Pd

ATLa

b

c

d

ARRA

KLDPMCMT

1

asceo

RT

0h

Lung Cancer 81 (2013) 397– 403

Contents lists available at SciVerse ScienceDirect

Lung Cancer

j ourna l ho me page: www.elsev ier .com/ locate / lungcan

otentialities of aberrantly methylated circulating DNA foriagnostics and post-treatment follow-up of lung cancer patients

nastasia A. Ponomaryovaa,∗, Elena Yu. Rykovab, Nadezda V. Cherdyntsevaa,c,atiana E. Skvortsovab, Alexey Yu. Dobrodeeva, Alexander A. Zav’yalova,eonid O. Bryzgalovd, Sergey A. Tuzikova,c, Valentin V. Vlassovb, Pavel P. Laktionovb

Cancer Research Institute of Siberian Branch of the Russian Academy of Medical Sciences, Tomsk, RussiaInstitute of Chemical Biology and Fundamental Medicine of Siberian Branch of the Russian Academy of Sciences, Novosibirsk, RussiaSiberian State Medical University, Tomsk, RussiaInstitute of Cytology and Genetics of Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia

a r t i c l e i n f o

rticle history:eceived 26 March 2013eceived in revised form 22 May 2013ccepted 24 May 2013

eywords:ung canceriagnosticsrognosisonitoring

irculating DNAethylation

umor suppressor gene

a b s t r a c t

To date, aberrant DNA methylation has been shown to be one of the most common and early causes ofmalignant cell transformation and tumors of different localizations, including lung cancer. Cancer cell-specific methylated DNA has been found in the blood of cancer patients, indicating that cell-free DNAcirculating in the blood (cirDNA) is a convenient tumor-associated DNA marker that can be used as aminimally invasive diagnostic test. In the current study, we investigated the methylation status in bloodsamples of 32 healthy donors and 60 lung cancer patients before and after treatment with neoadjuvantchemotherapy followed by total tumor resection. Using quantitative methylation-specific PCR, we foundthat the index of methylation (IM), calculated as IM = 100 × [copy number of methylated/(copy numberof methylated + unmethylated gene)], for the RASSF1A and RARB2 genes in the cirDNA isolated from bloodplasma and cell-surface-bound cirDNA was elevated 2- to 3-fold in lung cancer patients compared withhealthy donors. Random forest classification tree model based on these variables combined (RARB2 andRASSF1A IM in both plasma and cell-surface-bound cirDNA) lead to NSCLC patients’ and healthy subjects’differentiation with 87% sensitivity and 75% specificity. An association of increased IM values with anadvanced stage of non-small-cell lung cancer was found for RARB2 but not for RASSF1A. Chemotherapyand total tumor resection resulted in a significant decrease in the IM for RARB2 and RASSF1A, in both

cirDNA fractions, comparable to the IM level of healthy subjects. Importantly, a rise in the IM for RARB2was detected in patients within the follow-up period, which manifested in disease relapse at 9 months,confirmed with instrumental and pathologic methods. Our data indicate that quantitative analysis ofthe methylation status of the RARB2 and RASSF1A tumor suppressor genes in both cirDNA fractions isa useful tool for lung cancer diagnostics, evaluation of cancer treatment efficiency and post-treatmentmonitoring.

Crow

. Introduction

The majority of lung cancer patients are diagnosed at thedvanced disease stage [1,2]. Combined treatment increases theuccess rate of lung cancer therapy. For the increase of survival

hance it is necessary to develop early tumor detection andfficient tumor recurrence monitoring. Molecular genetic markersf cancer cells which can be detected in the blood have been used

∗ Corresponding author at: Cancer Research Institute of Siberian Branch of theussian Academy of Medical Sciences, 5, Kooperativny Street, Tomsk 634050, Russia.el.: +7 3822 51 25 29; fax: +7 3822 51 40 97.

E-mail address: anastasia-ponomaryova@rambler.ru (A.A. Ponomaryova).

169-5002/$ – see front matter. Crown Copyright © 2013 Published by Elsevier Ireland Ltttp://dx.doi.org/10.1016/j.lungcan.2013.05.016

n Copyright © 2013 Published by Elsevier Ireland Ltd. All rights reserved.

for diagnostics, theranostics and tumor process monitoring [3].Numerous studies have investigated DNA circulating in bloodplasma (cirDNA) as a phenomenon related to tumor pathogenesisand as tool for cancer diagnostics and prognosis [4–9]. Epigeneticand genetic changes identical to those observed in tumor tissueshave been reported to be detectable in the plasma cirDNA of cancerpatients [10–13]. Tumor-derived cirDNA appeared to be not onlydetected in blood plasma but also bound to the surface of bloodcells (csb-cirDNA), in both erythrocytes and leukocytes [14–18].It should be noted that cancer-related DNA usually represents

a minor part of cirDNA and is found at low concentrations, thusrequiring high PCR specificity [19].

Aberrantly methylated DNA was shown to have analytic bene-fits, related to chemical conversion, as compared to point mutation,

d. All rights reserved.

