.

40. Ferrari S, Mudde GC, Rieger M, Veyradier A, Kremer Hovinga JA, Scheiflinger F. IgG subclass distribution of anti­ ADAMTS13 antibodies in patients with acquired thrombotic thrombocytopenic purpura. J Thromb Haemost 2009;7:1703–10.

41. Van Zelm MC. B cells take their time: sequential IgG class switching over the course of an immune response? Immunol Cell Biol 2014;92:645–6.

42. Milburn HJ, Poulter LW, Dilmec A, Cochrane GM, Kemeny DM. Corticosteroids restore the balance between locally produced Th1 and Th2 cytokines and immunoglobulin isotypes to normal in sarcoid lung. Clin Exp Immunol 1997;108:105–13.

43. Rock B, Martins CR, Theofilopoulos AN, Balderas RS, Anhalt GJ, Labib RS, et al. The pathogenic effect of IgG4 autoantibodies in endemic pemphigus foliaceus (fogo selvagem). N Engl J Med 1989;320:1463–9.

44. Koneczny I. A new classification system for IgG4 autoantibodies. Front Immunol 2018;9:97.

45. Uruha A, Allenbach Y, Charuel JL, Musset L, Aussy A, Boyer O, et al. Diagnostic potential of sarcoplasmic MxA expression in subsets of dermatomyositis. Neuropathol Appl Neurobiol 2018. E­pub ahead of print.

46. Huard C, Gullà SV, Bennett DV, Coyle AJ, Vleugels RA, Greenberg SA. Correlation of cutaneous disease activity with type 1 interferon gene signature and interferon β in dermatomyositis. Br J Dermatol 2017;176:1224–30.

47. Pinal­Fernandez I, Ferrer­Fabregas B, Trallero­Araguas E, Balada E, Martínez MA, Milisenda JC, et al. Tumour TIF1 mutations and loss of

heterozygosity related to cancer­ associated myositis. Rheumatology (Oxford) 2018;57:388–96.

48. Teraishi M, Nakajima K, Ishimoto T, Yamamoto M, Maeda N, Muro Y, et al. Anti­ transcription intermediary factor 1γ antibody titer correlates with clinical symptoms in a patient with recurrent dermatomyositis associated with ovarian cancer. Int J Rheum Dis 2018;21:900–2.

49. Kasuya A, Hamaguchi Y, Fujimoto M, Tokura Y. TIF1γ­ overexpressing, highly progressive endometrial carcinoma in a patient with dermato­ myositis positive for malignancy­ associated anti­ p155/140 autoantibody. Acta Derm Venereol 2013;93:715–6.

50. Aggarwal R, Oddis CV, Goudeau D, Koontz D, Qi Z, Reed AM, et al. Autoantibody levels in myositis patients correlate with clinical response during B cell depletion with rituximab. Rheumatology (Oxford) 2016;55:991–9.

51. Kassem L, Deygas M, Fattet L, Lopez J, Goulvent T, Lavergne E, et al. TIF1γ interferes with TGFβ1/SMAD4 signaling to promote poor outcome in operable breast cancer patients. BMC Cancer 2015;15:453.

52. Jain S, Singhal S, Francis F, Hajdu C, Wang JH, Suriawinata A, et al. Association of overexpression of TIF1γ with colorectal carcinogenesis and advanced colorectal adenocarcinoma. World J Gastroenterol 2011;17:3994–4000.

53. Aussy A, Boyer O, Cordel N. Dermatomyositis and immune­ mediated necrotizing myopathies: a window on autoimmunity and cancer. Front Immunol 2017;8:992.

54. Mohassel P, Rosen P, Casciola­Rosen L, Pak K, Mammen AL. Expression of the dermatomyositis autoantigen transcription intermediary factor 1γ in regenerating muscle. Arthritis Rheumatol 2015;67:266–72.

55. Honnorat J, Antoine JC. Paraneoplastic neurological syndromes. Orphanet J Rare Dis 2007;2:22.

56. Joseph CG, Darrah E, Shah AA, Skora AD, Casciola­Rosen LA, Wigley FM, et al. Association of the autoimmune disease scleroderma with an immunologic response to cancer. Science 2014;343:152–7.

57. Habers GE, Huber AM, Mamyrova G, Targoff IN, O’Hanlon TP, Adams S, et al. Association of myositis autoantibodies, clinical features, and environmental exposures at illness onset with disease course in juvenile myositis. Arthritis Rheumatol 2016;68:761–8.

58. Preusse C, Allenbach Y, Hoffmann O, Goebel HH, Pehl D, Radke J, et al. Differential roles of hypoxia and innate immunity in juvenile and adult dermatomyositis. Acta Neuropathol Commun 2016;4:45.

59. Gitiaux C, Latroche C, Weiss­Gayet M, Rodero MP, Duffy D, Bader­Meunier B, et al. Myogenic progenitor cells exhibit type I interferon–driven proangiogenic properties and molecular signature during juvenile dermatomyositis. Arthritis Rheumatol 2018;70:134–45.

APPENDIX A: THE ONCOMYOSITIS STUDY GROUPMembers of the OncoMyositis study group are as follows:

Zahir Amoura, Jessie Aouizerate, François­ Jérôme Authier, Ygal Benhamou, Jacques Bénichou, Bilade Cherqaoui, Olivier Chosidow, Sébastien Debarbieux, Laurence Fardet, Julie Graveleau, Jean­ Emmanuel Kahn, Vincent Langlois, Hervé Lévesque, Jérôme Martin, Nihal Martis, Agathe Masseau, Lucile Musset, Antoine Néel, Viviane Queyrel­ Moranne, Benjamin Terrier, and Stéphane Vinzio.

1371

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1371–1376DOI 10.1002/art.40883 © 2019 American College of Rheumatology. This article has been contributed to by US Government employees and their work is in the public domain in the USA.

B R I E F R E P O R T

Myositis Autoantigen Expression Correlates With Muscle Regeneration but Not Autoantibody SpecificityIago Pinal-Fernandez,1 David R. Amici,2 Cassie A. Parks,2 Assia Derfoul,2 Maria Casal-Dominguez,3 Katherine Pak,2 Richard Yeker,2 Paul Plotz,2 Jose C. Milisenda,4 Josep M. Grau-Junyent,4 Albert Selva-O’Callaghan,5 Julie J. Paik,6 Jemima Albayda,6 Andrea M. Corse,6 Thomas E. Lloyd,6 Lisa Christopher-Stine,6 and Andrew L. Mammen3

Objective. Although more than a dozen myositis- specific autoantibodies (MSAs) have been identified, most pa-tients with myositis are positive for a single MSA. The specific overexpression of a given myositis autoantigen in myo-sitis muscle has been proposed as initiating and/or propagating autoimmunity against that particular autoantigen. The present study was undertaken to test this hypothesis.

Methods. In order to quantify autoantigen RNA expression, RNA sequencing was performed on muscle biopsy samples from control subjects, MSA- positive patients with myositis, regenerating mouse muscles, and cultured hu-man muscle cells.

Results. Muscle biopsy samples were available from 20 control subjects and 106 patients with autoantibodies recognizing hydroxymethylglutaryl- coenzyme A reductase (n = 40), signal recognition particles (n = 9), Jo- 1 (n = 18), nuclear matrix protein 2 (n = 12), Mi- 2 (n = 11), transcription intermediary factor 1γ (n = 11), or melanoma differen-tiation–associated protein 5 (n = 5). The increased expression of a given autoantigen in myositis muscle was not associated with autoantibodies recognizing that autoantigen (all q > 0.05). In biopsy specimens from both myositis muscle and regenerating mouse muscles, autoantigen expression correlated directly with the expression of muscle regeneration markers and correlated inversely with the expression of genes encoding mature muscle proteins. Myo-sitis autoantigens were also expressed at high levels in cultured human muscle cells.

Conclusion. Most myositis autoantigens are highly expressed during muscle regeneration, which may relate to the propagation of autoimmunity. However, factors other than overexpression of specific autoantigens are likely to govern the development of unique autoantibodies in individual patients with myositis.

INTRODUCTION

Myositis compromises a heterogeneous group of dis-eases, including dermatomyositis (DM), immune- mediated necrotizing myopathy (IMNM), antisynthetase syndrome, and

inclusion body myositis (1). Each type of myositis is charac-terized by weakness, elevated levels of muscle enzymes, and muscle biopsy findings of inflammation, necrotic myofibers, and/or regenerating muscle cells (2). Most myositis patients have only 1 of more than a dozen myositis- specific autoan-

Supported by the Intramural Research Program of the National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH. The Myositis Research Database is supported by the the Huayi and Siuling Zhang Discovery Fund. Dr. Pinal-Fernandez’s work was supported by a Myositis Association Fellowship.

1Iago Pinal-Fernandez, MD, PhD: National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH, Bethesda, Maryland, Johns Hopkins University School of Medicine, Baltimore, Maryland, and Universitat Oberta de Catalunya, Barcelona, Spain; 2David R. Amici, BS, Cassie A. Parks, BS, Assia Derfoul, PhD, Katherine Pak, MD, Richard Yeker, BS, Paul Plotz, MD, PhD: National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH, Bethesda, Maryland; 3Maria Casal-Dominguez, MD, PhD, Andrew L. Mammen, MD, PhD: National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH, Bethesda, Maryland, and Johns Hopkins University School of Medicine, Baltimore, Maryland; 4Jose C. Milisenda, MD, Josep M. Grau-Junyent, MD, PhD: Clinic Hospital and the University of Barcelona,

Barcelona, Spain; 5Albert Selva-O’Callaghan, MD, PhD: Vall d’Hebron Hospital and Autonomous University of Barcelona, Barcelona, Spain; 6Julie J. Paik, MD, Jemima Albayda, MD, Andrea M. Corse, MD, Thomas E. Lloyd, MD, PhD, Lisa Christopher-Stine, MD, MPH: Johns Hopkins University School of Medicine, Baltimore, Maryland.

Dr. Pinal-Fernandez and Mr. Amici contributed equally to this work.No potential conflicts of interest relevant to this article were reported.Address correspondence to Iago Pinal-Fernandez, MD, PhD, or Andrew

L. Mammen, MD, PhD, Muscle Disease Unit, Laboratory of Muscle Stem Cells and Gene Expression, National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH, 50 South Drive, Room 1141, Building 50, MSC 8024, Bethesda, MD 20892. E-mail: iago.pinalfernandez@nih.gov or andrew.mammen@nih.gov.

Submitted for publication November 19, 2018; accepted in revised form March 7, 2019.

PINAL-­FERNANDEZ­ET­AL­1372       |

tibodies (MSAs), each of which is associated with a unique clinical phenotype (3).

Regenerating muscle cells in biopsy tissue from human myositis muscle have been shown to express high levels of several myositis autoantigens, including Mi- 2, transcription intermediary factor 1γ (TIF1γ), Jo- 1, hydroxymethylglutaryl- coenzyme A reductase (HMGCR), and signal recognition parti-cles (SRPs) (4–8). Given this observation, it has been proposed that increased expression of myositis autoantigens may initiate and/or maintain autoimmunity against these particular proteins. However, it has not been determined if autoantigens other than Mi- 2, TIF1γ, and Jo- 1 are expressed at high levels in regener-ating muscle, if autoantigen expression patterns differ between myositis subgroups (e.g., IMNM versus DM), or if there is a relationship between the expression level of an autoantigen and the presence of its corresponding autoantibody. Here, we have addressed these issues by using high- throughput next- generation sequencing (NGS) to quantify the expression of myositis autoantigens and other genes in muscle biopsy sam-ples obtained from patients with defined MSAs.

MATERIALS AND METHODS

Patient samples and autoantibody testing. Muscle biopsy samples obtained from patients enrolled in longitudinal cohorts at the National Institutes of Health, Johns Hopkins Myositis Center, Clinic Hospital, and Vall d’Hebron Hospital were included in the study if patients had one of the following MSAs: anti- HMGCR, anti- SRP, anti–Jo- 1, anti–nuclear matrix protein 2 (NXP- 2), anti–Mi- 2, anti- TIF1γ, or anti–melanoma differentiation–associated protein 5 (MDA- 5). Autoantibody testing was performed for anti- HMGCR as previously described (9) and by line blot for the remaining MSAs (Euroline Myositis Profile 4). Normal muscle biopsy specimens were obtained from the Johns Hopkins Neuromuscular Pathology Laboratory (n = 10) and the Skeletal Muscle Biobank of the Univer-sity of Kentucky (Lexington, KY) (n = 10).

Cultured human skeletal muscle cells. Normal human skeletal muscle myoblasts (HSMMs) were cultured according to recommended protocol by the manufacturer (Lonza). When 80% confluent, the cultures were induced to differentiate into myotubes

Figure 1. Mean RNA expression levels of the different myositis autoantigens in muscle biopsy samples obtained from control subjects and patients within different autoantibody groups. For example, SRP54 is expressed at uniformly high levels in all autoantibody groups and is not more highly expressed in patients positive for anti–signal recocognition particles (anti- SRPs). Of note, the anti–melanoma differentiation–associated protein 5 (anti–MDA- 5) autoantigen (IFIH1) is highly expressed in patients with dermatomyositis (i.e., anti–Mi- 2, anti–nuclear matrix protein 2 [anti–NPX- 2], anti–transcription intermediary factor 1γ [anti- TIF1γ], and anti–MDA- 5), moderately expressed in patients with anti–Jo- 1 autoantibodies, and expressed at low levels in patients with immune- mediated necrotizing myositis (i.e., anti- SRP and anti–hydroxymethylglutaryl- coenzyme A reductase [anti- HMGCR]). White circles represent the expression levels of each autoantigen in its corresponding autoantibody group. TPM = transcripts per kilobase million.

MYOSITIS AUTOANTIGEN EXPRESSION IN MUSCLE REGENERATION |      1373

by replacing the growth medium with differentiaton medium (Dul-becco’s modified Eagle’s medium supplemented with 2% horse serum and L- glutamine). Two plates of cells were harvested before differentiaton and then daily for 6 days.

Mouse muscle injury. Muscle injury and regeneration were induced in mice using cardiotoxin (CTX), as previously described (5). Briefly, the tibialis anterior muscles of C57BL/6 mice were uni-laterally injured via intramuscular injection of 0.1 ml of 10 μM CTX. Muscle was harvested on day 3 (n = 2), day 5 (n = 2), day 7 (n = 2), day 10 (n = 4), day 14 (n = 4), and day 28 (n = 3) postinjury. Contralateral uninjured tibialis anterior muscle tissue was also col-lected (n = 9). All animal experiments were performed according to the NIH ethical guidelines.

