Correction

APPLIED BIOLOGICAL SCIENCESCorrection for “Glycosylation profiles determine extravasationand disease-targeting properties of armed antibodies,” by DarioVenetz, Christian Hess, Chia-wei Lin, Markus Aebi, and DarioNeri, which appeared in issue 7, February 17, 2015, of Proc Natl

Acad Sci USA (112:2000–2005; first published February 2, 2015;10.1073/pnas.1416694112).The authors note that Fig. 2 appeared incorrectly. The cor-

rected figure and its legend appear below.

www.pnas.org/cgi/doi/10.1073/pnas.1503039112

B C

TGE

SE

Ex vivo In vivo

0 60 120 180 240 300 36010

100

101

102

103

Time [min]

A

tumor liver lung spleen heart kidney intestine blood0

4

8

12

16

F8-IL

9 [%

ID/g

]

F8-IL

9 in

blo

od

[%ID

/g]

1-

Fig. 2. In vivo biodistribution profiles, microscopic analysis, and pharmacokinetic data. (A) Quantitative biodistribution profiles 24 h after i.v. administrationto F9-tumor–bearing mice. TGE (blue), SE (red), desialylated TGE (DS; gray), and deglycosylated SE (DG; black) derived F8-IL9 products are shown. (B) Jux-taposition of immunofluorescence detection images of F8-IL9 at 24 h after i.v. administration or after ex vivo application onto F9 tumor sections. (Scale bar:100 μm.) (C) Pharmacokinetic data of F8-IL9 from SE (red) and TGE (blue) during the first 6 h after injection.

E1508 | PNAS | March 24, 2015 | vol. 112 | no. 12 www.pnas.org

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

12, 2

020

Glycosylation profiles determine extravasation anddisease-targeting properties of armed antibodiesDario Venetza,1, Christian Hessa,1, Chia-wei Linb, Markus Aebib, and Dario Neria,2

aInstitute of Pharmaceutical Sciences, Department of Chemistry and Applied Biosciences, and bInstitute of Microbiology, Department of Biology, Swiss FederalInstitute of Technology, CH-8093 Zürich, Switzerland

Edited by Richard A. Lerner, The Scripps Research Institute, La Jolla, CA, and approved January 9, 2015 (received for review September 2, 2014)

The ability of antibodies to extravasate out of blood vessels iscritical for therapeutic activity, because molecular targets formost diseases are located outside of the endothelial lining. Byperforming detailed biodistribution studies with a novel IL9-armed cancer-specific antibody, we identified a clear correlationbetween N-linked glycan structures and tumor-targeting effi-ciencies. Site-specific glycan analysis provided a detailed viewof the glycan microheterogeneity present on the IL9 portion ofthe recombinant protein. Nonsialylated glycan structures havea negative impact on disease-homing activity, highlighting theimportance of glycosylation control and characterization duringprocess development.

tumor targeting | glycosylation | armed antibody | interleukin-9 |site-specific glycan analysis

Monoclonal antibodies represent the largest and fastest growingclass of pharmaceutical biotechnology products (1). Most

biopharmaceuticals, including antibody-based therapeutics, fea-ture posttranslational modifications such as protein N-glycosyl-ation and therefore rely on mammalian cell expression systems(2). For example, monoclonal antibodies, used in a human IgGformat, contain structurally distinct N-linked glycans at con-served positions within the Fc region. Depending on the IgGsubtype, Fc-glycosylation has been recognized to have a pro-found effect on the activation of immune cells (3, 4). Indeed,the first glyco-engineered antibody product, obinutuzumab(Gazyva), has recently been approved for chronic lymphocyticleukemia. The impact of protein glycosylation on pharmacoki-netics has been extensively studied for glycoprotein hormones,including the prominent examples of recombinant erythropoietinand its glyco-engineered derivative Darbepoetin alfa (5). Theintroduction of additional N-glycosylation motifs into the pep-tide sequence of erythropoietin can result in increased serumhalf-lives (6). Glycosylation also dictates the serum half-life ofglycoprotein drugs, capable of neonatal Fc-receptor–mediatedrecycling, as shown for the systemic TNF inhibitor Lenercept,a fusion protein consisting of an IgG1 Fc portion and the ex-tracellular p55 TNF receptor domain (7). However, quantitativestudies investigating the impact of protein glycosylation on dis-ease-homing properties of therapeutic proteins are rare.Exploiting their exquisite target selectivity and ability to lo-