3 ung C

crgcaatll[sdc

2

2

(ilctApodtwbaspaI

2e

pawEl0t(wwpwi

2c

RTl(pbf

98 A.A. Ponomaryova et al. / L

opy number variation or DNA rearrangements [20,21]. Withespect to lung cancer, aberrantly methylated tumor suppressorenes, such as RARB2, RASSF1A, CDKN2A, FHIT and APC, could beonsidered candidates for diagnostic and prognostic usage. RARB2nd RASSF1A were selected from the list for the present study as fars hypermethylation frequency for these two genes was reportedo be significantly higher in lung tumors as compared with normalung tissues [22,23]. In addition, RARB2 and RASSF1A hypermethy-ation was recently associated with poor prognosis in lung cancer24,25]. The purpose of this study was to estimate the diagnosticignificance of alterations in RARB2 and RASSF1A methylation levelsetected in plasma cirDNA and csb-cirDNA from patients with lungancer before and after combined treatment.

. Materials and methods

.1. Patients and healthy controls

Untreated patients (n = 60) with non-small cell lung cancerNSCLC) under the care of the Tomsk Cancer Research Institute werencluded in this study. Diagnoses of patients were confirmed histo-ogically according to the TNM classification (T1–3N0–3M0). Patientlinical data, including sex, age, smoking status, disease stage andumor histology, were obtained from the Hospital Cancer Registry.ll patients received 2–3 courses of neoadjuvant chemotherapy:aclitaxel at a dose of 175 mg/m2 body surface area over a periodf 1 day and carboplatin AUC 6 over a period of 1 day, repeated onay 21. Surgical resection of the tumor and intraoperative radia-ion therapy (IORT) were scheduled after the chemotherapy periodas completed. Blood samples from NSCLC patients were taken

efore treatment, 15–30 days after the last injection of drugs and in follow-up period ranging from 10 to 15 days up to 9 months afterurgery. The control group consisted of 32 healthy volunteers, com-arable in age to the lung cancer patients. The current research waspproved by the Ethical Committee of the Tomsk Cancer Researchnstitute.

.2. Preparation of plasma and cell-surface-bound fractions, DNAxtraction and bisulfite treatment

Venous blood samples were stabilized and fractionated intolasma and blood cells. The csb-cirDNA fractions were obtaineds previously described [15]. Briefly, blood cells were incubatedith 9 volumes of phosphate buffered saline (PBS) containing 5 mM

DTA (PBS/EDTA) (5 min, room temperature). The cells were pel-eted by centrifugation and incubated with an equal volume of.25% trypsin solution, and the enzyme was then inactivated using arypsin inhibitor from soybeans (Sigma–Aldrich, St. Louis, MO, USA)4 min, room temperature). Plasma, PBS/EDTA and trypsin eluatesere centrifuged for an additional 20 min at 800 × g, and aliquotsere stored frozen at −40 ◦C. cirDNA was isolated from 1 mL oflasma, 3 mL of PBS/EDTA and 1 mL of trypsin eluates and modifiedith sodium bisulfite, followed by purification using a blood DNA

solation kit (BioSilica Ltd., Novosibirsk, Russia).

.3. Measurement of tumor suppressor gene methylation levels inirDNA

Concentrations of methylated and unmethylated forms of theASSF1A gene were assessed by quantitative methylation-specificaqMan PCR (MSP). Concentrations of methylated and unmethy-ated forms of the RARB2 gene were assessed by EvaGreen PCR

Biotium, Hayward, California, USA). The primer, master mix com-osition and PCR conditions for both RASSF1A and RARB2 haveeen described in previous studies [15,26]. The methylation dif-erence between duplicate measurements for RASSF1A and RARB2

ancer 81 (2013) 397– 403

was 1.7% and 2.1%, respectively; the sensitivity of 30 and 40 genecopies per reaction, respectively; and the MSP efficiency was inthe range of 93–95%. Standard curves were generated using serialdilutions of the purified methylated and unmethylated MSP ampli-fication products, stored frozen as stock solutions (1012 copies/mL)and freshly diluted before each use. The index of gene methylation(IM) was calculated as % IM = 100 × [copy numbers of methylatedgene/(copy numbers of methylated gene + unmethylated gene)].

2.4. Statistical analysis

The Mann–Whitney U-test and ANOVA analyses were usedto assess the statistical relationships between gene methyla-tion levels and the clinicopathologic variables. The R packageade4 using instrumental principal component analysis (PCA) wasapplied for the global analysis of the differences between the lungcancer patients and healthy controls parameters (cirDNA methy-lation level, subject age, sex, patient stage, histotype, smokingstatus). PCA allowed visualize representation of point classes fac-torial maps. Monte Carlo test on the between-groups inertia wasperformed 1000 permutations to calculate an observation and p-value. MANOVA was additionally performed to test for differencesbetween control and patient groups.

Random forest classification algorithm was used to estimate theperformance of a predictive model. The random forest algorithmwas originally developed by Breiman [27] and has been imple-mented as the Random Forest package in R [28]. Random Forest isan Ensemble Classifier in which the base classifier is an un-prunedDecision Tree built from a random selection of feature variables (inthis case methylation indexes) for a randomly selected subset oftraining samples (patients). The method allows assess the affect ofa feature variable upon the classification, identified as importancescore. Using the randomForest package (v.4.6-2) in the R program-ming language a random forest of 10.000 trees was generated forclassification.