RNA sequencing. RNA sequencing (RNA- seq) was per-formed as previously described (10). Briefly, muscle biopsy sam-ples were homogenized in TRIzol using 1.4- mm ceramic bead low- binding tubes, and RNA was extracted using the standard TRIzol protocol. Concentration and quality of the resulting RNA were assessed using standard NanoDrop and TapeStation proto-cols, respectively. The median RNA integrity number (RIN) of the muscle biopsy samples was 7 (interquartile range [IQR] 5.9–7.4), and the RIN of the cultured human skeletal muscle cells was 9.5 (IQR 9.1–9.6). Libraries were prepared using the NeoPrep system according to the TruSeqM Stranded mRNA Library Prep proto-col (Illumina) and sequenced with Illumina HiSeq 2500 or 3000. Reads were aligned with STAR version 2.5 (11), and the abun-dance of each gene was quantified with StringTie version 1.3.3 (12). Differential gene expression was assessed with DESeq2 version 1.20.0 (13). Benjamini- Hochberg correction was used to

adjust for multiple comparisons, and corrected P values (q values) less than 0.05 were considered significant.

Data analysis. Gene expression values (transcripts per kilobase million [TPM]) were log- transformed (logTPM = log2[TPM + 1]) or referenced to the values in normal biopsy samples (log2[fold change]). Data were processed and visualized using Python (packages Numpy, Pandas, and Seaborn). Spearman’s rho was used to quantify correlation between genes of interest.

RESULTS

Correlation of myositis autoantigen RNA expression with muscle regeneration in myositis muscle. To assess myositis autoantigen expression in patients with different MSAs, RNA- seq was performed on 20 normal muscle biopsy speci-mens and on muscle biopsy specimens obtained from patients with the following MSAs: anti- SRP (n = 9), anti- HMGCR (n = 40), anti–Mi- 2 (n = 11), anti–NXP- 2 (n = 12), anti–TIF1γ (n = 11), anti–MDA- 5 (n = 5), and anti–Jo- 1 (n = 18). There was no correlation between the autoantibody for which a patient was positive and the RNA expression level of its corresponding autoantigen (all q > 0.05). For example, muscles from anti–Mi- 2–positive patients did not have increased expression of chromodomain- helicase DNA binding protein 4 RNA encoding the Mi- 2 autoantigen compared to patients with other MSAs (Figure 1). Of note, the expression of IFIH1 (an interferon [IFN] inducible gene) encoding the MDA- 5 autoantigen was elevated in all DM autoantibody groups (i.e., anti–Mi- 2, anti–NXP- 2, anti- TIF1γ, and anti–MDA- 5) compared to the anti- SRP– or anti- HMGCR–positive patients. In general, all myositis autoantigens except NXP- 2 and TIF1γ were expressed

Figure 2. Matrices of correlation coefficients for the RNA expression of myositis autoantigens versus the expression of genes found in T cells (CD3E, CD4, and CD8A), macrophages (CD14), regenerating muscle cells (NCAM1, MYOG, MYOD1, PAX7, MYH3, and MYH8), and mature muscle cells (ACTA1, MYH1, and MYH2) in myositis muscle biopsy samples (a) and regenerating mouse muscle (b). All time points were pooled for analysis. See Figure 1 for definitions.

PINAL-­FERNANDEZ­ET­AL­1374       |

at higher levels in biopsy samples obtained from myositis patients than from control subjects (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40883/ abstract).

As prior studies have shown that TIF1γ (4) and Mi- 2 (5,6) are highly expressed in regenerating myofibers, we sought to deter-mine whether there was a correlation between the expression of other myositis autoantigens and the expression of genes asso-ciated with regenerating muscle fibers. Indeed, among patients with all MSAs, there was a positive correlation between the RNA expression of each myositis autoantigen and the expression of regeneration genes (i.e., myogenin [MYOG], MyoD, PAX7, and perinatal and embryonic myosin heavy chains [MYH3 and MYH8, respectively]). In contrast, there was an inverse correla-tion between the expression of RNA for genes encoding myositis autoantigens and markers of mature muscle (i.e., skeletal mus-cle actin [ACTA1] and adult skeletal muscle myosin heavy chains [MYH1 and MYH2]) (Figure 2a).

Correlation of myositis autoantigens with muscle regeneration in regenerating mouse muscle. Myositis mus-cle may include mature myofibers as well as muscle cells at vari-ous stages of degeneration and regeneration. To define myositis autoantigen RNA expression during muscle regeneration, we uti-lized a mouse model in which muscle is injured with CTX and then allowed to regenerate; myoblast proliferation, myocyte differentia-tion, and myotube formation are synchronized. As was observed in human myositis muscle, myositis autoantigen RNA expression in the mouse muscles was positively correlated with markers of muscle differentiation and inversely correlated with the expression of adult muscle genes (Figure 2b). Indeed, the RNA expression of all myositis autoantigens increased after CTX injection (Figure 3) and paralleled the expression of genes associated with muscle regen-eration (Myog, MyoD, Pax7, Myh3, and Myh8), which transiently increased after muscle injury (see Supplementary Figure 2, avail-able on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40883/ abstract). The expression levels of genes encoding mature muscle proteins (Acta1, Myh1, and Myh2) transiently decreased after injury and subsequently increased following muscle repair (Supplementary Figure 2, available at http://onlin elibr ary.wiley.com/doi/10.1002/art.40883/ abstract).

High expression of myositis autoantigens in cul-tured human myoblasts. Myositis muscle may contain infiltrat-ing macrophages and T cells along with regenerating myofibers. Similarly, macrophages and T cells infiltrate regenerating mouse muscle, where they remove cellular debris and promote mus-cle repair. Not surprisingly, T cell and macrophage- specific gene expression were correlated with levels of myositis autoantigens in both myositis muscle and regenerating mouse muscle (Figure 2).

To confirm that myositis autoantigens are expressed at high levels in muscle cells (rather than exclusively by inflam-

matory cells), myositis autoantigen RNA expression in human myoblast cultures was analyzed as they proliferated and then differentiated into myotubes. While genes specifically expressed by inflammatory cells were not present in the cul-tured muscle cells (Supplementary Figure 3, available at http://onlin elibr ary.wiley.com/doi/10.1002/art.40883/ abstract), markers of muscle regeneration were expressed at levels equivalent to those measured in regenerating mouse muscles (Supplementary Figure 2, available at http://onlin elibr ary.wiley.com/doi/10.1002/art.40883/ abstract). Moreover, all myosi-tis autoantigens were expressed at levels equal to or higher than those observed in regenerating mouse muscle (Figure 3).

Figure 3. Evolution of the RNA expression levels (log2[TPM + 1]) of different myositis autoantigens during human skeletal muscle myoblast (HSMM) differentiation into myotubes and during the regeneration of mouse muscle following injury with cardiotoxin (CTX). In the HSMM model, proliferating myoblasts are placed in differentiation media on day 0 and allowed to differentiate into myotubes over the next 6 days. In the mouse injury model, the tibialis anterior muscle is injected with CTX on day 0, and the muscle is allowed to regenerate for as long as 28 days. CTX day 0 corresponds to the contralateral (uninjured) tibialis anterior muscle. Vertical lines indicate the 95% confidence interval for each value. See Figure 1 for other definitions.

MYOSITIS AUTOANTIGEN EXPRESSION IN MUSCLE REGENERATION |      1375

Taken together, these results demonstrate that proliferat-ing myoblasts, differentiating myocytes, and newly formed myotubes contribute substantially to the expression levels of myositis antigens in myositis muscle.

DISCUSSION

As several myositis autoantigens (i.e., Mi- 2, TIF1γ, and Jo- 1) were previously shown to be expressed at high levels in regenerat-ing muscle cells, it has been proposed that the overexpression of specific autoantigens in myositis muscle might drive autoantigen- specific immune response (6). We used RNA- seq to systemati-cally investigate autoantigen expression levels in muscle biopsy specimens obtained from myositis patients with each major MSA. RNA levels of each myositis autoantigen were found to be posi-tively correlated with markers of muscle regeneration, but levels of a given autoantigen were found not to be associated with the presence of its corresponding autoantibody. Therefore, restricted autoantigen overexpression alone does not explain why patients with myositis typically are positive for only a single MSA. Rather, it is likely that factors such as aberrant posttranslational processing (14), mislocalization of autoantigen, immunogenetic susceptibility (15), and/or exposure to molecularly similar antigens (e.g., tumor antigens) (16) determine which autoantigens will be targeted by the immune system in patients with myositis.

Our findings also showed that all myositis autoantigens are expressed at high levels not just in regenerating myositis muscle, but also in regenerating mouse muscle and in cultured human myoblasts. This indicates that elevated myositis autoantigen expression is a normal part of muscle regeneration/differentia-tion. Nonetheless, disease- related factors may also contribute to myositis autoantigen overexpression. For example, IFIH1 is expressed at low levels (<2) during all phases of cultured mus-cle cell differentiation compared to the expression levels of other myositis autoantigens (4–6). However, IFIH1 is expressed at especially high levels in muscle specimens obtained from patients with DM autoantibodies. Since IFN levels are high in DM patients (17) and IFIH1 is an IFN- inducible gene, we hypothesize that muscle regeneration and IFN both contribute to the high levels of IFIH1 in DM muscle.

The primary limitation of this study is the reliance on RNA quantitation to assess gene expression levels. However, the uti-lization of high- throughput NGS also allowed us to analyze the expression of many genes in several samples. For example, Fig-ure 2 shows the expression levels of 20 genes in >106 human myositis muscle specimens and 26 mouse muscle specimens. Such an analysis would be impractical using immunoblotting techniques to quantify protein expression levels. Further, we and others have previously shown that Mi- 2, TIF1γ, Jo- 1, HMGCR, and SRP proteins are upregulated in regenerating cells of myosi-tis muscle (4–8), validating a correlation between RNA levels and protein levels for these autoantigens.

In summary, by utilizing RNA- seq to quantitate autoantigen expression in a large number of myositis muscle biopsy samples obtained from patients with defined autoantibodies, we have demonstrated that autoantigen expression is highly correlated with muscle regeneration but that expression of a given autoan-tigen is not associated with the presence of its corresponding autoantibody. Future studies are needed to determine why only a single autoantigen is typically targeted by the immune system in a given myositis patient.

ACKNOWLEDGMENTS

We would like to thank Dr. Gustavo Gutierrez- Cruz, Dr.  Stefania Dell’Orso, and Faiza Naz (National Institute of Arthri-tis and Musculoskeletal and Skin Diseases Sequencing Facility) for all technical collaboration in RNA- seq library construction and sequencing. We would like to thank the University of Kentucky Center for Muscle Biology for providing control human muscle samples.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final ver-sion to be submitted for publication. Dr. Mammen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study conception and design. Pinal- Fernandez, Amici, Mammen.Acquisition of data. Pinal- Fernandez, Amici, Parks, Derfoul, Casal- Dominguez, Pak, Yeker, Plotz, Milisenda, Grau- Junyent, Selva- O’Callaghan, Paik, Albayda, Corse, Lloyd, Christopher- Stine, Mammen.Analysis and interpretation of data. Pinal- Fernandez, Amici.

REFERENCES 1. Selva-O’Callaghan A, Pinal-Fernandez I, Trallero-Araguás E,

Milisenda JC, Grau-Junyent JM, Mammen AL. Classification and management of adult inflammatory myopathies. Lancet Neurol 2018;17:816–28.

2. Dalakas MC. Inflammatory muscle diseases. N Engl J Med 2015;373:393–4.

3. Mammen AL, Casciola-Rosen L, Christopher-Stine L, Lloyd TE, Wagner KR. Myositis- specific autoantibodies are specific for myositis compared to genetic muscle disease. Neurol Neuroimmunol Neuroinflamm 2015;2:e172.

4. Mohassel P, Rosen P, Casciola-Rosen L, Pak K, Mammen AL. Expression of the dermatomyositis autoantigen transcription intermediary factor 1γ in regenerating muscle. Arthritis Rheumatol 2015;67:266–72.

5. Mammen AL, Casciola-Rosen LA, Hall JC, Christopher-Stine L, Corse AM, Rosen A. Expression of the dermatomyositis autoantigen Mi- 2 in regenerating muscle. Arthritis Rheum 2009;60:3784–93.

6. Casciola-Rosen L, Nagaraju K, Plotz P, Wang K, Levine S, Gabrielson E, et al. Enhanced autoantigen expression in regenerating muscle cells in idiopathic inflammatory myopathy. J Exp Med 2005;201:591–601.

7. Arouche-Delaperche L, Allenbach Y, Amelin D, Preusse C, Mouly V, Mauhin W, et al. Pathogenic role of anti- signal recognition protein and anti- 3- hydroxy- 3- methylglutaryl- CoA reductase antibodies in

PINAL-­FERNANDEZ­ET­AL­1376       |

necrotizing myopathies: myofiber atrophy and impairment of muscle regeneration in necrotizing autoimmune myopathies. Ann Neurol 2017;81:538–48.

8. Allenbach Y, Arouche-Delaperche L, Preusse C, Radbruch H, Butler-Browne G, Champtiaux N, et al. Necrosis in anti- SRP+ and anti- HMGCR+ myopathies: role of autoantibodies and complement. Neurology 2018;90:e507–17.

9. Mammen AL, Chung T, Christopher-Stine L, Rosen P, Rosen A, Doering KR, et al. Autoantibodies against 3- hydroxy- 3- methylglutaryl- coenzyme A reductase in patients with statin- associated autoimmune myopathy. Arthritis Rheum 2011;63:713–21.

10. Amici DR, Pinal-Fernandez I, Mázala DA, Lloyd TE, Corse AM, Christopher-Stine L, et al. Calcium dysregulation, functional calpainopathy, and endoplasmic reticulum stress in sporadic inclusion body myositis. Acta Neuropathol Commun 2017;5:24.

11. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA- seq aligner. Bioinformatics 2013;29:15–21.

12. Pertea M, Pertea GM, Antonescu CM, Chang TC, Mendell JT, Salzberg SL. StringTie enables improved reconstruction of a

transcriptome from RNA- seq reads. Nat Biotechnol 2015;33: 290–5.

13. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 2014;15:550.

14. Levine SM, Raben N, Xie D, Askin FB, Tuder R, Mullins M, et al. Novel conformation of histidyl–transfer RNA synthetase in the lung: the target tissue in Jo- 1 autoantibody–associated myositis. Arthritis Rheum 2007;56:2729–39.

15. Miller FW, Chen W, O’Hanlon TP, Cooper RG, Vencovsky J, Rider LG, et al. Genome- wide association study identifies HLA 8.1 ancestral haplotype alleles as major genetic risk factors for myositis phenotypes. Genes Immun 2015;16:470–80.

16. Pinal-Fernandez I, Ferrer-Fabregas B, Trallero-Araguas E, Balada E, Martínez MA, Milisenda JC, et al. Tumour TIF1 mutations and loss of heterozygosity related to cancer- associated myositis. Rheumatology (Oxford) 2018;57:388–96.