calize at sites of disease, there is an emerging trend to usemonoclonal antibodies as pharmacodelivery vehicles, thus mov-ing from intact antibodies toward armed antibody products (8,9). The attachment of therapeutic payloads to a targeting anti-body can be accomplished either by chemical conjugation, in thecase of small molecules, or by genetic fusion of protein domains(9). The fusion of bioactive protein payloads (e.g., cytokines)may lead to additional O- or N-glycans in the resulting armedantibody when expressed in eukaryotic cell expression systems.Most antibody-mediated pharmacodelivery approaches rely

on extravasation of the biopharmaceutical product to diffuseinto tissues and reach the site of disease. Glycans present onthe therapeutic protein can modulate this process by differentglycan–receptor interactions in the bloodstream. For example,

hepatocytes express the asialoglyoprotein receptor with speci-ficity for nonsialylated proteins with terminally exposed galac-tose residues (10). Another receptor involved in glycoproteinhomeostasis, primarily expressed by macrophages and dendriticcells, is the mannose receptor, recognizing terminal mannose orN-acetylglucosamine (11). Certain glycan epitopes are also knownto be immunogenic and can lead to antidrug antibody responsesin humans (12).Our group has worked extensively on the production and in

vivo characterization of armed antibody products directed againstsplice isoforms of extracellular matrix components, which areundetectable in normal adult tissues, but abundantly expressedat sites of cancer and other inflammatory conditions (13). Inparticular, we have studied the high-affinity human monoclonalantibody F8, which is specific to the alternatively spliced extra-domain A (EDA) of fibronectin, a marker of angiogenesis ex-pressed in the subendothelial extracellular matrix of tumor bloodvessels (13, 14). The F8 antibody has been used for the phar-macodelivery of drugs, radionuclides, and cytokines to varioustypes of disease lesions (9, 15). Small bivalent antibody frag-ments without Fc portion may be preferred for the delivery ofhighly potent payloads, because they are rapidly cleared fromcirculation while exhibiting favorable biodistribution profiles(8). In this study, we found that variations in N-linked glycanstructures, present on the interleukin-9 (IL9) moiety of differentF8-based diabody fusion protein preparation, led to dramaticchanges in tumor targeting efficiencies, as revealed by quanti-tative biodistribution analysis.

Significance

Therapeutic antibodies represent the largest and fastest grow-ing class of biopharmaceuticals. There is a trend in movingfrom intact antibodies toward “armed” antibody products, inwhich the antibody moiety serves as pharmacodelivery vehicle.The impact of glycosylation on the targeting performance ofarmed antibodies is still largely unknown. Our article sheds lighton the surprising finding that relatively small variations in gly-costructures and sialic acid content can have dramatic effects ontherapeutic agent performance. A better understanding of theimpact of glycosylation on pharmaceutical activity is likely to berelevant not only for future antibody development activities, butalso for changes in current manufacturing processes and for thedevelopment of biosimilar products.

Author contributions: D.V., C.H., and D.N. designed research; C.-W.L. designed MS-basedglycan analysis; D.V., C.H., and C.-W.L. performed research; D.V., C.H., C.-W.L., and M.A.contributed new reagents/analytic tools; D.V., C.H., C.-W.L., M.A., and D.N. analyzed data;and D.V., C.H., C.-W.L., M.A., and D.N. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1D.V. and C.H. contributed equally to this work.2To whom correspondence should be addressed. Email: neri@pharma.ethz.ch.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1416694112/-/DCSupplemental.

2000–2005 | PNAS | February 17, 2015 | vol. 112 | no. 7 www.pnas.org/cgi/doi/10.1073/pnas.1416694112

Results and DiscussionWe focused our attention on IL9 as a therapeutic payload, basedon reports on its potent T-cell–mediated antitumor activities (16,17). IL9 is a special cytokine because it contains four distinct N-glycosylation sites, while being devoid of O-linked glycans (Fig.S1). In this study, we genetically fused IL9 to the C terminus ofa nonglycosylated F8-based diabody (Fig. S2) and expressed therecombinant immunocytokine either by transient gene expres-sion (TGE) or by stable expression (SE) in stably transfectedChinese hamster ovary (CHO) cells (Fig. 1 A and B) (18). Stablytransfected polyclonal CHO cells were cultured at a higher celldensity and modified medium composition. SDS/PAGE analysisshowed similar patterns for protein preparations obtained invarious experimental conditions. The mass difference of ∼10 kDaobserved for glycosylated F8-IL9 samples in SDS/PAGE sug-gested similar glycosylation site occupancies, irrespective of theproduction method. F8-IL9 preparations tested in vivo elutedas a single peak in gel filtration and displayed comparable EDA-binding kinetics in surface plasmon resonance (SPR) analysis(Fig. 1 D and E) . The products could be converted into a fullydeglycosylated form upon peptide-N-glycosidase F (PNGase F)treatment (Fig. 1C and Figs. S1 and S3). Ex vivo, all F8-IL9preparations selectively stained the subendothelial extracellular