3. Results

3.1. Methylation of the RASSF1A and RARB2 genes in cirDNA andcsb-cirDNA in NSCLC patients

Our earlier studies showed that the sensitivity of the methy-lated gene detection is increased when cirDNA and csb-cirDNAfrom the blood plasma of cancer patients are analyzed simultane-ously [6,7,29,30]. In this study, we also analyzed both plasma andcsb-cirDNA to increase the sensitivity of the lung cancer diagnos-tics. The concentration of methylated and unmethylated RASSF1Aand RARB2 tumor suppressor genes circulating in the blood wasquantified by quantitative MSP in duplicates for all the samples.The Mann–Whitney U-test showed the significant increase of theindex of methylation values for RASSF1A, RARB2 genes in the plasmacirDNA and csb-cirDNA from the primary NSCLC patients (p < 0.01)(Table 1).

Mann–Whitney U-test revealed no association of the IM val-ues for RARB2, RASSF1A genes with patient sex, age, smokingstatus, tumor histology. The advanced stage of disease appearedto be moderately associated with hypermethylation of RARB2 incsb-cirDNA: the average IM value for RARB2 in csb-cirDNA wasincreased in stage III patients compared with stage I-II patients(p = 0.04) (Table 1). We used multivariate analysis to further eval-uate the clinical and demographic correlates with the 4 variables

of two genes methylation level in the circulating DNA. Multivariateanalysis of variance (MANOVA) revealed that the two gene methy-lation indexes in two cirDNA fractions have the high associationwith lung cancer diagnosis, whereas other variables do not provide

A.A. Ponomaryova et al. / Lung Cancer 81 (2013) 397– 403 399

Table 1RARB2 and RASSF1A index of methylation in the circulating DNA from NSCLC patients and healthy subjects.

Index of methylation, mean and standard error (M ± SE)

RARB2 RASSF1A

N Csb-cirDNA Cell-free cirDNA Csb-cirDNA Cell-free cirDNA

Healthy subjects 33 11 ± 2 17 ± 2 19 ± 3 21 ± 3

NSCLC patientsTotal 60 35 ± 4 50 ± 4 38 ± 4 42 ± 5

p = 0.002 p = 0.0000 p = 0.0007 p = 0.004Age (years)

≤60 35 (35/60 – 58%) 29 ± 4 55 ± 6 36 ± 4 41 ± 5>60 25 (25/60 – 42%) 41 ± 6 43 ± 5 40 ± 5 43 ± 5

p = 0.06 p = 0.08 p = 0.62 p = 0.81Gender

Female 8 (8/60 – 13%) 26 ± 9 35 ± 9 47 ± 10 53 ± 10Male 52 (52/60 – 87%) 35 ± 4 52 ± 4 36 ± 4 40 ± 4

p = 0.36 p = 0.11 p = 0.28 p = 0.29Smoking

Smokers 50 (50/60 – 83%) 34 ± 4 51 ± 4 35 ± 3 40 ± 4Non-smokers 10 (10/60 – 17%) 34 ± 9 43 ± 10 52 ± 10 52 ± 10

p = 0.99 p = 0.38 p = 0.08 p = 0.28Stage

I, II 20 (20/60 – 33%) 27 ± 4 50 ± 5 39 ± 5 41 ± 5III 40 (40/60 – 67%) 42 ± 5 51 ± 6 37 ± 4 42 ± 5

p = 0.04 p = 0.96 p = 0.82 p = 0.95Histology

SCC 40 (40/60 – 67%) 33 ± 4 47 ± 5 38 ± 5 42 ± 5AC 20 (20/60 – 33%) 35 ± 5 54 ± 6 37 ± 4 43 ± 5

p = 0.79 p = 0.39 p = 0.85 p = 0.95

Abbreviations: NSCLC, non-small-cell lung cancer; SCC, squamous cell carcinoma; AC, adenocarcinoma.

Table 2Between-subjects effects from MANOVA analysis and principal component analysis (PCA): Monte Carlo test.

Factors PCA (test Mote Carlo) MANOVA

Observation p-Value Pr(>F) Approx F

Histology 0.003 0.964 0.952 0.17Age 0.04 0.025 0.006** 3.9Group (healthy subject or lung cancer patient) 0.15 >0.001 >0.0000001*** 14.9Stage 0.06 0.025 0.176 1.4Smoking 0.08 0.093 0.1 1.5

*

sinictaaaoIt

3p

adiRhfia

** <0.01 significance.**<0.001 significance.

ignificant influence (Table 2, MANOVA). According to this test thempact of the methylation status variables in the lung cancer diag-osis decreases in the raw: RARB2 IM in cell-free cirDNA > RARB2 IM

n csb-cirDNA > RASSF1A IM in csb-cirDNA > RASSF1A IM in cell-freeirDNA (Supplementary Data, Table S1). Using PCA (Monte Carloest) we also demonstrated for two genes methylation the highestssociation with disease (p < 0.001), no significant association with

tumor histotype, smoking history, and low association with stagend age of a patient (Table 2, PCA). Fig. 1 visualize representationf point classes factorial maps, contribution of RARB2 and RASSF1AM variables in the components selected by PCA are represented inhe Supplementary Data (Table S2).