17. Greenberg SA, Pinkus JL, Pinkus GS, Burleson T, Sanoudou D, Tawil R, et al. Interferon- α/β- mediated innate immune mechanisms in dermatomyositis. Ann Neurol 2005;57:664–78.

DOI 10.1002/art.41035

Expression of Concern: Demasi M, Cleland LG, Cook- Johnson RJ, James MJ. Effects of hypoxia on the expression and activity of cyclooxygenase 2 in fibroblast- like synoviocytes:

interactions with monocyte- derived soluble mediators. Arthritis Rheum 2004;50:2441–9. https ://onlin elibr ary.wiley.com/doi/10.1002/art.20429

Arthritis & Rheumatology (formerly Arthritis & Rheumatism) and John Wiley & Sons are issuing an Expression of Concern to inform readers that questions regarding the similarity of images used in Figures 1, 2, and 6 of the above- referenced article were raised, which may be pertinent to the data and conclusions of the article. An inquiry convened by the University of Adelaide, SA, Australia did not find research misconduct.

PINAL-­FERNANDEZ­ET­AL­1376       |

necrotizing myopathies: myofiber atrophy and impairment of muscle regeneration in necrotizing autoimmune myopathies. Ann Neurol 2017;81:538–48.

8. Allenbach Y, Arouche-Delaperche L, Preusse C, Radbruch H, Butler-Browne G, Champtiaux N, et al. Necrosis in anti- SRP+ and anti- HMGCR+ myopathies: role of autoantibodies and complement. Neurology 2018;90:e507–17.

9. Mammen AL, Chung T, Christopher-Stine L, Rosen P, Rosen A, Doering KR, et al. Autoantibodies against 3- hydroxy- 3- methylglutaryl- coenzyme A reductase in patients with statin- associated autoimmune myopathy. Arthritis Rheum 2011;63:713–21.

10. Amici DR, Pinal-Fernandez I, Mázala DA, Lloyd TE, Corse AM, Christopher-Stine L, et al. Calcium dysregulation, functional calpainopathy, and endoplasmic reticulum stress in sporadic inclusion body myositis. Acta Neuropathol Commun 2017;5:24.

11. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA- seq aligner. Bioinformatics 2013;29:15–21.

12. Pertea M, Pertea GM, Antonescu CM, Chang TC, Mendell JT, Salzberg SL. StringTie enables improved reconstruction of a

transcriptome from RNA- seq reads. Nat Biotechnol 2015;33: 290–5.

13. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 2014;15:550.

14. Levine SM, Raben N, Xie D, Askin FB, Tuder R, Mullins M, et al. Novel conformation of histidyl–transfer RNA synthetase in the lung: the target tissue in Jo- 1 autoantibody–associated myositis. Arthritis Rheum 2007;56:2729–39.

15. Miller FW, Chen W, O’Hanlon TP, Cooper RG, Vencovsky J, Rider LG, et al. Genome- wide association study identifies HLA 8.1 ancestral haplotype alleles as major genetic risk factors for myositis phenotypes. Genes Immun 2015;16:470–80.

16. Pinal-Fernandez I, Ferrer-Fabregas B, Trallero-Araguas E, Balada E, Martínez MA, Milisenda JC, et al. Tumour TIF1 mutations and loss of heterozygosity related to cancer- associated myositis. Rheumatology (Oxford) 2018;57:388–96.

17. Greenberg SA, Pinkus JL, Pinkus GS, Burleson T, Sanoudou D, Tawil R, et al. Interferon- α/β- mediated innate immune mechanisms in dermatomyositis. Ann Neurol 2005;57:664–78.

DOI 10.1002/art.41035

Expression of Concern: Demasi M, Cleland LG, Cook- Johnson RJ, James MJ. Effects of hypoxia on the expression and activity of cyclooxygenase 2 in fibroblast- like synoviocytes:

interactions with monocyte- derived soluble mediators. Arthritis Rheum 2004;50:2441–9. https ://onlin elibr ary.wiley.com/doi/10.1002/art.20429

Arthritis & Rheumatology (formerly Arthritis & Rheumatism) and John Wiley & Sons are issuing an Expression of Concern to inform readers that questions regarding the similarity of images used in Figures 1, 2, and 6 of the above- referenced article were raised, which may be pertinent to the data and conclusions of the article. An inquiry convened by the University of Adelaide, SA, Australia did not find research misconduct.

1377

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1377–1390DOI 10.1002/art.40881 © 2019 The Authors. Arthritis & Rheumatology published by Wiley Periodicals, Inc. on behalf of American College of Rheumatology.This is an open access article under the terms of the Creat ive Commo ns Attri bution-NonCo mmercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Galectin- 9 and CXCL10 as Biomarkers for Disease Activity in Juvenile Dermatomyositis: A Longitudinal Cohort Study and Multicohort ValidationJudith Wienke,1 Felicitas Bellutti Enders,2 Johan Lim,3 Jorre S. Mertens,4 Luuk L. van den Hoogen,1 Camiel A. Wijngaarde,1 Joo Guan Yeo,5 Alain Meyer,6 Henny G. Otten,1 Ruth D. E. Fritsch-Stork,7 Sylvia S. M. Kamphuis,8 Esther P. A. H. Hoppenreijs,9 Wineke Armbrust,10 J. Merlijn van den Berg,11 Petra C. E. Hissink Muller,12 Janneke Tekstra,1 Jessica E. Hoogendijk,1 Claire T. Deakin,13 Wilco de Jager,1 Joël A. G. van Roon,1 W. Ludo van der Pol,1 Kiran Nistala,14 Clarissa Pilkington,14 Marianne de Visser,3 Thaschawee Arkachaisri,5 Timothy R. D. J. Radstake,1 Anneke J. van der Kooi,3 Stefan Nierkens,1 Lucy R. Wedderburn,13 Annet van Royen-Kerkhof,1 and Femke van Wijk1

Objective. Objective evaluation of disease activity is challenging in patients with juvenile dermatomyositis (DM) due to a lack of reliable biomarkers, but it is crucial to avoid both under- and overtreatment of patients. Recently, we identified 2 proteins, galectin- 9 and CXCL10, whose levels are highly correlated with the extent of juvenile DM disease activity. This study was undertaken to validate galectin- 9 and CXCL10 as biomarkers for disease activity in juvenile DM, and to assess their disease specificity and potency in predicting the occurrence of flares.

Methods. Levels of galectin- 9 and CXCL10 were measured by multiplex immunoassay in serum samples from 125 unique patients with juvenile DM in 3 international cross- sectional cohorts and a local longitudinal cohort. The disease specificity of both proteins was examined in 50 adult patients with DM or nonspecific myositis (NSM) and 61 patients with other systemic autoimmune diseases.

Results. Both cross- sectionally and longitudinally, galectin- 9 and CXCL10 outperformed the currently used lab-oratory marker, creatine kinase (CK), in distinguishing between juvenile DM patients with active disease and those in remission (area under the receiver operating characteristic curve [AUC] 0.86–0.90 for galectin- 9 and CXCL10; AUC 0.66–0.68 for CK). The sensitivity and specificity for active disease in juvenile DM was 0.84 and 0.92, respectively, for galectin- 9 and 0.87 and 1.00, respectively, for CXCL10. In 10 patients with juvenile DM who experienced a flare and were prospectively followed up, continuously elevated or rising biomarker levels suggested an imminent flare up to several months before the onset of symptoms, even in the absence of elevated CK levels. Galectin- 9 and CXCL10 distinguished between active disease and remission in adult patients with DM or NSM (P = 0.0126 for galectin-9 and P < 0.0001 for CXCL10) and were suited for measurement in minimally invasive dried blood spots (healthy controls versus juvenile DM, P = 0.0040 for galectin-9 and P < 0.0001 for CXCL10).

Conclusion. In this study, galectin- 9 and CXCL10 were validated as sensitive and reliable biomarkers for disease activity in juvenile DM. Implementation of these biomarkers into clinical practice as tools to monitor disease activity and guide treatment might facilitate personalized treatment strategies.

The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.

Supported by the NIHR Great Ormond Street Biomedical Research Centre at Great Ormond Street Hospital for Children NHS Foundation Trust. The Dutch Juvenile Dermatomyositis (DM) cohort was supported by the Princess Beatrix Fund and the De Bas Stichting and the Healthcare Insurers Innovation Fund. The Singaporean Juvenile DM cohort was supported by the National Research Foundation Singapore (National Medical Research Council grant NMRC/CG/M003/2017) and the Singapore Ministry of Health’s National Medical Research Council. The UK Juvenile DM Cohort and Biomarker Study was supported by Wellcome Trust UK (grants 085860 and 097259), Myositis UK, Arthritis Research UK (grants 14518, 20164, and 21593), Cure JM, The Myositis Association, Great Ormond Street Children’s Charity (grant V1268), and the NIHR Rare Diseases Translational Research Collaboration.

1Judith Wienke, MD, Luuk L. van den Hoogen, MD, PhD, Camiel A. Wijngaarde, MD, Henny G. Otten, PhD, Janneke Tekstra, MD, PhD, Jessica E. Hoogendijk, MD, PhD, Wilco de Jager, PhD, Joël A. G. van Roon, PhD, W. Ludo van der Pol, MD, PhD, Timothy R. D. J. Radstake, MD, PhD, Stefan Nierkens, PhD, Annet van Royen-Kerkhof, MD, PhD, Femke van Wijk, PhD: University Medical Centre Utrecht, Utrecht, The Netherlands; 2Felicitas Bellutti Enders, MD, PhD: Lausanne University Hospital, Lausanne, Switzerland, and University Hospital Basel, Basel, Switzerland; 3Johan Lim, MD, Marianne de Visser, MD, PhD, Anneke J. van der Kooi, MD, PhD: Academic Medical Centre Amsterdam, Amsterdam, The Netherlands; 4Jorre S. Mertens, MD, PhD: University Medical Centre Utrecht, Utrecht, The Netherlands, and Radboud University Medical Centre, Nijmegen, The Netherlands; 5Joo Guan Yeo, MBBS, Thaschawee Arkachaisri, MD, PhD: K. K. Women’s and Children’s Hospital, Duke-NUS Medical School, SingHealth Duke-NUS Academic

WIENKE ET AL 1378       |

INTRODUCTION

Juvenile dermatomyositis (DM) is a rare, chronic systemic immune- mediated disease with a high disease burden. In children with juvenile DM, the disease is characterized by inflammation of the skeletal muscles and skin, leading to muscle weakness and a pathognomonic skin rash. Vital organs such as the lung and heart can also be involved. Although the pathogenesis is still largely unknown, environmental and genetic factors may predispose chil-dren to the disease (1–5). The autoimmune process is character-ized by a type I interferon signature and by infiltration of immune cells such as plasmacytoid dendritic cells, B cells, CD4+ T cells, and macrophages into the skin and muscle tissue (6–9).

Children with juvenile DM are at risk of both under- and overtreatment due to a lack of reliable biomarkers that could be used to gauge the extent of disease activity. Current treatment guidelines recommend immunosuppression for at least 2 years, tapering steroids over the first year, and withdrawing treatment if a patient has been taken off steroids and has achieved disease remission with methotrexate (or an alternative disease- modifying antirheumatic drug) for a minimum of 1 year (10–12). However, for some patients, this standardized regimen may not be optimal. Approximately 50% of patients do not respond to initial treatment or present with disease flares during follow- up, resulting in addi-tional tissue damage and impaired physical recovery (13–15). Of the other 50% of patients, some could likely benefit from a shorter treatment duration, taking into account that overtreatment with steroids can result in serious side effects in children, such as Cushing’s syndrome, osteoporosis, and growth delay (16–18).

To determine the rate of medication tapering and to avoid both under- and overtreatment, objective measurement of disease activity and subclinical inflammation is crucial. However, validated and reliable biomarkers for disease activity in juvenile DM are lack-ing (19). Disease activity is currently assessed by a combination of muscle enzyme testing and clinical evaluation (10,20–22); the lat-ter depends on the experience of the health care professional and the patient’s collaboration. Muscle enzymes, including serum cre-atine kinase (CK) activity, have been shown to correlate only mod-erately with disease activity in juvenile DM, and the erythrocyte sedimentation rate and C- reactive protein level are rarely elevated

in patients with juvenile DM (23–25). Lack of objective tools or bio-markers to monitor the response to therapy also hampers clinical trial design. Thus, there is an unmet need for an objective and reliable measure of disease activity.

Recently, in a cross- sectional cohort of patients with juvenile DM, we demonstrated that 3 proteins, galectin- 9, CXCL10, and tumor necrosis factor receptor type II, can distinguish between juvenile DM patients with active disease and those in remission, with galectin- 9 and CXCL10 being the most discriminative mark-ers (26,27). CXCL10 and galectin- 9 can be produced by a vari-ety of cells, both immune and nonimmune, upon stimulation with interferons (28,29). CXCL10 has been recognized as a biomarker in several human autoimmune diseases, including myositis (29–33), whereas galectin- 9 has been investigated mainly as a bio-marker in cancer and viral infections (28,34). Reports on the role of galectin- 9 in autoimmunity are conflicting, suggesting either an attenuating or an aggravating effect on autoimmune manifesta-tions in experimental models (35,36). Its role in human autoim-mune diseases has yet to be elucidated.

We aimed to validate galectin- 9 and CXCL10 as biomarkers for active disease in patients with juvenile DM, to examine their disease specificity in adult patients with DM, adult patients with nonspecific myositis (NSM), and patients with other systemic autoimmune diseases, to assess their potency in predicting flares, and to test the applicability of the biomarkers in minimally invasive dried blood spots, in order to aid broad implementation into clin-ical practice.