matrix of blood vessels in murine F9 teratocarcinoma, regardlessof their production method and enzymatic modification (Fig. S4).When various batches of F8-IL9 were studied by quantitative

biodistribution analysis in immunocompetent 129/Sv mice bear-ing s.c. F9 tumors, a strikingly different tissue distribution profilewas observed for proteins produced by using either TGE or SEmethodologies (Fig. 2A). F8-IL9 derived from TGE cultures wasable to efficiently and selectively localize to tumors 24 h after i.v.injection with 11.53 ± 0.71% injected dose per gram (%ID/g) inthe neoplastic lesions while exhibiting favorable tumor-to-organratios (Fig. 2A). By contrast, batches of the same protein derivedfrom SE cultures failed to target tumors in vivo, with only 0.36 ±0.05%ID/g in the tumor and 0.16 ± 0.01%ID/g in blood. In linewith the biodistribution data, immunofluorescence microscopyrevealed that only TGE-derived F8-IL9 could successfully lo-calize to the perivascular space of tumor blood vessels in vivo(Fig. 2B). These biodistribution data were surprising, becausethe stable polyclonal expression cultures were directly derivedfrom preceding TGE cultures via antibiotic selection. Our find-ing proved to be highly reproducible because the data originatefrom three independent experiments for each type of sample.Additional biodistribution experiments, performed after treatingF8-IL9 with PNGase F, revealed that deglycosylated F8-IL9 fromSE cultures had regained its tumor-targeting ability (Fig. 2A).

0 8 16 24

[mAU

][R

U][R

U]

0 8 16 240

10

20

[mAU

]

0 8 16 240

4

8

0 8 16 240

4

80 8 16 24

0

10

20

0 8 16 240

5

10

0 8 16 240

5

100 8 16 240

6

12

[mL]

A

B

VL VH

IL9

N

N

C

C

IL9

C

• 5 106 cells/mL • PowerCHO

• PEI transfection • 1 106 cells/mL • ProCHO/PowerCHO

(1:1, 31°C/6d)

Transient Gene Expression (TGE) Stable

Expression (SE) SIP IL9VH VLpcDNA3.1_F8-IL9

NotI15aa 5aaNheI

Transfection

Selection

Neo>28d

1 2 3 1 2 3

TGE SE

70 50

35 25

DS DG [kDa]

(CHO-Scells)

D TGE1 TGE2 TGE3 DS

SE1 SE2 SE3 DG

0 200 4000

400

800

0 200 4000

400

800

0 200 4000

400

800

0 200 4000

400

800

0 200 4000

400

800

0 200 4000

400

800

0 200 4000

400

800

[s]

0 200 4000

400

800

[s]

E TGE1 TGE2 TGE3 DS

SE1 SE2 SE3 DG

15

100

0

5

10

(31°C/5d)

[mL] [mL] [mL]

[s][s]

Fig. 1. Production methods and protein characterization data of F8-IL9 preparations investigated in vivo. Data from TGE (blue) and SE (red) F8-IL9 batches(1–3) are displayed. Desialylated SE (DS; gray) and deglycosylated TGE (DG; black) F8-IL9 were compared. (A) Expression vector and production methods. (B)Schematic of homobivalent diabody format. (C) Analytical SDS/PAGE analysis. (D) Gel-filtration chromatograms. (E) EDA-binding sensograms (SPR analysis).