.2. Methylation index of tumor suppressor genes in cirDNA as aotential diagnostic marker for lung cancer

We tested the possibility to enhance the lung cancer predictiveccuracy using the combination of the two methylation markersetection in plasma and csb-cirDNA fractions compared with their

ndividual analysis. Using the machine learning techniques called

andom Forests we could discriminate lung cancer patients fromealthy controls with 81% accuracy (85% sensitivity and 75% speci-city) with totally 4 variables enrolled (Table 3). To note, if a personge value is added as the additional variable in classification tree

model it leads to the increase of lung cancer prediction accuracyup to 82% (87% sensitivity and 75% specificity). This multivari-ate assay provides much more effective discrimination of cancerpatients from healthy controls compared with the assay based onthe individual methylation indexes of RARB2 and RASSF1A in plasmacirDNA and csb-cirDNA, alone or in pair combinations (Table 3).The highest accuracy of 89% (91% sensitivity and 87% specificity)was found for the III stage patients’ discrimination from healthycontrols when 4 methylation variables were assessed independentof the age variable enrolment in the model (Table 3 and Fig. 1).The importance of quantitative components decreased in the raw:RARB2 IM in cell-free cirDNA > RARB2 IM in csb-cirDNA > RASSF1AIM in cell-free cirDNA > RASSF1A IM in csb-cirDNA (SupplementaryData, Table S3).

The positive and negative likelihood ratios (LR+ and LR−) werecalculated separately for the IM of RARB2 and RASSF1A in both frac-tions of cirDNA as well as in their combination (Table 3). IndividualLR+ values are in the range from 1.5 to 3.2, LR− are in the rangefrom 0.3 to 0.6 and thus yield small to moderate changes in thepost-test probability of disease. LR+ for the analysis of two genes

in two cirDNA fractions combination has the value of 3.5 whichslightly increase the change in the post-test probability in case ofthe positive result [31]. Otherwise LR− for all fraction combinationhas the value 0.17 which is close to the range of high level changes

400 A.A. Ponomaryova et al. / Lung Cancer 81 (2013) 397– 403

Fig. 1. Challenge data inspection by principal component analysis (PCA). Scatter plots with representation of the various classes were produced with the s.class commandof the ade4 R package. The various classes are: (A) age (young – ≤60 age, old – >60 age); (B) group (healthy subjects, cancer patients); (C) stage (0 – control, 1 – I–II stage, 3 –III stage); (D) histotype (SCC, squamous cell carcinoma; AC, adenocarcinoma).

Table 3Evaluation of a diagnostic test value.

Healthy subjects vs lung cancer patients

Index of methylation Sensitivity (%) Specificity (%) LR+ LR− PPV (%) NPV (%) OOB estimate oferror rate (%)

RARB2b Cell-free DNA 72 62 1.89 0.45 74 61 31.65RARB2b Cell-free DNA + csb-cirDNA 81 75 3.24 0.25 83 73 21.52RASSF1Ab Cell-free DNA 66 57 1.53 0.60 69 53 37.97RASSF1Ab Cell-free DNA + csb-cirDNA 79 56 1.80 0.38 73 64 30.38RARB2 + RASSF1Aa Cell-free DNA + csb-cirDNA 85 75 3.40 0.20 83 77 18.99RARB2 + RASSF1Ab Cell-free DNA + csb-cirDNA 87 75 3.48 0.17 84 80 17.72RARB2 + RASSF1Ac Cell-free DNA + csb-cirDNA 91 87 7.00 0.10 89 90 10.61RARB2 + RASSF1Ad Cell-free DNA + csb-cirDNA 91 87 7.00 0.10 89 90 10.61

Abbreviations: LR+, positive likelihood ratio; LR− , negative likelihood ratio; PPV, positive predictive value; NPV, negative predictive value.a All patients vs control, without age.

itlwpfiivfRt

3p

tpo

b All patients vs control, with age.c III stage patients vs control, without age.d III stage patients vs control, with age.

n the post-test probabilities in case of the negative result (lesshan 0.1). The highest LR+/LR− values are obtained for the III stageung cancer patients’ discrimination from normal controls 7/0.1,

hich indicate that the test make moderate/high changes of theost-test probabilities compared with the pre-test ones. The bene-t of simultaneously analyzing two genes in two cirDNA fractions

s additionally confirmed by the positive and negative predictionalue estimations (PPV and NPV). Indeed, PPV/NPV values increaserom 74/61 when analyzing the most powerful variable – IM forARB2 in plasma cirDNA alone – up to 84/80 when analyzing IM forwo genes in two fractions of cirDNA (Table 3).