PATIENTS AND METHODS

Cohorts. In total, 125 unique patients with juvenile DM from 3 independent cross- sectional international cohorts and 1 Dutch prospective cohort participated in the present study, with inclusion between May 2001 and May 2017. Two large cohorts from Utre-cht, The Netherlands and London, UK were used for validation of the biomarkers; a third smaller cohort from Singapore was used to assess international generalizability. An overview of all cohorts is shown in Table  1. The internal validation cohort (IVC) from Utrecht does not overlap with the previously reported discovery cohort (26). For specific questions, including disease specificity,

Medical Center, Singapore; 6Alain Meyer, MD, PhD: CHU de Strasbourg, Strasbourg, France; 7Ruth D. E. Fritsch-Stork, MD, PhD: University Medical Centre Utrecht, Utrecht, The Netherlands, Sigmund Freud Private University, Vienna, Austria, and Hanusch Krankenhaus und Ludwig Boltzmann Institut für Osteologie, Vienna, Austria; 8Sylvia S. M. Kamphuis, MD, PhD: Sophia Children’s Hospital, Erasmus University Medical Centre, Rotterdam, The Netherlands; 9Esther P. A. H. Hoppenreijs, MD: Amalia Children’s Hospital, Radboud University Medical Centre, Nijmegen, The Netherlands; 10Wineke Armbrust, MD, PhD: Beatrix Children’s Hospital, University Medical Centre Groningen, Groningen, The Netherlands; 11J. Merlijn van den Berg, MD, PhD: Emma Children’s Hospital AMC, University of Amsterdam, Amsterdam, The  Netherlands; 12Petra C. E. Hissink Muller, MD: Sophia Children’s Hospital,  Erasmus University Medical Centre, Rotterdam, The Netherl - ands, and Leiden University Medical Centre, Leiden, The Netherlands; 13Claire T. Deakin, PhD, Lucy R. Wedderburn, MD, PhD: University College

London, University College London Hospital, and the NIHR Biomedical Research Centre at Great Ormond Street Hospital, London, UK; 14Kiran Nistala, MD, PhD, Clarissa Pilkington, MD: University College London, London, UK.

Drs. Wienke and Bellutti Enders contributed equally to this work. Drs. van Royen-Kerkhof and van Wijk contributed equally to this work.

Dr. Nistala is currently employed by GlaxoSmithKline and owns stock or stock options in GlaxoSmithKline. No other disclosures relevant to this article were reported.

Address correspondence to Annet van Royen-Kerkhof, MD, PhD, or Femke van Wijk, PhD, Suite KC.02.085.0, Lundlaan 6, P. O. Box 85090, 3508 AB Utrecht, The Netherlands. E-mail: a.vanroyen@umcutrecht.nl or f.vanwijk@umcutrecht.nl.

Submitted for publication October 16, 2018; accepted in revised form March 5, 2019.

GALECTIN- 9 AND CXCL10 AS BIOMARKERS FOR JUVENILE DM |      1379

longitudinal follow- up, and measurements in dried blood spots, a combination of blood samples from the IVC and blood samples

from new patients was used.

Participants. Patients with juvenile DM were included if they met the Bohan and Peter criteria for definite or probable juvenile DM (37,38). The Childhood Myositis Assessment Scale (CMAS; scale 0–52) (39), Manual Muscle Testing of 8 muscle groups (MMT- 8; scale 0–80) (40), and physician’s global assess-ment of disease activity (PhGA; scale 0–10) were recorded as clin-ical measures of muscle and global disease activity. In addition, cutaneous assessment tool (CAT) scores measuring the severity of skin disease (scale 0–116) (41) were recorded in Dutch and Singaporean patients. Disease remission was defined according to the updated criteria for clinically inactive disease and, in the case of missing data, was defined by clinical description (42). All other patients were considered to have active disease. Flares were defined as the combination of the following 3 items: a pre-vious response to treatment with the decision to start tapering steroids, worsening of at least 1 of 3 clinical scores (CMAS, PhGA, and CAT) by ≥2 points, and the decision to start new immunosup-pressive treatment or increase the current dose.

Adult patients with DM and those with NSM were classified according to the European Neuromuscular Centre criteria (43). Myositis was confirmed by biopsy unless typical skin manifesta-

tions of DM were present. Patients with cancer- associated myosi-tis were excluded. Disease activity was determined by combined evaluation of muscle strength with the Medical Research Council Muscle Scale (44), skin symptoms, and muscle enzyme levels. To determine the disease specificity of the biomarkers, different disease controls were added in the study, including pediatric and adult patients with systemic lupus erythematosus (SLE), pediatric patients with localized scleroderma, adult patients with eosino-philic fasciitis (EF), and pediatric and adult patients with hereditary proximal spinal muscular atrophy (SMA). All controls had either systemic inflammation, inflammation of the skin or muscles, or a noninflammatory neuromuscular disorder.

Patients with SLE fulfilled the American College of Rheu-matology classification criteria for SLE (45). Active disease was defined as an SLE Disease Activity Index score of ≥4 of 105 (46). Patients with localized scleroderma were diagnosed based on the typical clinical picture, with active disease being defined as a mod-ified Localized Scleroderma Skin Severity Index (mLoSSi) score of ≥5 of 162 (47). Patients with EF were diagnosed based on the clinical picture and histopathologic evaluation of skin biopsy spec-imens containing the fascia. As the mLoSSi may stay high in these patients due to the presence of extensive, irreversible sclerosis despite a reduction of inflammation, active disease was defined as a PhGA score of ≥5 (on 100- mm visual analog scale) (47). Patients with hereditary proximal SMA, a progressive, noninflammatory

Table 1. Overview of the juvenile DM cohorts*

Abbreviation City, countryNo. of

patientsNo. of

samples

No. of active disease–remission

paired samples

International validation cohorts

External validation cohort

EVC London, UK 61 79 16

Internal validation cohort

IVC; JDM NL Utrecht, The Netherlands

47; 47 83; 58 26; 11

Asian cohort JDM Sing Singapore 12 13 –Analysis- specific sub-

cohorts from UtrechtSystemic

autoimmune disease cohort

– Utrecht, The Netherlands

14 16 2

Longitudinal cohort – Utrecht, The Netherlands

28 286 –

Dried blood spot cohort

– Utrecht, The Netherlands

7 10 –

* Data are listed as follows: for the London external validation cohort (EVC), see Figure 1, Supplementary Table 1, and Supplementary Figure 1; for the Utrecht internal validation cohort (IVC), see Figure 1, Supplementary Table 2, and Supplementary Figure 1; for the Juvenile Dermatomyositis The Netherlands (JDM NL) and Juvenile Dermatomyositis Singapore (JDM Sing) cohorts, see Figure 2 and Supplementary Table 4; for the systemic autoimmune disease cohort, see Figure 2 and Supplementary Table 5; for the longitudinal cohort, see Figure 3, Supplementary Table 6, and Sup-plementary Figure 2; for the dried blood spot cohort, see Figure 4 and Supplementary Table 7. All supplementary tables and figures are available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract.

WIENKE ET AL 1380       |

neuromuscular disorder, were diagnosed by genetic confirmation of a homozygous loss of function of the survival motor neuron 1 gene (48); these patients served as disease controls. Adult healthy volunteers were included as healthy controls.

Ethics approval. The study was approved by the institutional ethics committees of the involved centers (UMC Utrecht [approval nos. METC 15- 191 and 12- 466], UK [approval no. MREC1/3/22], CHUV Lausanne, CHU Strasbourg, SingHealth centralized IRB, AMC Amsterdam) and conducted according to the Declaration of Helsinki. Written informed consent was obtained prior to inclusion in the study, both from patients and from parents or legal repre-sentatives when the patient was younger than 12 years old.

Blood samples. Blood was collected in serum tubes in accordance with the local study protocol (all participating centers). At the UMC Utrecht, blood samples were collected in sodium- heparin tubes in addition to serum tubes. All samples were spun down and aliquoted within 4 hours after collection, and subse-quently stored at −80°C until analyzed.

Measurement in dried blood spots. Dried blood spots were made by application of 50 μl sodium- heparin full blood to each spot on Whatman 903 filter paper within 4 hours after the blood sample was obtained. Spotted filter papers were dried for 2 days at room temperature to mimic mail delivery times, and sub-sequently stored with desiccant in individual air- tight polyethylene bags at −80°C under constant monitoring of humidity levels until analyzed. Two circles of 3.0 mm in diameter (containing ~3 μl of whole blood each) were punched from the central part of 1 spot and eluted in 100 μl buffer (phosphate buffered saline contain-ing 5 ml/liter Tween 20, 10 gm/liter bovine serum albumin, and Complete protease inhibitor cocktail with EDTA [1 tablet per 25 ml buffer; Roche]) in 96- well plates. Plates were sealed and placed overnight at 4°C on a microshaker (600 revolutions per minute) and were spun down at 2,100g for 2 minutes. The analysis was performed on the obtained eluate.

Biomarker analysis. Galectin- 9 and CXCL10 were measured in 50 μl of serum, plasma, or eluate by multiplex assay (xMAP; Luminex). CXCL10 was measured in undiluted mate-rial. Galectin- 9 was measured in 10× diluted plasma or serum, except in the serum/plasma samples paired with dried blood spots (in which case galectin- 9 was measured undiluted from the eluate and serum/plasma). The multiplex immunoassay was performed as described previously (49). Heterophilic immuno-globulins were preabsorbed from all samples with HeteroBlock (Omega Biologicals). Acquisition was performed with a Bio- Rad FlexMAP3D in combination with xPONENT software version 4.2 (Luminex). Data analysis was performed with Bioplex Manager version 6.1.1 (Bio- Rad).

Between measurement of the internal and external vali-dation cohorts in 2015, the recombinant protein for galectin- 9 was replaced, which affected the standard curve. Therefore, absolute values between these cohorts may not be compa-rable. Since 2015, the interassay variability has been negligi-ble (50). All biomarker analyses were performed at the UMC Utrecht, thereby minimizing intercenter variation. Treating physicians were blinded with regard to biomarker levels, and technicians performing the multiplex assay were blinded with regard to clinical data.

Statistical analysis. Basic descriptive statistics were used to describe the patient population. Statistical analyses were performed using either GraphPad Prism version 7.0 or SPSS Statistics version 21 (IBM). Correlations were assessed using Spearman’s rank correlation coefficients. For comparisons between 2 groups, the Mann- Whitney U test (unpaired analysis) or Wilcoxon’s matched- pairs signed rank test (paired analysis) was used. For comparisons between multiple groups, nonpara-metric variants of analysis of variance (ANOVA) with post hoc correction for multiple testing were used (Dunn’s post hoc test for Kruskal- Wallis, and Šídák’s or Tukey’s post hoc test for 2- way ANOVA, as appropriate). Multiplicity- adjusted P values less than 0.05 were considered significant.

To assess diagnostic accuracy, area under the receiver oper-ating characteristic (ROC) curves (AUCs) were constructed. Cut-off values for the diagnostic accuracy of galectin- 9 and CXCL10 were determined based on the maximal Youden’s Index, with a sensitivity of at least 80%.

RESULTS

Cross- sectional validation of galectin- 9 and CXCL10. To validate the biomarker potential of galectin- 9 and CXCL10, we measured the proteins in blood samples from patients with juvenile DM from 2 independent validation cohorts: an external validation cohort (EVC) from London and an internal validation cohort (IVC) from Utrecht. The clinical characteristics of these cohorts are shown in Supplementary Tables 1 and 2 (available on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract). As observed in the previously reported discovery cohort (26), the levels of galectin- 9 and CXCL10 were sig-nificantly higher in patients with active disease compared to patients in remission (P < 0.0001) (results in Supplementary Figures 1A and B [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). The levels were highest at the time of diagnosis (before treatment), decreased steadily under treat-ment, and were comparably low in remission regardless of whether the patient was receiving or not receiving medication while in remission (Figures 1A and B).

GALECTIN- 9 AND CXCL10 AS BIOMARKERS FOR JUVENILE DM |      1381

The wide range of biomarker levels in the group of juvenile DM patients with active disease who were on treatment cor-responded to a wide range of clinical disease activity scores within this group (CMAS scores ranging 3–44 in the EVC and 10–52 in the IVC; PhGA scores ranging 2–8 in the EVC and 1–9 in the IVC). Based on the levels of both galectin- 9 and CXCL10, we were able to differentiate patients with active disease while receiving medication from patients in remission while receiving medication (Figures 1A and B), which is clinically important to assess the response to treatment. Paired analysis within individual patients, in which we compared samples from

a period of active disease and from a period of remission in each patient, showed decreasing biomarker levels in response to therapy and confirmed the high discriminative power of both proteins (each P = 0.0078 in the EVC and P = 0.0002 in the IVC) (Figures 1C and D).

To further assess the discriminative power of galectin- 9 and CXCL10 for distinguishing between a status of active disease and a status of remission in juvenile DM, we examined the AUCs in the 2 separate cohorts. In comparing active disease and remission in patients regardless of their treatment status, the levels of galec-tin- 9 and CXCL10 had AUCs of 0.894 and 0.863, respectively, in

Figure  1. Cross- sectional validation of galectin- 9 and CXCL10 as biomarkers for disease activity in juvenile dermatomyositis (DM) in 2 independent validation cohorts. A and B, Measurement of galectin- 9 and CXCL10 by multiplex immunoassay in serum from patients with active disease before start of treatment (group A), active disease while receiving medication (group AM), in remission while receiving medication (group RM), and in remission while not receiving medication (group R), in the external validation cohort (EVC) (n = 12 group A, n = 27 group AM, n = 28 group RM, n = 12 group R) (A) and the internal validation cohort (IVC) (n = 25 group A, n = 30 group AM, n = 16 group RM, n = 12 group R) (B). In group AM, 3 samples from 1 patient (from different time points at least 3 months apart) and 2 samples from 6 patients (from different time points 2–11 months apart) were included. Data are shown as box plots. Each box represents the interquartile range. Lines inside the boxes represent the median (log scale). Lines outside the boxes represent the 10th and 90th percentiles. Symbols represent individual patients. Multiplicity- adjusted P values were determined by Kruskal- Wallis test with Dunnett’s post hoc test. C and D, Measurement of galectin- 9 and CXCL10 in paired samples from individual patients (regardless of treatment status) during active disease and remission, from the EVC (median time between samples 23 months) (C) and IVC (median time between samples 12 months) (D). P values were determined by Wilcoxon’s matched- pairs signed rank test. E and F, Area under the receiver operating characteristic (ROC) curves (AUCs) for diagnostic accuracy of galectin- 9, CXCL10, and creatine kinase (CK) in patients (regardless of treatment status) from the EVC (E) and IVC (F). Only patients with complete data for the specific ROC curve were included. G and H, Spearman’s rank correlations of galectin- 9, CXCL10, and CK levels with Childhood Myositis Assessment Scale (CMAS) scores in the EVC (n = 79) (G) and IVC (n = 61) (H).