Venetz et al. PNAS | February 17, 2015 | vol. 112 | no. 7 | 2001

APP

LIED

BIOLO

GICAL

SCIENCE

S

Importantly, TGE- and SE-produced F8-IL9 displayed closeto identical biodistribution profiles after enzymatic deglycosy-lation with PNGase F (Fig. 2A and Fig. S5), indicating that theprotein components were of equivalent quality. F8-IL9 pro-duced by TGE also failed to target the tumor neovasculature ofF9 tumors upon enzymatic removal of terminal sialic acids byα2-3,6,8,9 neuraminidase (Fig. S5), pointing out the special roleof this carbohydrate residue.The quantitative biodistribution profiles shown in Fig. 2A

provide a global view of potential glycan-mediated F8-IL9interactions in vivo. A potential lectin-trapping mechanism (e.g.,by immune cells) would be detectable by elevated radioactivitylevels in blood and in the spleen. However, only low levels ofradiolabeled F8-IL9 were found in normal organs, with the ex-ception of intestinal uptake, which is often observed using anti-EDA antibody products. We therefore assumed that changes inglycostructures could have an impact both on drug clearance andon extravasation. To support this conclusion, a formal pharma-cokinetic analysis comparing radiolabeled F8-IL9 samples fromeither TGE or SE cultures was performed (Fig. 2C). However,F8-IL9 produced by the SE method displayed only slightly fasterblood clearance, suggesting that the two products could also differin their ability to cross the tumor endothelium.The model system used is appropriate to study extravasation

and vascular targeting because of the specific site of expressionof the target antigen EDA in the subendothelial matrix of thetumor-associated vasculature (Fig. 2B and Fig. S4). High ex-pression levels of EDA at this site prevent saturation effects overa wide dose range (19). Additionally, the turnover rate for thistype of antigens is considered to be very low because boundantibody can be detected up to 5 d after injection (10). Targetingof solid tumors and metastases is often limited by the buildup ofan antigen barrier in proximity to the neovasculature (20). Hence,our findings may also be relevant for other antibody-based phar-macodelivery approaches.F8-IL9 glycoforms produced by either SE or TGE were ex-

tensively characterized by using several complementary methods.Hydrophilic interaction chromatography (HILIC)-HPLC–basedglycoprofiling of fluorescently labeled glycan pools after PNGaseF treatment showed substantial differences between TGE andSE samples (Fig. 3A). We also quantified the amount of terminalN-acetylneuraminic acid (Neu5Ac) moieties in various protein

preparations, using two orthogonal assays based either onchemical or enzymatic sialic acid release, followed by differentfluorescent labeling and detection methods as described inMethods.The TGE-produced batches of F8-IL9 exhibited significantly higherNeu5Ac-to-protein ratios, compared with the same protein derivedfrom stably transfected cells in both assays (Fig. 3 B and C). Bycomparison, the glycan profiles before and after α2-3,6,8,9 neur-aminidase treatment revealed characteristic peak shifts for theTGE samples, again indicating terminally sialylated N-glycans(Fig. 3A). Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF/TOF MS) analysis ofpermethylated glycans further confirmed this finding (Fig. 3D).The major N-glycan structures from TGE-derived F8-IL9 weresialylated N-glycans with core fucosylation. On the contrary, MSprofiling showed that incomplete galactosylated and neutral N-glycans were the main species present on SE-derived F8-IL9.Further, we have predominantly found α-2,3–linked sialic acidson TGE product as confirmed by enzymatic treatment with α-2,3-neuraminidase. Surprisingly, we have also observed antennary N-acetyllactosamine (LacNAc) repeats on the glycans of SE F8-IL9.Such poly-LacNAc structures represent ligands for galectins,which are known to be involved in cell adhesion and tumor pro-gression (21, 22). A detailed list of all observed N-glycan struc-tures is given in Table S1.IL9 features four putative glycosylation sites. To obtain a de-

tailed view of the glycan microheterogeneity, site-specific gly-cosylation analysis was performed. After proteolytic cleavage,the four sets of F8-IL9 derived glycopeptides were analyzed bynanoLC–higher-energy collisional dissociation (HCD) tandemmass spectrometry (MS/MS). Sialylated glycans were againidentified with the help of α2-3,6,8,9 neuraminidase. Site-specificN-linked glycan structures for representative SE and TGE prep-arations are summarized in Fig. 4. All annotated glycan structuresand corresponding glycopeptides are summarized in Table S2. MSand MS/MS spectra for glycosylation site 2 are exemplified in Fig.S6. Consistent with the data mentioned above, sialylatedN-glycanswere again predominantly found in the TGE preparations butwere also present in the SE product. Notably, sialylation differ-ences were site-specific; i.e., terminal sialic acids as well as galactoseresidues at the penultimate position were absent at glycosylation site3 in both types of products (Fig. 4).