.3. Suppressor gene methylation in the cirDNA ofost-chemotherapy and post-surgery lung cancer patients

To determine whether combined treatment affects the concen-ration of methylated alleles, we evaluated the IM in lung canceratients after neoadjuvant chemotherapy and surgery with intra-perative radiation therapy (IORT). Blood samples from 43 patients

with NSCLC were obtained before treatment, after neoadjuvantchemotherapy (15–30 days before surgery) and 10–15 days aftersurgery. After chemotherapy a trend of decreasing of the IM valuefor both genes was detected in the cirDNA from treated patientscompared with the IM value before treatment (Fig. 2). Moreover,in the post-tumor resection period, the majority of the patientsdemonstrated a more pronounced decrease in the IM values ofRASSF1A and RARB2 in both cirDNA and csb-cirDNA, which becamecomparable with IM values for healthy subjects. It should be notedthat the decrease in the IM value was statistically significant in bothfractions being more pronounced in csb-cirDNA than in the cirDNAfraction (*p < 0.01, Fig. 2).

3.4. RARB2 and RASSF1A methylation during the post-treatmentfollow-up period

Reductions in the IM values of the RARB2 and RASSF1A genesin the cirDNA of lung cancer patients after therapy demonstratethat the concentration of methylated markers in the blood was

A.A. Ponomaryova et al. / Lung C

Fb

dTstbm2ltepittvms

Fal

ig. 2. RARB2 and RASSF1A genes index of methylation in cirDNA of NSCLC patientsefore treatment, after neoadjuvant chemotherapy (post-NAC) and surgery.

ramatically decreased following partial or total tumor resection.o test the potential applicability of the methylated markers undertudy for tumor recurrence monitoring, we compared quantita-ive changes in RARB2 and RASSF1A gene methylation in cirDNAetween patients who experienced disease recurrence within 9onths after surgery and relapse-free patients. Blood cirDNA from

6 patients with NSCLC was tested for the presence of methy-ated genes 10–15 days and 3, 6 and 9 months after the combinedreatment. From this group, 5 patients showed evidence of dis-ase progression, which was clinically observed 9 months afteratient treatment. An increase in methylated allele concentration

n at least one of the studied genes, up to the level detected beforereatment, was observed in 100% of these patients (Fig. 3A). In con-rast, all patients without recurrence did not show increased IM

alues in either fraction of cirDNA (Fig. 3B). Similar changes ofethylation level were found also for RASSF1A gene (results are not

hown).

ig. 3. Long term study of RARB2 index of methylation in cirDNA of NSCLC patientsfter combined treatment. (A) N patient with clinically confirmed progression ofung cancer; (B) F patient, without clinically confirmed tumor recurrence.

ancer 81 (2013) 397– 403 401

4. Discussion

One of the mechanisms contributing to malignant transfor-mation is tumor suppressor gene inactivation due to epigeneticchanges, such as aberrant cytosine methylation in the CpG-dinucleotides. Studies have reported hypermethylation of varioustumor suppressor genes in lung tumors, including the RARB2 andRASSF1A genes, which are shown to be methylated at a high fre-quency [19,24,32–35]. RARB2, the gene encoding the vitamin A(retinoid acid) nuclear receptor, plays a critical role during embry-onic development, homeostasis, cell growth and differentiation. Ithas been shown to be methylated in up to 60% of NSCLC cells com-pared with 9% in healthy lung tissue [33,35]. The methylation of theRASSF1A gene, which is involved in the regulation of cell prolifer-ation and apoptosis, was observed in up to 50% of primary NSCLCtumors compared to 10% in healthy subjects [33,35,36]. However,the increase in RARB2 and RASSF1A methylation levels was loweror not detected in plasma cirDNA samples compared with tumortissues [19,36].

In this study using larger patient groups we confirmed our pre-vious findings [6] that RARB2 methylation level is increased inlung cancer patient blood compared with healthy subjects and thatRARB2 methylation level is associated with disease stage. However,according to earlier reports, a single gene methylation analysisin cirDNA from blood plasma does not provide sufficient predic-tion accuracy for lung cancer detection [11,32,35,37]. Therefore,to improve the sensitivity and specificity of lung cancer diag-nostics, we additionally tested the methylation status of anotherpotential marker, the RASSF1A tumor suppressor gene. Moreover,we analyzed two genes methylation changes in both plasma andcell-surface-bound cirDNA fractions. Indeed, the accuracy of dif-ferentiating lung cancer patients from healthy subjects was higherusing the two-marker combination than using each marker alone.Random forest classification tree model estimated high importancescore for plasma RARB2 IM, moderate for csb RARB2 IM, plasmaRASSF1A IM, csb RASSF1A IM variables and low for a patient agevariable. Machine learning model based on these variables com-bined lead to NSCLC patients’ and healthy subjects’ differentiationwith 87% sensitivity and 75% specificity.