B

A AM RM R

10000

100000

Gal

ectin

-9 (p

g/m

L)

A AM RM R

100

1000

10000

CXC

L10

(pg/

mL)

<0.0001<0.0001

0.46720.1457

<0.0001

<0.0001<0.0001

0.13950.0260

<0.0001

Active versus remission (EVC)

1 - Specificity1,00,80,60,40,20

Sens

itivi

ty

1,0

0,8

0,6

0,4

0,2

0

CXCL10 (AUC: 0.877)Galectin-9 (AUC: 0.894)

E

CK (AUC: 0.682)

Active versus remission (IVC)

1 - Specificity

Sens

itivi

ty

F

CXCL10 (AUC: 0.902)Galectin-9 (AUC: 0.863)

CK (AUC: 0.664)1,00,80,60,40,2

1,0

0,8

0,6

0,4

0,2

00A/AM R/RM

10000

1000000.0002 0.0002

A/AM R/RM

100

1000

10000

CXC

L10

(pg/

mL)

D

0.0078 0.0078

A/AM R/RM10000

100000

Gal

ectin

-9 (p

g/m

L)

A/AM R/RM10

100

1000

10000C

XCL1

0 (p

g/m

L)

C

A AM RM R10

100

1000

10000

Gal

ectin

-9 (p

g/m

L)

0 20 40 60

100

1000

10000

0 20 40 60

10

100

1000

10000

100000

CMAS

CK

(IU

/L)

0 20 40 60

10

100

1000

10000

CMAS

CXC

L10

(pg/

mL)

0 20 40 60

10000

100000

CMAS

Gal

ectin

-9 (p

g/m

L)

rs = -0.709, p<0.0001 rs = -0.711, p<0.0001 rs = -0.323, p=0.0053G

H

0 20 40 60

10000

100000

0 20 40 60

10

100

1000

10000

100000

CMAS

CK

(IU

/L)

CMAS

CXC

L10

(pg/

mL)

CMAS

Gal

ectin

-9 (p

g/m

L)

rs = -0.744, p<0.0001 rs = -0.673, p<0.0001 rs = -0.381, p=0.0027

External validation cohortInternal validation cohort

CXC

L10

(pg/

mL)

A AM RM R10000

100000

Gal

ectin

-9 (p

g/m

L)

A

<0.0001<0.0001

0.00390.0007

0.0004<0.0001

0.0001

Internal validation cohort

Internal validation cohort

External validation cohort

External validation cohort External validation cohort Internal validation cohort

WIENKE ET AL 1382       |

the EVC and 0.877 and 0.902, respectively, in the IVC (Figures 1E and F, and Supplementary Table 3 [online at http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]).

To take into account the effect of treatment, we also assessed the AUC for differentiating active disease from dis-ease remission in patients who were taking medication. Dur-ing treatment, the levels of galectin- 9 and CXCL10 had AUCs of 0.844 and 0.776, respectively, in the EVC and 0.860 and 0.840, respectively, in the IVC (see Supplementary Figures 1C and D and Supplementary Table 3 [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). Moreover, galectin- 9 and CXCL10 performed better than the current standard labo-ratory marker, CK, in both cohorts (AUCs for CK, 0.682 in the EVC and 0.662 in the IVC).

To calculate the optimal cutoff value for distinguishing active disease from disease remission, we analyzed the ROC curves in the IVC, as blood samples from this cohort were assessed according to the most recently optimized and standardized pro-tocol of the multiplex immunoassay (50). Based on the coordi-nates of this ROC curve, we determined the cutoff values for discriminating active disease from remission, yielding a cutoff value of 19,396 pg/ml for galectin- 9 and 805 pg/ml for CXCL10, with a high sensitivity (0.84 for galectin- 9 and 0.87 for CXCL10) and a high negative predictive value (0.83 for galectin- 9 and 0.87 for CXCL10) (Table 2); these values ensured a low risk of ongo-ing inflammation in the case of a test result that was below the cutoff. The specificity of the galectin- 9 and CXCL10 cutoff lev-els was 0.92 and 1.00, respectively, and the positive predictive

value was 0.93 and 1.00, respectively.Consistent with the previously reported discovery cohort

(26), the levels of galectin- 9 and CXCL10 correlated strongly with 3 clinical scores of global or muscle disease activity: the PhGA, the CMAS, and the MMT- 8. The correlation coefficients

for association with either of the biomarkers, which ranged between 0.67 and 0.81 (P < 0.0001), were notably higher than those for CK (rs = 0.32–0.51, P < 0.01) (Figures  1G and H, and Supplementary Figures 1E–G [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). Thus, these results in 2 independent validation cohorts validate galectin- 9 and CXCL10 as strong biomarkers for disease activity in patients with juvenile DM, outperforming the currently used laboratory marker CK.

To assess the international generalizability of galectin- 9 and CXCL10, we tested the biomarkers in a small cohort of patients with juvenile DM from a different geographic region (i.e., Singa-pore). Observations in this cohort confirmed the discriminative potential of galectin- 9 and CXCL10 between active disease and remission, and their levels were comparable to those seen in the IVC (P = 0.0006 for galectin- 9 and P = 0.0025 for CXCL10) (Figures 2A and B).

Disease specificity of galectin- 9 and CXCL10. We next investigated the disease specificity of galectin- 9 and CXCL10 and explored the applicability of each as a biomarker in adult patients with DM or NSM and patients with other systemic autoimmune diseases. The biomarkers were first measured in a cohort of adult patients with DM (n = 36), patients with NSM (n = 14), and patients with EF (n = 18), as well as 43 disease con-trol patients with SMA, a genetic neuromuscular disorder without systemic inflammation, and 22 healthy controls (the characteris-tics of these subjects are listed in Supplementary Table 4 [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). The levels of both galectin- 9 and CXCL10 were elevated in adult patients with active DM (P < 0.0001), patients with NSM (P < 0.0003), and patients with EF (P < 0.05) as compared to healthy controls. Both biomarkers distinguished between active dis-ease and remission in the adult DM cohort (P = 0.0126 and P < 0.0001 for galectin- 9 and CXCL10, respectively), and CXCL10 was also discriminative for disease activity in patients with NSM (P = 0.0139) and those with EF (P = 0.0497) (Figures 2A and B). As expected, the biomarkers were not elevated in control patients with SMA.

A second cohort consisted of pediatric and adult patients with 2 other systemic immune- mediated diseases: localized scleroderma (n = 15) and SLE (n = 36) (the characteristics of these patients are listed in Supplementary Table 5 [http:// onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). In patients with localized scleroderma and those with SLE, the 2 biomarkers did not distinguish significantly between active dis-ease and remission, but galectin- 9 levels in patients with SLE were elevated compared to healthy controls (P = 0.0105) (Fig-ures 2C and D).

Thus, these results demonstrate that galectin- 9 and CXCL10 are applicable as biomarkers for disease activity in both pediatric and adult patients with myositis.

Table 2. Sensitivity, specificity, NPV, and PPV of the determined cutoff values for diagnostic accuracy of galectin- 9 and CXCL10 in the juvenile dermatomyositis internal validation cohort*

Galectin- 9 CXCL10

Cutoff value, pg/ml 19,396 805Sensitivity 0.839 0.871Specificity 0.923 1.000NPV 0.828 0.867PPV 0.929 1.000

* Cutoff values for galectin- 9 and CXCL10 were determined based on the maximal Youden’s Index with a sensitivity of >0.80, in or-der to ensure a low risk of ongoing active inflammation with a bio-marker value below the set cutoff. Only 1 sample per patient per category (active disease or in remission) was included in the analy-sis (i.e., the cohort designated “JDM NL” [Juvenile Dermatomyositis The Netherlands], as shown in Figure 2 and Supplementary Table 5 on the Arthritis & Rheumatology web site at http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). NPV = negative pre-dictive value; PPV = positive predictive value.

GALECTIN- 9 AND CXCL10 AS BIOMARKERS FOR JUVENILE DM |      1383

Figure 2. Biomarker potential of galectin- 9 and CXCL10 in adult inflammatory myopathies and systemic autoimmune diseases with skin involvement. A and B, Galectin- 9 (A) and CXCL10 (B) were measured in serum samples from patients with juvenile dermatomyositis from The Netherlands (JDM NL) (the internal validation cohort [IVC]) and Singapore validation cohort (JDM Sing), adult patients with DM, adult patients with nonspecific myositis (NSM), adult patients with eosinophilic fasciitis (EF), a mixed cohort of adult and juvenile patients with hereditary proximal spinal muscular atrophy (SMA), and adult healthy controls (HC). C and D, Galectin- 9 (C) and CXCL10 (D) were measured in serum samples from the Dutch juvenile DM cohort, juvenile patients with localized scleroderma (LoS), a mixed cohort of juvenile and adult patients with systemic lupus erythematosus (SLE), and adult healthy controls. In A–D, patients were stratified into 2 groups based on disease activity (active [A] or in remission [R] regardless of treatment status). Only 1 sample per patient per activity group was included in the analysis; therefore, the numbers of patients in the IVC differ from those in Figure 1. Data are shown as box plots. Each box represents the interquartile range. Lines inside the boxes represent the median (log scale). Lines outside the boxes represent the 10th and 90th percentiles. Symbols represent individual patients. Multiplicity- adjusted P values above boxes are for comparison between active disease and remission, by 2- way analysis of variance with Šídák’s post hoc test for multiple comparisons. Multiplicity- adjusted P values below boxes are for comparison between each disease group and healthy controls, by Kruskal- Wallis test with Dunnett’s post hoc test for multiple comparisons. P values >0.999 are not shown.

WIENKE ET AL 1384       |

Prospective analysis and flare prediction. To deter-mine the prognostic value of galectin- 9 and CXCL10 during clinical follow- up in patients with juvenile DM, we measured the biomarkers in a prospective cohort of 28 patients, with a median follow- up time of 2.8 years per patient (the charac-

teristics of these patients are listed in Supplementary Table 6 [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). First, we established the biomarker dynamics after diagnosis in 15 patients who reached sustained remission within the first months of treatment and did not have a flare later. The biomarker

0

20000

40000

60000

80000

100000

>12m<12m0

100000

200000

300000

400000

500000

600000

700000

AUC

Gal

ectin

-9

-

0.0026

Time of flare

0.0009

- <12m >12m0

50000

100000

150000

200000

Gal

ectin

-9 a

t dia

gnos

is (p

g/m

L) 0.02540.0728

Time of flareTime since diagnosis (months)0 1 2 3 4 5 6 7 8 9 10 11

0

Gal

ectin

-9 (p

g/m

L)

C

0.02220.07280.0254 <0.0001

0.0012

Flare >12 months (>12m) Flare <12 months (<12m) No flare (-)

0.00890.06160.0814

0

20000

40000

60000

80000

100000

120000

Time since diagnosis (months)

Gal

ectin

-9 (p

g/m

L)

0

2000

4000

6000

8000

CXC

L10

(pg/

mL)

0

200

400

600

8001000

6000

CK

(IU

/L)

E

0 1 2 3 4 5 6 7 8 9 10 11 12

0 1 2 3 4 5 6 7 8 9 10 11 12 12 24 36 48 60 72

Time since diagnosis (months)

Gal

ectin

-9 (p

g/m

L)

A

0

2000

4000

6000

0 1 2 3 4 5 6 7 8 9 10 11 12 12 24 36 48 60 72

Time since diagnosis (months)

CXC

L10

(pg/

mL)

B

0

20000

40000

60000

80000

100000

F

10

100

1000

10000

100000

(IU/L

),(p

g/m

L)

CXCL10

Galectin-9

CK

Patient 1

Months since DxMonths to flare

CMASPhGA

CATPredMTX

0322463--

2.829-31

0.513

3.5293822

0.512

725470

0.50.412

11214803

0.112

17154900-

12

20125200-

12

256.94900-

12

293.4-01-6

320

4745--

33-1.0

-32

1.011

Diagnosis Flare

0

10000

20000

30000

40000

AUC

CXCL

10

0.0997

>12m<12m-

0.0219

Time of flare

- <12m >12m0

2000

4000

6000

8000

10000

CXC

L10

at d

iagn

osis

(pg/

mL) 0.02650.0525

Time of flareTime since diagnosis (months)0 1 2 3 4 5 6 7 8 9 10 11

0

2000

4000

6000

8000

CXC

L10

(pg/

mL)

D

Flare >12 months (>12m) Flare <12 months (<12m) No flare (-)

0.0265 0.11620.1634

0.00070.0441

0.05250.15920.4064

Figure 3. Galectin- 9 and CXCL10 serum levels from longitudinal follow- up of patients with juvenile dermatomyositis (DM) in a prospective cohort. A and B, Dynamics of galectin- 9 (A) and CXCL10 (B) serum levels up to 6 years after juvenile DM diagnosis in 15 patients without flares. The first sample was obtained a maximum of 6 months after treatment start. Both patients with and those without intensification of therapy within the first 3 months were included. Each point contains between 3 and 13 samples, pooled over the time span around the data point. The median interval between 2 samples from a patient was 3.6 months. Per patient, 4–14 samples (median 9) were included. Values are the mean ± SD (linear scale). C and D, Galectin- 9 (C) and CXCL10 (D) serum levels in longitudinal samples from juvenile DM patients with a flare within the first year (<12m) (n = 6), after the first year (>12m) (n = 7), or without flares (n = 15) (same patients as in A and B). Only patients with a first sample obtained a maximum of 6 months after treatment start were included. Left, Longitudinal data (mean ± SD) within the first year. Multiplicity- adjusted P values, by 2- way analysis of variance (ANOVA) with Tukey’s post hoc test, were for flare <12 months versus no flare (top) or flare <12 months versus flare >12 months (bottom) Middle, Galectin- 9 and CXCL10 levels at diagnosis, before treatment start. Data are shown as box plots. Boxes represent the interquartile range, lines inside the boxes show the median, and lines outside the boxes show the 10th and 90th percentiles. Symbols represent individual patients. All P values, by 2- way ANOVA with Tukey’s post hoc test, were corrected. Right, Total area under the receiver operating characteristic curves (AUCs) for each group, calculated by the trapezoidal method. Values are the mean and 95% confidence interval. P values were determined by one- way ANOVA with Tukey’s post hoc test. E, Galectin- 9, CXCL10, and creatine kinase (CK) serum levels in 6 individuals with a flare within the first year after treatment start. In A–F, Gray shading indicates the previously determined cutoffs for galectin- 9 and CXCL10 (19,396 pg/ml and 805 pg/ml, respectively) and the standard cutoff for CK (150 IU/liter). F, Levels of galectin- 9, CXCL10, and CK measured longitudinally in an individual with a disease flare after the first year. Broken horizontal lines indicate the previously determined cutoffs for galectin- 9, CXCL10, and CK. Biomarker levels are shown on a log scale. Shading in the rows for prednisone (Pred) and methotrexate (MTX) represent the relative medication dose (in mg/kg/day for prednisone; in mg/m2/week for MTX). Dark gray shading = high dose; lighter gray shading = low dose. Dx = diagnosis; CMAS = Childhood Myositis Assessment Scale; PhGA = physician’s global assessment; CAT = cutaneous assessment tool.