B C

TGE

SE

In vivo Ex vivo

0 60 120 180 240 300 36010

100

101

102

103

Time [min]

A

tumor liver lung spleen heart kidney intestine blood0

4

8

12

16

F8-IL

9 [%

ID/g

]

F8-IL

9 in

blo

od

[%ID

/g]

1-

Fig. 2. In vivo biodistribution profiles, microscopic analysis, and pharmacokinetic data. (A) Quantitative biodistribution profiles 24 h after i.v. administrationto F9-tumor–bearing mice. TGE (blue), SE (red), desialylated TGE (DS; gray), and deglycosylated SE (DG; black) derived F8-IL9 products are shown. (B) Jux-taposition of immunofluorescence detection images of F8-IL9 at 24 h after i.v. administration or after ex vivo application onto F9 tumor sections. (Scale bar:100 μm.) (C) Pharmacokinetic data of F8-IL9 from SE (red) and TGE (blue) during the first 6 h after injection.

2002 | www.pnas.org/cgi/doi/10.1073/pnas.1416694112 Venetz et al.

Together, the absence of terminal sialic acids and the exposureof terminal galactose or N-acetylglucosamine residues controlboth the elimination of the fusion protein from blood and itsextravasation properties. Low levels of terminal sialic acids canlead to unacceptably fast clearance, as previously described (5).A contribution to protein clearance of either the asialoglyco-protein receptor and of the mannose receptor would be com-patible with our analytical data.Changes in bioavailability and extravasation of biophar-

maceuticals, including therapeutic antibodies, can have a pro-found impact on safety and therapeutic action (23). Indeed,substantial differences in pharmacokinetics and biological ac-tivities have been reported for products undergoing changes inmanufacturing processes (24). However, variations in extrava-sation rates and disease-homing properties may often go un-noticed, unless quantitative biodistribution studies are performed.The findings of this study suggest that the engineering and devel-opment of biopharmaceuticals may favor either a complete ab-sence of glycostructures or, alternatively, the engineering of well-defined sialylated carbohydrates (25–27). Pharmacokinetic andbiodistribution studies combined with thorough glycan profiling

appear mandatory for therapeutic glycoproteins, especially whenthe production process is modified or when biosimilar productsare developed.

MethodsCloning of F8-IL9 Fusion Proteins. Mouse IL9 (mIL9) cDNA (Sino Biological) wasamplified by PCR using the forward primer link15-mIL9_for (5′-TCA-GGCGGAGGTGGCTCTGGCG GTGGCGGATCACAGAGATGCAGCACCACAT-GGGGC-3′) that appends the C-terminal part of the flexible (Gly4Ser)3-linkersequence and the reverse primer mIL9-NotI_rev (5′-TTTTCCTTTTG CGGCCG-CTCACTATGGTCGGCTTTTCTGCCTTTGCATCTC-3′) encoding two stop codonsand a NotI restriction site. The F8 diabody gene was PCR-amplified by usingthe primers SIP-F8_for (5′-CCTGTTCCTCGTCGCTGTGGCTACAGGTGTGCACT-CGGAGGTGCAGCTGTTGGAGTCTGGGG-3′) that appends part of the signalpeptide (SIP) sequence and F8-link15_rev (5′-CCGCCA-GAGCCACCTCCGCCT-GAACCGCCTCCACCTTTGATTTCCACCTTGGT CCCTTGG-3′), which encodes theN-terminal part of the (Gly4Ser)3-linker. The mIL9 and F8-diabody DNAfragments were assembled by PCR and amplified by using the primers NheI-SIP_for (5′-CCCGCTAGCGTCGACCATGGGCTGGAGCCTGATCCTCCTGTTCCTC-GTCGCTGTGGC-3′), containing an NheI restriction site followed by the N-terminal part of the SIP sequence, and mIL9-NotI_rev. The PCR-assembledfull-length F8-IL9 immunocytokine gene was double-digested with NheI andNotI restriction endonucleases (New England Biolabs) and ligated into the

Fluo

resc

ence

A

C

D[GU] [GU]

0 1 2 3 4DGDS

3 2 1 3 2 1

TGE

SE

α2-3,6,8,9 Neuraminidase

5 7.5 10 12.5 150

60120180240

5 7.5 10 12.5 150

255075

100

5 7.5 10 12.5 150

4080

120160

5 7.5 10 12.5 150

20406080

Neu5NAc/F8-IL9 [mol/mol]