Sensitivity and specificity have limited use in day-to-day clin-ical practice, and few clinicians will know this information wheninterpreting a test result. A more useful approach is to combine thesensitivity and specificity results into single measures that tell ushow much more likely a positive or negative test result is to havecome from someone with the disease than from someone with-out it. These are known as the likelihood ratios for a positive test(LR+) and for a negative test (LR−) [38]. According to Grimes andSchulz opinion one of the most important reasons to use LRs is thatlikelihood ratios refine clinical judgment [31]. Application of a like-lihood ratio to a working diagnosis generally changes the diagnosticprobability. Likelihood ratios from 2 to 5 yield small increases in thepost-test probability of disease, from 5 to 10 moderate increases,and above 10 large increases. In this study, we have shown that theLR+ = 3.5 which according to a rough estimate [31] increases thepost-test probability of lung cancer in the presence of a positivetest, by approximate 25%, and the LR− = 0.2 which decreases in thepost-test probability of lung cancer in the presence of a negativetest, by approximate 30%.

The knowledge of disease prevalence will provide an estimateof the baseline probability of cancer, and the result of the rapidtest will modify this probability on the basis of the known testcharacteristics. Instead of taking a positive or negative test result

for granted, the consideration of predictive values as conditionalprobabilities could prevent inappropriate interpretations of testresults. In practice, the PPVs and NPVs provide the most useful clin-ical information. This is necessary integration of the laboratory test

4 ung C

cpcft

mccdippiphs(stble

ctipnistfDbc

datmocdsdiPiagpIr

5

ciesumo

[

[

[

[

[

[

[

[

[

02 A.A. Ponomaryova et al. / L

haracteristics (sensitivity and specificity) with knowledge of therevalence of the disease [39]. Our findings allow us to draw theonclusion that the methylation assay for both genes in both cirDNAractions is characterized by low false-positive and false-negativeest rates (i.e., a high PPV = 0.84 and NPV = 0.80).

According to data by Fujiwara et al. [25], who determinedethylation status by conventional methyl-specific PCR, the asso-

iation exists between RARB2 gene methylation levels in serumirDNA and lung cancer clinical stage. Our quantitative PCRata demonstrated that RARB2 and RASSF1A methylation levels

ncreased significantly in cell-free cirDNA of stage I–II patients com-ared to healthy subjects and did not further increase in stage IIIatients (p > 0.05). However, the mean RARB2 methylation level

n csb-cirDNA was found to be higher in stage III patients com-ared with stage I–II patients (p < 0.05). RF model demonstratedigh accuracy of stage III patients’ discrimination from healthyubjects using circulating RARB2, RASSF1A gene methylation levels89%). Elevation of the IM is thought to indicate that overrepre-entation of tumor DNA in the blood sample caused an increase inhe tumor-derived DNA concentration in cell-free and cell-surface-ound circulating DNA. Cell surface-bound DNA was shown to be

ess degraded compared with cell-free cirDNA and to be thus a morefficient PCR template.

The reasons why tumor DNA is bound to the surface of bloodells are not clear, but this relationship may be related to the struc-ure of nucleoprotein complexes, as well as cancer-induced changesn the composition and amount of the blood cell-surface and bloodlasma proteins (such as amyloid P increasing the solubility ofucleosomes). It has been recently shown, that DNA degradation

s dependent on its nucleotide structure [40] and during apoptosisome sequences are more abundant in the pool of cell-free cirDNAhan others [41]. In addition, the acquisition of cirDNA by the sur-ace of the blood cells may be considered that leukocytes “secrete”NA on to the cell surface apparently forming a network. It coulde that this DNA complex could also act as a binding site for theirDNA.

The IM values of RARB2 and RASSF1A in cirDNA were shown toecrease in lung cancer patients after chemotherapy, resulting in

partial reduction of tumor burden, as confirmed by instrumen-al methods. After total tumor resection, the IM values decreased

ore markedly, reaching healthy subject levels. According to databtained by Diehl et al. [3], the half-life of mutated tumor-derivedirculating DNA in the blood plasma is no more than 2 h. Theirata are consistent with other published results [11,42], whichhowed a rapid decrease of mutated DNA concentration in blooderived from tumor patients after surgery, as well as fetal DNA

n the mother’s blood circulation after newborn delivery [43].ost-treatment follow-up revealed that all NSCLC patients show-ng increased IM values for at least one of the two genes (RARB2nd/or RASSF1A) 9 months after surgery displayed cancer pro-ression (relapses), which was confirmed with instrumental andathologic methods. Our data demonstrate that RARB2 and RASSF1A

M determination represents a valuable tool for monitoring tumorecurrence.

. Conclusion

Methylation index determination in csb-cirDNA and plasmairDNA in blood samples of NSCLC patients appeared to be morenformative than analysis of plasma cirDNA alone. Our data providevidence that quantitative analysis of RARB2 and RASSF1A tumor

uppressor gene methylation status in both cirDNA fractions is aseful tool for lung cancer diagnostics, evaluation of cancer treat-ent efficiency and post-treatment monitoring. The clinical utility

f the suggested marker set from this study remains to be validated[

ancer 81 (2013) 397– 403

in large cohorts in the near future and may be further improved byaddition of other biomarkers.

Conflict of interest statement

None declared.

Acknowledgements

The research has been carried out with support of the grantsfrom Russian Foundation of Basic Research #11-04-12105-offi-m-2011, Siberian Branch of Russian Academy of Science Programin collaboration with other scientific organizations #65, RussianAcademy of Science Program “Fundamental Science for Medicine”#23, Federal Special-Purpose Program “Scientific, Academic, andTeaching Staff of Innovative Russia” 2009–2013 (# P256).