GALECTIN- 9 AND CXCL10 AS BIOMARKERS FOR JUVENILE DM |      1385

levels quickly declined after the start of treatment, reached lev-els below the previously determined cutoff value within several months, and remained low in remission (the “No flare” group, shown in Figures 3A and B). The biomarker dynamics in patients with a flare after the first year (the “Flare >12 months” group; n = 7) were similar to those in patients without flares (Figures 3C and D). However, patients who experienced a disease flare in the first year after the start of treatment (the “Flare <12 months” group; n = 6) had significantly higher biomarker levels at diagno-sis than did patients with later flares (P = 0.0254 for galectin- 9 and P = 0.0265 for CXCL10) (Figures 3C and D). In addition, these patients who experienced a flare at <12 months had ele-vated biomarker levels over the entire first year (Figures 3C–E). In contrast to the 2 biomarkers, CK activity normalized in 5 of 6 patients (Figure 3E).

To assess the predictive value of the biomarkers for flares after the first year, we analyzed 4 patients for whom longitudinal samples were available within 7 months before a flare (Figure 3F and Supplementary Figure 2 [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). In patients 1 and 2, raised levels of galectin- 9 and CXCL10 (even while remaining below the cutoff level) were observed from up to 7 months prior to the flare, with levels that were above the cutoff value up to 6 months prior to the flare for galectin- 9 and up to 3 months prior to the flare for CXCL10. These biomarker fluctuations were observed even before clinical symptoms of a flare became apparent. In patients 3 and 4, persistently borderline cutoff values were

observed for galectin- 9 and CXCL10 in the 12 months prior to occurrence of a flare, and biomarkers were elevated above the cutoff during the flare. In contrast, CK levels did not increase prior to or during a flare in patient 4, and did not demonstrate an increase until the occurrence of a flare in patients 2 and 3. Only in patient 1 did the CK level steadily increase by 3 months prior to a flare. It was also observed that galectin- 9 and CXCL10 levels stayed high during continued disease activity after the start of the flare in patients receiving medication, while in 3 of 4 individuals, the CK level decreased to within normal limits by the first time point following the start of the clinical flare, despite continued disease activity.

Thus, these results suggest that persistently high or rising galectin- 9 and CXCL10 levels above their cutoff values may be indicative of ongoing (sub)clinical inflammation or an imminent flare, even with a lack of clinical symptoms or elevated CK levels.

Levels of galectin- 9 and CXCL10 in dried blood spots. To facilitate minimally invasive (at- home) biomarker assessment and broad clinical applicability with centralization of diagnostic cores, we assessed galectin- 9 and CXCL10 measurements in dried blood spots and paired plasma and serum samples (the patients’ characteristics are shown in Supplementary Table 7 [http://onlin elibr ary.wiley.com/doi/10.1002/art.40881/ abstract]). Correlation between the biomarker levels in the circulation and biomarker levels in dried blood spots was higher for CXCL10 (rs = 0.93 in plasma and rs = 0.96 in serum) than for galectin- 9 (rs =

Figure 4. Measurement of galectin- 9 and CXCL10 levels in dried blood spots (DBS) as compared to paired plasma and serum samples from patients with active juvenile dermatomyositis (DM). A and B, Correlations between biomarker levels in the plasma (A) and serum (B) and in DBS (on a double log scale) were assessed using Spearman’s correlation coefficients. C, Paired representation of the biomarker levels in the plasma, serum, and DBS from healthy controls (HC), patients with active juvenile DM pretreatment (JDM A), and patients with active juvenile DM while receiving medication (JDM AM) are shown. D, Biomarker levels in DBS were compared between healthy controls and patients with active juvenile DM. P values were determined by Mann- Whitney U test.

WIENKE ET AL 1386       |

0.62 in plasma and rs = 0.58 in serum) (Figures 4A and B). Galec-tin- 9 and CXCL10 levels were similar in the plasma and serum (Figure 4C). Both galectin- 9 and CXCL10, as measured in dried blood spots, were capable of discriminating between patients with active juvenile DM and healthy controls (P = 0.0040 and P < 0.0001, respectively) (Figure 4D), with the healthy control subjects having biomarker levels that were similar to those in patients with juvenile DM in remission (Figure 2). Thus, measurements of both galectin- 9 and CXCL10 in dried blood spots are suitable as bio-markers for juvenile DM disease activity.

DISCUSSION

In this study, galectin- 9 and CXCL10 were validated as strong, reliable, and sensitive biomarkers for disease activity in juvenile DM, and both were identified as promising biomarkers both in adult patients with DM and in adult patients with NSM. The levels of galectin- 9 and CXCL10 strongly distinguished between juvenile DM patients with active disease and juvenile DM patients in remis-sion, even when the patient was receiving immunosuppressive treatment. Furthermore, we showed that galectin- 9 and CXCL10 were relatively specific for autoinflammatory myopathies in adult and pediatric patients, as their levels were not as highly increased or did not differentiate between active disease and remission in other autoimmune diseases such as localized scleroderma and SLE. Both cross- sectionally and longitudinally, galectin- 9 and CXCL10 outperformed CK, which is commonly used as a labo-ratory marker for disease activity and is one of the current criteria for determining clinically inactive disease in juvenile DM (42,51). Continuously elevated or rising biomarker levels, as determined in a prospective patient cohort, may be indicative of an imminent disease flare up to several months before clinical symptoms, even in the absence of elevated CK levels. The biomarkers may there-fore be promising to use in longitudinal follow- up of patients for monitoring of disease activity.

Furthermore, our results showed that galectin- 9 and CXCL10 can be reliably measured in the plasma, serum, and minimally invasive dried blood spots from patients with juvenile DM. It has recently been shown that capillary concentrations of CXCL10 correlate with venous concentrations; for galectin- 9, this has not yet been established (52). The moderate correlation between circulating levels of galectin- 9 and levels of galectin- 9 in dried blood spots could be attributed to either liberation of intra-cellularly stored galectin- 9 and/or release from its carrier proteins upon elution and dilution.

This study has several strengths. Although many biomark-ers are being identified for a variety of diseases, only a few have been implemented into clinical practice, due to a lack of reproducibility and diagnostic accuracy. However, the levels of galectin- 9 and CXCL10 have a high discriminative power and strong, reproducible correlation with disease activity. Thanks

to a large international collaborative effort, and despite the rar-ity of the disease, we have been able to extensively validate galectin- 9 and CXCL10 as biomarkers in a large number of patients with juvenile DM from 3 independent cross- sectional cohorts. The additional analyses in a prospective cohort of patients with juvenile DM with a long follow- up time added important information on the value of galectin- 9 and CXCL10 in clinical follow- up. In addition to the clinical validation in this study, the biomarkers have undergone a technical validation at the diagnostic department of the UMC Utrecht, which has demonstrated the stability of the biomarkers and reproducibil-ity of the measurements. In addition, we have explored a mini-mally invasive diagnostic method of measuring the biomarkers in dried blood spots.

The findings of this study need to be interpreted carefully, tak-ing into account the observational nature of the data and the use of a combination of clinical scores and CK levels (the Paediatric Rheumatology International Trials Organisation criteria for clinically inactive disease in juvenile DM) as the gold standard for assess-ment of disease activity in juvenile DM (42,51). Importantly, mea-surement of galectin- 9 and CXCL10 levels can complement, but not replace, clinical assessment by experienced health care pro-fessionals. However, both biomarkers outperformed the currently used marker, CK, a finding that underscores the gains that can be achieved by introducing the new biomarkers into clinical practice.

A recent study using the SOMAscan assay also identified both galectin- 9 and CXCL10 among the top up- regulated proteins in juvenile DM, correlating with disease activity as assessed by the PhGA (53). CXCL10 levels were previously shown to correlate with disease activity in juvenile DM (26,30–32,54), and CXCL10 is well known to be an interferon- inducible chemokine that can be ele-vated in other types of myositis and autoimmune diseases (29,33). In our study, galectin- 9 was a specific biomarker for inflammatory myopathies. In patients with juvenile DM, high circulating interfer-on- α levels have been found, and in one group of patients with juvenile DM, more than 75% of patients had a positive interferon signature (55,56). Circulating galectin- 9 and CXCL10 levels could therefore be a direct reflection of active, interferon- driven inflam-mation, which is supported by a recent study in which galectin- 9 was demonstrated to be a marker for the interferon signature in SLE and antiphospholipid syndrome (57).

Since the levels of these biomarkers are known to correlate with the extent of disease activity in various types of tissue, local tissue cells are the main candidate producers of the proteins. Indeed, galectin- 9 can be detected not only in the circulation, but also locally within inflamed muscle and skin, where it is mainly present in activated tissue macrophages and capillary endothe-lial cells (data not shown). A similar expression pattern, in tissue mononuclear cells and endothelial cells, was previously demon-strated for CXCL10 (58,59). Local biomarker production within the inflamed tissue is consistent with our previous observation that

GALECTIN- 9 AND CXCL10 AS BIOMARKERS FOR JUVENILE DM |      1387

the biomarker levels slowly decline after stem cell transplantation, as tissue- infiltrating immune cells (and endothelial cells) are likely to be less affected by immune- ablative preconditioning than are circulating immune cells (27).

Implementation of galectin- 9 and CXCL10 into clinical prac-tice, as tools to monitor disease activity and guide treatment, might enable personalized treatment strategies for patients with juvenile DM. It is an advantage that both biomarkers performed equally well in our study, suggesting that diagnostic centers can decide to use their biomarker of choice depending on its availability and feasibility. Biomarker levels below the set cutoff value reflect the absence of disease activity, which could allow tapering of immunosuppressive medication. Rising or persistently high levels might be indicative of an insufficient response to therapy and/or an imminent flare, even in the absence of clinical symptoms or elevated CK levels, possibly reflecting subclinical inflammation. Elevated biomarker levels might therefore indicate the need for intensification of treatment or slower tapering of steroids. With this envisioned personalized treatment strategy, we could respond to important patient- reported needs: a recently conducted patient survey by Cure JM, a US patient organization for juvenile myositis, has shown that “predictors for disease flares” and “new treatments, less side effects” are 2 of the top 3 research priorities chosen by patients (60).

Galectin- 9 and CXCL10 may also provide an objective out-come measure for response to therapy in future clinical trials that would be assessing novel therapeutics. Our study has shown that galectin- 9 and CXCL10 levels in dried blood spots correlate with venous levels and could differentiate patients with active juvenile DM from healthy controls. Longitudinal assessment of these bio-markers via dried blood spots, which requires further study, has potential for high utility in the future, since dried blood spots can be sampled at home by simple capillary finger- prick. Since protein levels in dried blood spots remain remarkably stable over time, even at room temperature (61,62), samples of dried blood spots can be sent to a diagnostic center through regular mail. This enables at- home diagnostics and centralization of diagnos-tic cores for both clinical care and multicenter studies. It also ensures maximum accessibility of the biomarker measurements for non–expert medical centers, which can also facilitate care in rural areas.

Galectin- 9 and CXCL10 measurements could add impor-tant information to the complex differential diagnosis of mus-cle symptoms during follow- up, and might aid in discriminating between steroid- induced myopathy, noninflammatory muscle pain, and muscle inflammation, all of which require different treatment strategies. In these complicated cases, in particular, the biomarkers may also help abrogate the need for invasive diagnostic muscle biopsy or costly magnetic resonance imag-ing scans, which can sometimes require sedation in young chil-dren. This specific potential use of these biomarkers will have to be further investigated in additional prospective studies. In addi-tion, future prospective studies will have to point out 1) whether

one biomarker may be superior to the other in answering spe-cific clinical questions concerning juvenile DM, 2) whether the biomarkers are able to detect mild disease activity, 3) whether the biomarkers also have prognostic value in adult patients with myositis, and 4) whether biomarker- guided disease manage-ment will improve the outcomes in patients with juvenile DM.

In conclusion, galectin- 9 and CXCL10 were identified and extensively validated as strong, reliable, and sensitive biomark-ers for disease activity in juvenile DM. Measurement of these biomarkers might facilitate personalized treatment strategies for patients with juvenile DM, by providing a diagnostic monitoring tool to guide treatment.

ACKNOWLEDGMENTS

We are grateful to all members of the Juvenile Dermat-omyositis Research Group (JDRG) who contributed to this study (see https ://www.juven ilede rmato myosi tis.org.uk/about- jdrg/colla borat ions). We also thank the pediatric rheumatol-ogy research group at KK Women’s and Children’s Hospital, Singapore, the neurology department at the AMC, Amster-dam, the rheumatology department at the University Hospi-tal, Strasbourg, and the research groups within the UMC Utrecht and the University Hospital Centre, Lausanne, for kindly providing the samples to perform this study. We thank the Dutch Juvenile DM network and, especially, Annette van Dijk- Hummelman and Ellen Schatorjé for their help and sup-port in the patient inclusion and sample collection. We also thank Ester van Leeuwen for helping us with the logistics in obtaining samples from adult patients with myositis, and staff at the Luminex core facility for performing all biomarker mea surements. A special thanks goes to the UK JDRG Lon-don for providing blood samples from patients, and to the board members of the patient group Myositis of the VSN (Vereniging Spierziekten Nederland), the Stichting KAISZ (Kinderen met Auto- ImmuunSysteemZiekten), and the Bas Stichting (a Dutch juvenile DM patient organization) for their explicit and continuous support of our biomarker research in juvenile DM.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it criti-cally for important intellectual content, and all authors approved the final version to be published. Drs. van Royen- Kerkhof and van Wijk had full access to all of the data in the study and take responsibility for the integ-rity of the data and the accuracy of the data analysis.Study conception and design. Wienke, Bellutti Enders, Lim, Otten, Fritsch- Stork, de Jager, de Visser, Arkachaisri, van der Kooi, Nierkens, Wedderburn, van Royen- Kerkhof, van Wijk.Acquisition of data. Wienke, Bellutti Enders, Lim, Mertens, van den Hoogen, Wijngaarde, Yeo, Meyer, Otten, Fritsch- Stork, Kamphuis, Hop-penreijs, Armbrust, van den Berg, Hissink Muller, Tekstra, Hoogendijk, Deakin, van Roon, van der Pol, Nistala, Pilkington, Arkachaisri, Radstake, van der Kooi, Nierkens, Wedderburn, van Royen- Kerkhof, van Wijk.

WIENKE ET AL 1388       |

Analysis and interpretation of data. Wienke, Bellutti Enders, Lim, Mertens, de Jager, de Visser, van der Kooi, Nierkens, Wedderburn, van Royen- Kerkhof, van Wijk.

REFERENCES 1. Feldman BM, Rider LG, Reed AM, Pachman LM. Juvenile dermato-

myositis and other idiopathic inflammatory myopathies of childhood. Lancet 2008;371:2201–12.

2. Meyer A, Meyer N, Schaeffer M, Gottenberg JE, Geny B, Sibilia J. Incidence and prevalence of inflammatory myopathies: a systematic review. Rheumatology (Oxford) 2015;54:50–63.