[mV]

Fluo

resc

ence

[m

V]

1590

.8

1579

.8

1835

.9

2081

.022

44.1

2605

.326

93.3

2966

.530

54.5

3415

.7 3776

.938

64.9 42

26.1

5036

.5

5485

.7

4587

.3

Inte

nsity

[%]

100

80

60

40

20

1300 2240 3180 4120 5060 6000

1449.7

0

1590

.8 1835

.9

2081

.1 2326

.2

2530

.324

89.3

2693

.427

34.4

2938

.5

3591

.8

4041

.1

3143

.2

2244

.1

) 4

) 5 ) 6

Mass [m/z]

Inte

nsity

[%]

100

80

60

40

20

0 1300 2240 3180 4120 5060 6000

4386.7

SE3 TGE3

SE3

TGE3

B TGE

SE

0 1 2 3 4DGDS

3 2 1 3 2 1

Fig. 3. Characterization data of N-linked glycan pools. F8-IL9 samples from TGE (blue), SE (red), TGE after desialylation (DS; gray), and SE after deglyco-sylation (DG; black) were analyzed. (A) Representative HILIC-HPLC profiles of 2-AB–labeled N-glycan pools after PNGase F release and neuraminidasetreatment. (B) Quantification of Neu5Ac by reversed-phase HPLC upon mild acid hydrolysis. (C) Fluorimetric quantification of terminal Neu5Ac after enzy-matic release with α2-3,6,8,9 neuraminidase. (D) MALD-TOF/TOF spectra of permethylated glycan pools. All color schemes followed the guidelines of theConsortium for Functional Glycomics (33).

Venetz et al. PNAS | February 17, 2015 | vol. 112 | no. 7 | 2003

APP

LIED

BIOLO

GICAL

SCIENCE

S

mammalian cell expression vector pcDNA3.1(+) (Life Technologies) with T4ligase (New England Biolabs).

Cell Lines and Protein Expression. Percentages (%) are based on vol/vol unlessstated differently. Transiently expressed F8-IL9 fusion proteins were pro-duced in suspension cultures of Freestyle CHO (CHO-S) cells (Life Technolo-gies). The medium was composed of a 1:1 mixture of ProCHO-4 (Lonza) andPowerCHO-2 CD (Lonza), both supplemented with HT supplement contain-ing 100 μM hypoxanthine and 16 μM thymidine (Life Technologies), 2 mMUltraglutamine (Lonza), and 1% antibiotic–antimycotic solution (Life Tech-nologies). Protein expression occurred in a shaking incubator at 31 °C for 6 d,starting at an initial cell density of 1 × 106 cells per mL.

To generate stably transfected cells, 10 mL of the TGE culture were taken24 h after transfection, centrifuged (at 1,000 × g for 4 min), and resuspendedin RPMI medium (Life Technologies) supplemented with 10% FCS (LifeTechnologies), 1% antibiotic–antimycotic solution, and 0.5 mg/mL Geneticin(Life Technologies). Stable integration into the CHO-S genome was achievedafter cultivation for >28 d at 37 °C and 5% CO2 under antibiotic selection.Polyclonal stably transfected cells were then grown in suspension at 37 °C inPowerCHO-2 CD medium supplemented as described above with HT sup-plement, Ultraglutamine, and antibiotic–antimycotic solution. As soon as thecells reached a density of 4.5–5 × 106 cells per mL, cultures were transferred toa 31 °C shaking incubator for protein expression until day 5. F9 teratocarci-noma cells (ATCC no. CRL-1720) were grown according to supplier’s protocol in0.1% gelatin-coated tissue culture flasks in DMEM (Life Technologies) sup-plemented with 10% FCS and 1% antibiotic–antimycotic solution.

Protein Purification and Characterization. Both stably and transiently ex-pressed fusion protein preparations were purified from the supernatant tohomogeneity by protein-A (Sino Biological) affinity chromatography andfurther analyzed by SDS/PAGE (NuPAGE system; Life Technologies), size-exclusion chromatography (gel-filtration) on a Superdex S200 10/300GL column(GEHealthcare), and SPR analysiswith a Biacore 3000 system (GEHealthcare) ona CM5 sensor chip coated with ∼1,500 resonance units of EDA antigen per-formed as described (13).