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at http://dx.doi.org/10.1016/j.lungcan.2013.05.016.

References

[1] Buccheri G, Ferrigno D. Lung cancer: clinical presentation and specialist referraltime. Eur Respir J 2004;24:898–904.

[2] Nisman B, Biran H, Ramu N, Heching N, Barak V, Peretz T, et al. Prognostic valueof ProGRP in lung cancer. Anticancer Res 2009;29:4827–32.

[3] Diehl F, Schmidt K, Choti MA, Romans K, Goodman S, Li M, et al. Circulatingmutant DNA to assess tumor dynamics. Nat Med 2008;14:985–90.

[4] Kumar S, Guleria R, Singh V, Bharti AC, Mohan A, Das BC. Plasma DNA levelin predicting therapeutic efficacy in advanced non small cell lung cancer. EurRespir J 2010;36:885–92.

[5] Van der Drift MA, Hol BE, Klaassen CH, Prinsen CF, van Aarssen YA, Donders R,et al. Circulating DNA is a non-invasive prognostic factor for survival in non-small cell lung cancer. Lung Cancer 2010;68:283–7.

[6] Ponomaryova AA, Rykova EY, Cherdyntseva NV, Skvortsova TE, Dobrodeev AY,Zav’yalov AA, et al. RARbeta2 gene methylation level in the circulating DNAfrom blood of patients with lung cancer. Eur J Cancer Prev 2011;20:453–5.

[7] Ponomaryova AA, Rykova EY, Cherdyntseva NV, Skvortsova TE, Dobrodeev AY,Zav’yalov AA, et al. Assay of methylated gene RARbeta2 in circulating DNA ofblood from patients with lung cancer as a potential prognostic marker. VoprOnkol (Russ) 2011;57:302–7.

[8] Cooke S, Circulating Campbell P. DNA and next-generation sequencing. RecentResults Cancer Res 2012;19:143–9.

[9] Gahan PB. Biology of circulating nucleic acids and possible roles in diagnosis andtreatment in diabetes and cancer. Infect Disord Drug Targets 2012;12:360–70.

10] Jahr S, Hentze H, Englisch S, Hardt D, Fackelmayer FO, Hesch RD, et al. DNAfragments in the blood plasma of cancer patients: quantitations and evidencefor their origin from apoptotic and necrotic cells. Cancer Res 2001;61:1659–65.

11] Fleischhacker M, Schmidt B. Circulating nucleic acids (CNAs) and cancer – asurvey. Biochim Biophys Acta 2007;1775:181–232.

12] Radpour R, Barekati Z, Kohler C, Lv Q, Bürki N, Diesch C, et al. Hypermethy-lation of tumor suppressor genes involved in critical regulatory pathways fordeveloping a blood-based test in breast cancer. PLoS ONE 2011;6:e16080.

13] Carethers JM. Proteomics, genomics, and molecular biology in the personalizedtreatment of colorectal cancer. J Gastrointest Surg 2012;16:1648–50.

14] Tamkovich SN, Bryzgunova OE, Rykova EY, Bryzgunova OE, Starikov AV,Kuznetsova NP, et al. Circulating nucleic acids in blood of healthy male andfemale donors. Clin Chem 2005;51:1317–9.

15] Skvortsova TE, Rykova EY, Tamkovich SN, Bryzgunova OE, Starikov AV,Kuznetsova NP, et al. Cell-free and cell-bound circulating DNA in breast tumors:DNA quantification and analysis of tumor-related gene methylation. Br J Cancer2006;94:1492–5.

16] Vlassov VV, Laktionov PP, Rykova EY. Circulating nucleic acids as a potentialsource for cancer biomarkers. Curr Mol Med 2010;10:142–65.

17] Ponomaryova AA, Rykova EY, Cherdyntseva NV, Skvortsov TE, Cherepanova AV,Morozkin ES, et al. Concentration and distribution of single-copy beta-actingene and LINE-1 repetitive elements in blood of lung cancer patients. In: GahanPB, editor. Circulating nucleic acids in plasma and serum. United Kingdom:Springer; 2011. p. 41–5.

18] Rykova EY, Morozkin ES, Ponomaryova AA, Loseva EM, Zaporozhchenko IA,

Cherdyntseva NV, et al. Cell-free and cell-bound circulating nucleic acid com-plexes: mechanisms of generation, concentration and content. Expert Opin BiolTher 2012;12:S141–53.

19] Kontic M, Stojsic J, Jovanovic D, Bunjevacki V, Ognjanovic S, Kuriger J, et al.Aberrant promoter methylation of CDH13 and MGMT genes is associated with

ung C

[[

[

[

[

[

[

[[

[

[

[

[

[

[

[

[

[

[

[

[

[

A.A. Ponomaryova et al. / L

clinicopathologic characteristics of primary non-small-cell lung carcinoma.Clin Lung Cancer 2012;13:297–303.

20] Esteller M. Epigenetics in cancer. N Engl J Med 2008;358:1148–59.21] Duffy MJ, Napieralski R, Martens JW, Span PN, Spyratos F, Sweep FC, et al.