3. Pachman LM, Lipton R, Ramsey-Goldman R, Shamiyeh E, Abbott K, Mendez EP, et al. History of infection before the onset of juve-nile dermatomyositis: results from the National Institute of Arthritis and Musculoskeletal and Skin Diseases Research Registry. Arthritis Rheum 2005;53:166–72.

4. Shah M, Targoff IN, Rice MM, Miller FW, Rider LG, with the Child-hood Mytosis Heterogeneity Collaborative Study Group. Ultravio-let radiation exposure is associated with clinical and autoantibody phenotypes in juvenile myositis. Arthritis Rheum 2013;65:1934–41.

5. Vegosen LJ, Weinberg CR, O’Hanlon TP, Targoff IN, Miller FW, Rider LG. Seasonal birth patterns in myositis subgroups suggest an etiologic role of early environmental exposures. Arthritis Rheum 2007;56:2719–28.

6. López de Padilla CM, Vallejo AN, Lacomis D, McNallan K, Reed AM. Extranodal lymphoid microstructures in inflamed muscle and disease severity of new- onset juvenile dermatomyositis. Arthritis Rheum 2009;60:1160–72.

7. Wedderburn LR, Varsani H, Li CK, Newton KR, Amato AA, Banwell B, et al. International consensus on a proposed score system for muscle biopsy evaluation in patients with juvenile dermatomyositis: a tool for potential use in clinical trials. Arthritis Rheum 2007;57:1192–201.

8. Vercoulen Y, Bellutti Enders F, Meerding J, Plantinga M, Elst EF, Varsani H, et al. Increased presence of FOXP3+ regulatory T cells in inflamed muscle of patients with active juvenile dermatomyositis compared to peripheral blood. PLoS One 2014;9:e105353.

9. Shrestha S, Wershil B, Sarwark JF, Niewold TB, Philipp T, Pachman LM. Lesional and nonlesional skin from patients with untreated juvenile dermatomyositis displays increased numbers of mast cells and ma-ture plasmacytoid dendritic cells. Arthritis Rheum 2010;62:2813–22.

10. Enders FB, Bader-Meunier B, Baildam E, Constantin T, Dolezalova P, Feldman BM, et al. Consensus- based recommendations for the management of juvenile dermatomyositis. Ann Rheum Dis 2017;76:329–40.

11. Huber AM, Giannini EH, Bowyer SL, Kim S, Lang B, Lindsley CB, et al. Protocols for the initial treatment of moderately severe juve-nile dermatomyositis: results of a Children’s Arthritis and Rheuma-tology Research Alliance consensus conference. Arthritis Care Res (Hoboken) 2010;62:219–25.

12. Huber AM, Robinson AB, Reed AM, Abramson L, Bout-Tabaku S, Carrasco R, et al. Consensus treatments for moderate ju-venile dermatomyositis: beyond the first two months. Results of the second Childhood Arthritis and Rheumatology Research Alliance consensus conference. Arthritis Care Res (Hoboken) 2012;64:546–53.

13. Huber AM, Lang B, LeBlanc CM, Birdi N, Bolaria RK, Malleson P, et al. Medium- and long- term functional outcomes in a multicenter cohort of children with juvenile dermatomyositis. Arthritis Rheum 2000;43:541–9.

14. Gowdie PJ, Allen RC, Kornberg AJ, Akikusa JD. Clinical features and disease course of patients with juvenile dermatomyositis. Int J Rheum Dis 2013;16:561–7.

15. Ravelli A, Trail L, Ferrari C, Ruperto N, Pistorio A, Pilkington C, et al. Long- term outcome and prognostic factors of juvenile dermatomy-ositis: a multinational, multicenter study of 490 patients. Arthritis Care Res (Hoboken) 2010;62:63–72.

16. Ravelli A, Lattanzi B, Consolaro A, Martini A. Glucocorticoids in pae-diatric rheumatology. Clin Exp Rheumatol 2011;29 Suppl 68:S148–52.

17. Santiago RA, Silva CA, Caparbo VF, Sallum AM, Pereira RM. Bone mineral apparent density in juvenile dermatomyositis: the role of lean body mass and glucocorticoid use. Scand J Rheumatol 2008;37:40–7.

18. Shiff NJ, Brant R, Guzman J, Cabral DA, Huber AM, Miettunen P, et al. Glucocorticoid- related changes in body mass index among children and adolescents with rheumatic diseases. Arthritis Care Res (Hoboken) 2013;65:113–21.

19. Wienke J, Deakin CT, Wedderburn LR, van Wijk F, van Royen- Kerkhof A. Systemic and tissue inflammation in juvenile dermato-myositis: from pathogenesis to the quest for monitoring tools. Front Immunol 2018;9:2951.

20. Rider LG, Aggarwal R, Pistorio A, Bayat N, Erman B, Feldman BM, et al. 2016 American College of Rheumatology/European League Against Rheumatism criteria for minimal, moderate, and major clin-ical response in juvenile dermatomyositis: an International Myositis Assessment and Clinical Studies Group/Paediatric Rheumatolo-gy International Trials Organisation collaborative initiative. Arthritis Rheumatol 2017;69:911–23.

21. Ruperto N, Ravelli A, Pistorio A, Ferriani V, Calvo I, Ganser G, et al. The provisional Paediatric Rheumatology International Trials Or-ganisation/American College of Rheumatology/European League Against Rheumatism disease activity core set for the evaluation of response to therapy in juvenile dermatomyositis: a prospective vali-dation study. Arthritis Rheum 2008;59:4–13.

22. Rider LG, Werth VP, Huber AM, Alexanderson H, Rao AP, Ruper-to N, et al. Measures of adult and juvenile dermatomyositis, poly-myositis, and inclusion body myositis: Physician and Patient/Parent Global Activity, Manual Muscle Testing (MMT), Health Assessment Questionnaire (HAQ)/Childhood Health Assessment Questionnaire (C- HAQ), Childhood Myositis Assessment Scale (CMAS), Myositis Disease Activity Assessment Tool (MDAAT), Disease Activity Score (DAS), Short Form 36 (SF- 36), Child Health Questionnaire (CHQ), Physician Global Damage, Myositis Damage Index (MDI), Quantita-tive Muscle Testing (QMT), Myositis Functional Index- 2 (FI- 2), Myosi-tis Activities Profile (MAP), Inclusion Body Myositis Functional Rating Scale (IBMFRS), Cutaneous Dermatomyositis Disease Area and Severity Index (CDASI), Cutaneous Assessment Tool (CAT), Dermat-omyositis Skin Severity Index (DSSI), Skindex, and Dermatology Life Quality Index (DLQI). Arthritis Care Res 2011;63 Suppl 11:S118–57.

23. Guzmán J, Petty RE, Malleson PN. Monitoring disease activity in ju-venile dermatomyositis: the role of von Willebrand factor and muscle enzymes. J Rheumatol 1994;21:739–43.

24. Rider LG. Assessment of disease activity and its sequelae in children and adults with myositis. Curr Opin Rheumatol 1996;8:495–506.

25. McCann LJ, Juggins AD, Maillard SM, Wedderburn LR, Davidson JE, Murray KJ, et al. The Juvenile Dermatomyositis Na-tional Registry and Repository (UK and Ireland): clinical charac-teristics of children recruited within the first 5 yr. Rheumatology (Oxford) 2006;45:1255–60.

26. Bellutti Enders F, van Wijk F, Scholman R, Hofer M, Prakken BJ, van Royen-Kerkhof A, et al. Correlation of CXCL10, tumor necrosis factor receptor type II, and galectin 9 with disease activity in juvenile dermatomyositis. Arthritis Rheumatol 2014;66:2281–9.

27. Enders FB, Delemarre EM, Kuemmerle-Deschner J, van der Torre P, Wulffraat NM, Prakken BP, et al. Autologous stem cell transplanta-

GALECTIN- 9 AND CXCL10 AS BIOMARKERS FOR JUVENILE DM |      1389

tion leads to a change in proinflammatory plasma cytokine profile of patients with juvenile dermatomyositis correlating with disease activ-ity [letter]. Ann Rheum Dis 2015;74:315–7.

28. Merani S, Chen W, Elahi S. The bitter side of sweet: the role of ga-lectin- 9 in immunopathogenesis of viral infections. Rev Med Virol 2015;25:175–86.

29. Antonelli A, Ferrari SM, Giuggioli D, Ferrannini E, Ferri C, Fallahi P. Chemokine (C- X- C motif) ligand (CXCL)10 in autoimmune diseases. Autoimmun Rev 2014;13:272–80.

30. Reed AM, Peterson E, Bilgic H, Ytterberg SR, Amin S, Hein MS, et al. Changes in novel biomarkers of disease activity in juvenile and adult dermatomyositis are sensitive biomarkers of disease course. Arthritis Rheum 2012;64:4078–86.

31. Bilgic H, Ytterberg SR, Amin S, McNallan KT, Wilson JC, Koeuth T, et  al. Interleukin- 6 and type I interferon–regulated genes and chemokines mark disease activity in dermatomyositis. Arthritis Rheum 2009;60:3436–46.

32. Baechler EC, Bauer JW, Slattery CA, Ortmann WA, Espe KJ, Novitzke J, et al. An interferon signature in the peripheral blood of dermatomyositis patients is associated with disease activity. Mol Med 2007;13:59–68.

33. Huard C, Gullà SV, Bennett DV, Coyle AJ, Vleugels RA, Greenberg SA. Correlation of cutaneous disease activity with type 1 interferon gene signature and interferon β in dermatomyositis. Br J Dermatol 2017;176:1224–30.

34. Thijssen VL, Heusschen R, Caers J, Griffioen AW. Galectin ex-pression in cancer diagnosis and prognosis: a systematic review. Biochim Biophys Acta 2015;1855:235–47.

35. Panda SK, Facchinetti V, Voynova E, Hanabuchi S, Karnell JL, Hanna RN, et al. Galectin- 9 inhibits TLR7- mediated autoimmunity in murine lupus models. J Clin Invest 2018;128:1873–87.

36. Zeggar S, Watanabe KS, Teshigawara S, Hiramatsu S, Katsuyama T, Katsuyama E, et al. Role of Lgals9 deficiency in attenuating nephri-tis and arthritis in BALB/c mice in a pristane- induced lupus model. Arthritis Rheumatol 2018;70:1089–101.

37. Bohan A, Peter JB. Polymyositis and dermatomyositis (first of two parts). N Engl J Med 1975;292:344–7.

38. Bohan A, Peter JB. Polymyositis and dermatomyositis (second of two parts). N Engl J Med 1975;292:403–7.

39. Rider LG, Feldman BM, Perez MD, Rennebohm RM, Lindsley CB, Zemel LS, et al, in cooperation with the Juvenile Dermatomyositis Disease Activity Collaborative Study Group. Development of validat-ed disease activity and damage indices for the juvenile idiopathic inflammatory myopathies. I. Physician, parent, and patient global assessments. Arthritis Rheum 1997;40:1976–83.

40. Rider LG, Koziol D, Giannini EH, Jain MS, Smith MR, Whitney-Mahoney K, et al. Validation of Manual Muscle Testing and a subset of eight muscles for adult and juvenile idiopathic inflammatory myopathies. Arthritis Care Res (Hoboken) 2010;62:465–72.

41. Huber AM, Dugan EM, Lachenbruch PA, Feldman BM, Perez MD, Zemel LS, et al. The cutaneous assessment tool: development and reliability in juvenile idiopathic inflammatory myopathy. Rheumatolo-gy (Oxford) 2007;46:1606–11.

42. Almeida B, Campanilho-Marques R, Arnold K, Pilkington CA, Wedderburn LR. Nistala K, on behalf of the Juvenile Dermatomy-ositis Research Group. Analysis of published criteria for clinically inactive disease in a large juvenile dermatomyositis cohort shows that skin disease is underestimated. Arthritis Rheumatol 2015;67: 2495–502.

43. Hoogendijk JE, Amato AA, Lecky BR, Choy EH, Lundberg IE, Rose MR, et al. 119th ENMC International Workshop: trial design in adult

idiopathic inflammatory myopathies, with the exception of inclusion body myositis, 10- 12 October 2003, Naarden, The Netherlands. Neuromuscul Disord 2004;14:337–45.

44. Medical Research Council. Aids to the examination of the peripher-al nervous system, Memorandum no. 45, Her Majesty’s Stationery Office. London; 1981.

45. Hochberg MC. Updating the American College of Rheumatology re-vised criteria for the classification of systemic lupus erythematosus [letter]. Arthritis Rheum 1997;40:1725.

46. Yee CS, Farewell VT, Isenberg DA, Griffiths B, Teh LS, Bruce IN, et  al. The use of Systemic Lupus Erythematosus Disease Activity Index- 2000 to define active disease and minimal clinically meaningful change based on data from a large cohort of systemic lupus erythe-matosus patients. Rheumatology (Oxford) 2011;50:982–8.

47. Arkachaisri T, Vilaiyuk S, Li S, O’Neil KM, Pope E, Higgins GC, et al. The Localized Scleroderma Skin Severity Index and physician global assessment of disease activity: a work in progress toward devel-opment of localized scleroderma outcome measures. J Rheumatol 2009;36:2819–29.

48. Lunn MR, Wang CH. Spinal muscular atrophy. Lancet 2008;371: 2120–33.

49. De Jager W, Prakken BJ, Bijlsma JW, Kuis W, Rijkers GT. Improved multiplex immunoassay performance in human plasma and synovial fluid following removal of interfering heterophilic antibodies. J Immu-nol Methods 2005;300:124–35.

50. Scholman RC, Giovannone B, Hiddingh S, Meerding JM, Malvar Fernandez B, van Dijk ME, et al. Effect of anticoagulants on 162 circulating immune related proteins in healthy subjects. Cytokine 2018;106:114–24.

51. Lazarevic D, Pistorio A, Palmisani E, Miettunen P, Ravelli A, Pilkington C, et al. The PRINTO criteria for clinically inactive disease in juvenile dermatomyositis. Ann Rheum Dis 2013;72:686–93.

52. Neesgaard B, Ruhwald M, Krarup HB, Weis N. Determination of anti- HCV and quantification of HCV- RNA and IP- 10 from dried blood spots sent by regular mail. PLoS One 2018;13:e0201629.

53. Kim H, Biancotto A, Cheung F, O’Hanlon TP, Targoff IN, Huang Y, et al. Novel serum broad- based proteomic discovery analysis identi-fies proteins and pathways dysregulated in juvenile dermato myositis (JDM) myositis autoantibody groups [abstract]. Arthritis Rheuma-tol 2017;69 Suppl 10. URL: https ://acrab strac ts.org/abstr act/ novel-serum-broad-based-prote omic-disco very-analy sis-ident ifies- prote ins-and-pathw ays-dysre gulat ed-in-juven ile-derma tomyo sitis- jdm-and-myosi tis-autoa ntibo dy-group s/.

54. Sanner H, Schwartz T, Flatø B, Vistnes M, Christensen G. Sjaastad I. Increased levels of eotaxin and MCP- 1 in juvenile dermatomyositis median 16.8 years after disease onset; associa-tions with disease activity, duration and organ damage. PLoS One 2014;9:e92171.

55. Rice GI, Melki I, Frémond ML, Briggs TA, Rodero MP, Kitabayashi N, et al. Assessment of type I interferon signaling in pediatric inflamma-tory disease. J Clin Immunol 2017;37:123–32.

56. Rodero MP, Decalf J, Bondet V, Hunt D, Rice GI, Werneke S, et al. Detection of interferon α protein reveals differential levels and cellular sources in disease. J Exp Med 2017;214:1547–55.

57. Van den Hoogen LL, van Roon JA, Mertens JS, Wienke J, Lopes AP, de Jager W, et al. Galectin- 9 is an easy to measure biomarker for the interferon signature in systemic lupus erythematosus and antiphos-pholipid syndrome. Ann Rheum Dis 2018;77:1810–4.

58. Fall N, Bove KE, Stringer K, Lovell DJ, Brunner HI, Weiss J, et al. Association between lack of angiogenic response in muscle tissue and high expression of angiostatic ELR- negative CXC chemokines in

WIENKE ET AL 1390       |

patients with juvenile dermatomyositis: possible link to vasculopathy. Arthritis Rheum 2005;52:3175–80.

59. Limongi F. The CXCR59 chemokines in inflammatory myopathies. Clin Ter 2015;166:e56–61.

60. Dave M, for Cure JM Foundation. Patient perspectives on juvenile myo sitis research priorities. URL: http://www.curejm.org/medic al_ profe ssion als/pdfs/2017-05-04 JM-Research-Priorities.pdf.

61. McDade TW, Williams S, Snodgrass JJ. What a drop can do: dried blood spots as a minimally invasive method for integrating biomark-ers into population- based research. Demography 2007;44:899–925.

62. De Jesus VR, Zhang XK, Keutzer J, Bodamer OA, Mühl A, Orsini JJ, et al. Development and evaluation of quality control dried blood spot materials in newborn screening for lysosomal storage disorders. Clin Chem 2009;55:158–64.

1391

Arthritis & RheumatologyVol. 71, No. 8, August 2019, pp 1391–1392© 2019, American College of Rheumatology

L E T T E R

DOI 10.1002/art.40914

Excess deaths upon cessation of xanthine oxidase inhibitor treatment—data from the Cardiovascular Safety of Febuxostat and Allopurinol in Patients With Gout and Cardiovascular Morbidities trial: comment on the article by Choi et al

To the Editor:Choi et  al (1) reviewed implications of data from the Car­

diovascular Safety of Febuxostat and Allopurinol in Patients With Gout and Cardiovascular Morbidities (CARES) trial comparing the cardiovascular (CV) safety of febuxostat and allopurinol in a ran­domized study of 6,190 patients with gout and CV disease (2). They noted that nearly 85% of deaths occurred while subjects were not receiving therapy. More detailed inspection of the data reveals that 37 deaths attributed to CV disease occurred during therapy. Within 30 days of discontinuation of therapy, among the 56.6% of patients who discontinued treatment prematurely, there were an additional 66 deaths attributed to CV disease. Given a median duration of therapy of 724 days during which treated patients were at risk of death, this imbalance in gross numbers of deaths is notable and suggests that discontinuation of xanthine oxidase inhibitors may be dangerous. Comparing the rate of CV­ related death among the 56.6% of patients who had discontinued therapy for 30 days to that among the 100% of patients who were receiving therapy for a median of 724 days (used as an estimate of the mean in this calculation), the relative rate of death was 76 times higher after discontinuation. Relative event rates were also high for major CV events and all­ cause mortality occurring within 30 days after discontinuation of treatment compared to those same events occurring during therapy (Table 1). Interestingly, the reported excess of deaths seen with febuxostat relative to allopu­rinol (2) is not reflected in the high rates of death seen shortly after

discontinuation of therapy, with very high rates of death occurring

with discontinuation of either therapy.Why is the risk of death so high after discontinuation of

xanthine oxidase inhibitor therapy? Drug discontinuation syn­dromes are common for drugs with CV activity (3), although the event rates are much lower than observed here. Most patients in the CARES trial stopped therapy due to voluntary withdrawal (19.2%), separate from those with adverse events. The adverse events leading to 6.2% of the discontinuations are likely the same adverse events reported in the package insert for febuxostat, including minor liver function abnormalities, nausea, rash, and arthralgia, all reported to occur at a frequency similar to that observed with allopurinol, and none of them with an apparent causal relationship to mortality. Others have investigated the effects of discontinuing xanthine oxidase inhibitor treatment, not­ing a high rate of gout relapse and an increase in blood pressure and decline in renal function if renin­ angiotensin–blocking (RAB) agents are not taken (4), but no studies have been performed in a comparably high­ risk population. A high proportion of patients in the CARES trial were receiving RAB agents at the baseline visit (69.9%), and it may be informative to investigate whether mortal­ity during the 30 days after discontinuation of an xanthine oxidase inhibitor is higher in patients who never took or who discontinued RAB therapy. Even without a causal explanation, the observed high risk of discontinuation of xanthine oxidase inhibition in the CARES patient population is clinically important, with the magni­tude of this risk (~7,000%) greatly eclipsing that of the observed differences between febuxostat and allopurinol. Moreover, the high rates of drug discontinuation in the study population reflect an even higher rate of nonadherence with gout therapy in routine care (5), suggesting that this risk, if confirmed, will need to be factored into treatment recommendations.

Table  1. Relative event rates during the 30 days after discontinuation of xanthine oxidase treatment compared to during treatment*

Febuxostat Allopurinol Combined data

Median days receiving therapy 728 719 724% of patients who discontinued therapy 57.3 55.9 56.6Primary end point of major cardiovascular

events, relative rate11.3 8.4 10.0

Cardiovascular death, relative rate 71.8 82.7 76.1Death from any cause, relative rate 65.9 71.5 68.3Death from any cause with inclusion of 199

additional deaths identified retrospec-tively, relative rate

68.2 70.4 69.3

* Event rate = (number of events)/(days exposed × fraction exposed to those days).

LETTER 1392       |

Michael R. Bubb, MDMalcom Randall Department of Veterans

Affairs Medical Center and University of Florida College of Medicine

Gainesville, FL

1. Choi H, Neogi T, Stamp L, Dalbeth N, Terkeltaub R. Implications of the Cardiovascular Safety of Febuxostat and Allopurinol in Patients With Gout and Cardiovascular Morbidities (CARES) trial and the associated Food and Drug Administration public safety alert. Arthritis Rheumatol 2018;70:1702–9.

2. White WB, Saag KG, Becker MA, Borer JS, Gorelick PB, Whelton A, et al. Cardiovascular safety of febuxostat or allopurinol in patients with gout. N Engl J Med 2018;378:1200–10.

3. Reidenberg MM. Drug discontinuation effects are part of the pharmacology of a drug. J Pharmacol Exp Ther 2011;339:324–8.

4. Beslon V, Moreau P, Maruani A, Maisonneuve H, Giraudeau B, Fournier JP. Effects of discontinuation of urate­ lowering therapy: a systematic review. J Gen Intern Med 2018;33:358–66.

5. Scheepers LE, van Onna M, Stehouwer CD, Singh JA, Arts IC, Boonen A. Medication adherence among patients with gout: a systematic review and meta­ analysis. Semin Arthritis Rheum 2018;47:689–702.

DOI 10.1002/art.40907

Clinical Images: Dual- energy computed tomography for the noninvasive diagnosis of coexisting gout and calcium pyrophosphate deposition disease

The patient, an 82­ year­ old man with gout, calcium pyrophosphate deposition (CPPD) disease, alcoholic liver disease, and hyperten­sion, presented with acute­ onset pain in both hands. His regular treatment included allopurinol with colchicine and oral or intraarticular prednisolone. Physical examination demonstrated swelling and tenderness in the left wrist and second and third metacarpophalangeal (MCP) joints, with overlying erythema. Subcutaneous tophi were palpable bilaterally, notably on the dorsal aspect of the third MCP joints. Laboratory tests revealed markers of inflammation (elevated levels of C­ reactive protein, erythrocyte sedimentation rate, and white blood cell count) and prerenal acute kidney injury. Serum urate, calcium, and phosphate levels were normal. Synovial fluid (SF) analysis of the left third MCP joint identified monosodium urate monohydrate (MSU) crystals, with no evidence of CPP crystals or septic arthritis. Radiography demonstrated linear calcifications in the right ulnocarpal joint and possibly left third MCP joint (A). Joint space narrowing was observed in the second and third MCP joints, with dense soft tissue masses and subtle bone erosions. Dual­ energy computed tomography (DECT) was performed, revealing coexisting intra­ and juxtaarticular MSU and CPP crystal deposits, most notably in the left third MCP joint (B). Because initial treatment with colchicine and oral prednisolone was not effective, anakinra (100 mg daily) was administered subcutaneously for 3 days, providing rapid symptom relief. Although the identification of characteristic MSU and CPP crystals in SF or tophi remains the standard for the diagnosis of gout and CPPD (1,2), DECT and multi­ energy photon­ counting CT have the potential to become clinically relevant by enabling noninvasive detection, characterization, quantification, and mapping of MSU and CPP crystal deposition (3).

Rami Hajri, MD Steven D. Hajdu, MD Thomas Hügle, MD, PhD Pascal Zufferey, MDLausanne University Hospital and University of LausanneLaurent Guiral, MScGE Healthcare Buc, FranceFabio Becce, MDLausanne University Hospital and University of LausanneLausanne, Switzerland

1. Richette P, Doherty M, Pascual E, Barskova V, Becce F, Castaneda J, et al. 2018 updated European League Against Rheumatism evidence­based recommendations for the diagnosis of gout. Ann Rheum Dis 2019. E­pub ahead of print.

2. Zhang W, Doherty M, Bardin T, Barskova V, Guerne PA, Jansen TL, et al. European League Against Rheumatism recommendations for calcium pyrophosphate deposition. Part I: terminology and diagnosis. Ann Rheum Dis 2011;70:563–70.

3. Stamp LK, Anderson NG, Becce F, Rajeswari M, Polson M, Guyen O, et al. Clinical utility of multi­ energy spectral photon­ counting computed tomography in crystal arthritis. Arthritis Rheumatol 2019;71:1158–62.

LETTER 1392       |

Michael R. Bubb, MDMalcom Randall Department of Veterans

Affairs Medical Center and University of Florida College of Medicine

Gainesville, FL

1. Choi H, Neogi T, Stamp L, Dalbeth N, Terkeltaub R. Implications of the Cardiovascular Safety of Febuxostat and Allopurinol in Patients With Gout and Cardiovascular Morbidities (CARES) trial and the associated Food and Drug Administration public safety alert. Arthritis Rheumatol 2018;70:1702–9.

2. White WB, Saag KG, Becker MA, Borer JS, Gorelick PB, Whelton A, et al. Cardiovascular safety of febuxostat or allopurinol in patients with gout. N Engl J Med 2018;378:1200–10.

3. Reidenberg MM. Drug discontinuation effects are part of the pharmacology of a drug. J Pharmacol Exp Ther 2011;339:324–8.

4. Beslon V, Moreau P, Maruani A, Maisonneuve H, Giraudeau B, Fournier JP. Effects of discontinuation of urate­ lowering therapy: a systematic review. J Gen Intern Med 2018;33:358–66.

5. Scheepers LE, van Onna M, Stehouwer CD, Singh JA, Arts IC, Boonen A. Medication adherence among patients with gout: a systematic review and meta­ analysis. Semin Arthritis Rheum 2018;47:689–702.

DOI 10.1002/art.40907

Clinical Images: Dual- energy computed tomography for the noninvasive diagnosis of coexisting gout and calcium pyrophosphate deposition disease

The patient, an 82­ year­ old man with gout, calcium pyrophosphate deposition (CPPD) disease, alcoholic liver disease, and hyperten­sion, presented with acute­ onset pain in both hands. His regular treatment included allopurinol with colchicine and oral or intraarticular prednisolone. Physical examination demonstrated swelling and tenderness in the left wrist and second and third metacarpophalangeal (MCP) joints, with overlying erythema. Subcutaneous tophi were palpable bilaterally, notably on the dorsal aspect of the third MCP joints. Laboratory tests revealed markers of inflammation (elevated levels of C­ reactive protein, erythrocyte sedimentation rate, and white blood cell count) and prerenal acute kidney injury. Serum urate, calcium, and phosphate levels were normal. Synovial fluid (SF) analysis of the left third MCP joint identified monosodium urate monohydrate (MSU) crystals, with no evidence of CPP crystals or septic arthritis. Radiography demonstrated linear calcifications in the right ulnocarpal joint and possibly left third MCP joint (A). Joint space narrowing was observed in the second and third MCP joints, with dense soft tissue masses and subtle bone erosions. Dual­ energy computed tomography (DECT) was performed, revealing coexisting intra­ and juxtaarticular MSU and CPP crystal deposits, most notably in the left third MCP joint (B). Because initial treatment with colchicine and oral prednisolone was not effective, anakinra (100 mg daily) was administered subcutaneously for 3 days, providing rapid symptom relief. Although the identification of characteristic MSU and CPP crystals in SF or tophi remains the standard for the diagnosis of gout and CPPD (1,2), DECT and multi­ energy photon­ counting CT have the potential to become clinically relevant by enabling noninvasive detection, characterization, quantification, and mapping of MSU and CPP crystal deposition (3).

Rami Hajri, MD Steven D. Hajdu, MD Thomas Hügle, MD, PhD Pascal Zufferey, MDLausanne University Hospital and University of LausanneLaurent Guiral, MScGE Healthcare Buc, FranceFabio Becce, MDLausanne University Hospital and University of LausanneLausanne, Switzerland

1. Richette P, Doherty M, Pascual E, Barskova V, Becce F, Castaneda J, et al. 2018 updated European League Against Rheumatism evidence­based recommendations for the diagnosis of gout. Ann Rheum Dis 2019. E­pub ahead of print.

2. Zhang W, Doherty M, Bardin T, Barskova V, Guerne PA, Jansen TL, et al. European League Against Rheumatism recommendations for calcium pyrophosphate deposition. Part I: terminology and diagnosis. Ann Rheum Dis 2011;70:563–70.

3. Stamp LK, Anderson NG, Becce F, Rajeswari M, Polson M, Guyen O, et al. Clinical utility of multi­ energy spectral photon­ counting computed tomography in crystal arthritis. Arthritis Rheumatol 2019;71:1158–62.