Methylated genes as new cancer biomarkers. Eur J Cancer 2009;45:335–46.22] Li W, Deng J, Jiang P, Zeng X, Hu S, Tang J. Methylation of the RASSF1A and RARˇ

genes as a candidate biomarker for lung cancer. Exp Ther Med 2012;3:1067–71.23] Lee JU, Sul HJ, Son JW. Promoter methylation of CDKN2A, RARˇ, and RASSF1A in

non-small cell lung carcinoma: quantitative evaluation using pyrosequencing.Tuberc Respir Dis (Seoul) 2012;73:11–21.

24] Kim JS, Kim JW, Han J, Shim YM, Park J, Kim DH. Cohypermethylation of CDKN2Aand FHIT promoters as a prognostic factor of recurrence in surgically resectedstage I non-small cell lung cancer. Cancer Res 2006;66:4049–54.

25] Ko E, Lee BB, Kim Y, Lee EJ, Cho EY, Han J, et al. Association of RASSF1A and p63with poor recurrence-free survival in node-negative stage I–II non-small celllung cancer. Clin Cancer Res 2013;19:1204–12.

26] Fackler MJ, Veigh M, Mehrotra J, Blum MA, Lange J, Lapides A, et al. Quantitativemultiplex methylation-specific PCR assay for the detection of promoter hyper-methylation in multiple genes in breast cancer. Cancer Res 2004;64:4442–52.

27] Breiman L. Random Forests. Mach Learn 2001;45(1):5–32.28] Liaw A, Wiener M. Classification and regression by randomForest 2002. R News

2002;2(3):18–22.29] Skvortsova TE, Bryzgunova OE, Lebedeva AO, Mak VV, Vlassov VV, Laktionov

PP. Methylated cell-free DNA in vitro and in vivo. In: Gahan PB, editor. Cir-culating nucleic acids in plasma and serum. United Kingdom: Springer; 2011.p. 185–94.

30] Elistratova EV, Shelestyuk PI, Permyakova VI, Chikova ED, Tuzikov SA, Vlassov

VV, et al. Blood-based methylated DNA and tumor-specific protein analysisin gastric cancer diagnostics. In: Gahan PB, editor. Circulating nucleic acids inplasma and serum. United Kingdom: Springer; 2011. p. 57–62.

31] Grimes DA, Schulz KF. Refining clinical diagnosis with likelihood ratios. Lancet2005;365:1500–5.

[

[

ancer 81 (2013) 397– 403 403

32] Fujiwara K, Fujimoto N, Tabata M, Nishii K, Matsuo K, Hotta K, et al. Identifica-tion of epigenetic aberrant promoter methylation in serum DNA is useful forearly detection of lung cancer. Clin Cancer Res 2005;11:1219–25.

33] Hsu HS, Chen TP, Hung CH, Wen CK, Lin RK, Lee HC, et al. Characterization of amultiple epigenetic marker panel for lung cancer detection and risk assessmentin plasma. Cancer 2007;110:2019–26.

34] De Jong WK, Verpooten GF, Kramer H, Louwagie J, Groen HJ. Promoter methyla-tion primarily occurs in tumor cells of patients with non-small cell lung cancer.Anticancer Res 2009;29:363–70.

35] Zhang Y, Wang R, Song H. Methylation of multiple genes as a candidatebiomarker in non-small cell lung cancer. Cancer Lett 2011;303:21–8.

36] Schmiemann V, Bocking A, Kazimirek M, Onofre AS, Gabbert HE, Kappes R,et al. Methylation assay for the diagnosis of lung cancer on bronchial aspirates:a cohort study. Clin Cancer Res 2005;11:7728–34.

37] Chen H, Suzuki M, Nakamura Y, Ohira M, Ando S, Iida T, et al. Aberrant methy-lation of RASGRF2 and RASSF1A in human non-small cell lung cancer. Oncol Rep2006;15:1281–5.

38] Grijalva CG, Poehling KA, Edwards KM, Weinberg GA, Staat MA, Iwane MK,et al. Accuracy and interpretation of rapid influenza tests in children. Pediatrics2007;119:e6–11.

39] Greene MH, Feng Z, Gail MH. The importance of test positive predictive valuein ovarian cancer screening. Clin Cancer Res 2008;14:7574 [author 7577–9].

40] Valuev VP, Afoninkov DA, Petrenko OI, Grigorovich DA. Properties of artificialevolution of proteins and peptides. Mol Biol (Russ) 2001;35:1048–55.

41] Morozkin ES, Loseva EM, Morozov IV, Kurilshikov AM, Bondar AA, Rykova EY,et al. A comparative study of cell-free apoptotic and genomic DNA using FISHand massive parallel sequencing. Expert Opin Biol Ther 2012;12:S141–53.

42] Goebel G, Zitt M, Zitt M, Muller HM. Circulating nucleic acids in plasma or serum(CNAPS) as prognostic and predictive markers in patients with solid neoplasias.Dis Markers 2005;21:105–20.

43] Lo YM. Recent advances in fetal nucleic acids in maternal plasma. J HistochemCytochem 2005;53:293–6.