Immunofluorescence Detection. Ex vivo immunofluorescence staining withF8-IL9 preparations was performed with F9 teratocarcinoma tumors from129/Sv mice (Charles River), which served as control mice in previous therapyexperiments. In vivo immunofluorescence analysis was performed with F9tumors, which were excised 24 h after i.v. injection of F8-IL9 samples andembedded in NEG-50 cryo-embedding medium (Thermo Scientific). Tumorsections of 10 μm were stained by using the following primary antibodies:PECAM1 goat anti-mCD31 (M-20; Santa Cruz Biotechnology) and RM4A9 ratanti-mouse IL9 antibody (BioLegend). Detection was accomplished with anti-goat Alexa Fluor 488 and anti-rat Alexa Fluor 594-coupled secondary anti-bodies (Life Technologies). The slides were analyzed with an Axioskop 2 Plusmicroscope (Zeiss) and processed with Adobe Photoshop.

Biodistribution and Pharmacokinetic Analysis. Twelve-week-old female 129/Svmice were s.c. injected in the flank with 2.5 × 107 F9 teratocarcinoma cells. Assoon as tumor sizes were >50 mm3, mice were grouped (n = 4), and ∼10 μg

of 125I-labeled protein was injected into the lateral tail vein as described (13).Mice were killed 24 h after injection, organs were excised and weighed, andradioactivity was quantified with a Packard Cobra γ-counter. Values aregiven in %ID/g ± SD (wt/wt). The pharmacokinetic analysis was performed asfollows: The radiolabeling of F8-IL9 samples and i.v. administration intohealthy 129/Sv mice (n = 4) was carried out as described for the quantitativebiodistribution analysis. Blood samples at suitable time points were taken bypuncturing the lateral tail vein and withdrawing 1–5 μL of blood, which wasanalyzed by scintillation counting. One mouse per group was killed after 20min to obtain short-time biodistribution profiles. Experiments were per-formed under the project license granted by the local authority Verter-inäramt des Kantons Zürich, Switzerland (License No. 42/2012).

Sialic Acid-Release and Quantification. Sialic acids were released enzymaticallyvia α2-3,6,8,9 neuraminidase (New England Biolabs). Specifically, 3 U of en-zyme per μg of F8-IL9 was incubated at 37 °C for 16 h. The samples forsubsequent in vivo experiments were then purified via protein-A affinitychromatography. Alternatively, sialic acids were released by mild hydrolysisin 0.5 M NaHSO4 (Sigma Aldrich), requiring an incubation period of 40 minat 80 °C (28). Enzymatically released Neu5Ac was quantified by usinga fluorimetric Neu5Ac assay kit (BioVision). Neu5Ac (Sigma Aldrich) repre-sents the most common sialic acid species and was used as a standard forcalibration for both assays. Fluorescence was measured in a SpectraMaxParadigm plate reader (Molecular Devices) at excitation/emission (Ex/Em)wavelengths of 535/587 nm. Hydrolyzed sialic acids were fluorescently la-beled with o-phenylenediamine (Sigma Aldrich) in 0.5 M NaHSO4 for 2 h at80 °C. RP-HPLC was performed with a Hitachi Lachrom D-7000 HPLC-system(Merck) equipped with an Xterra 5-μm, 4.6- × 150-mm C18 column (Waters).Sialic acid derivatives were eluted by using an isocratic buffer system anddetected at Ex/Em 280/425 nm as described (28). Data points were normal-ized per F8-IL9 monomer (average of triplicates ± SD).

Enzymatic Release of N-Glycan Pools. N-linked oligosaccharides were releasedby glycerol-free PNGase F (New England Biolabs). Specifically, 10 U of en-zyme per μg of protein were incubated for 24 h at 37 °C. Samples for bio-distribution experiments were deglycosylated in PBS buffer, followed by anadditional protein A purification step to remove the enzyme and con-taminants. All other samples were dialyzed against 50 mM ammonium bi-carbonate (Sigma Aldrich) (pH 8.0) buffer before deglycosylation reactionsfor a prolonged incubation time of up to 24 h. Released glycans were thenseparated form proteins by Vivaspin 500 centrifugal filter units (SartoriusStedim) with a 10-kDa cutoff and washed with 4 × 400 μL of deionizedwater. The flow-through was collected and vacuum-dried.

HILIC-HPLC–Based Glycoprofiling. Vacuum-dried N-glycan samples were flu-orescently labeled with 2-aminobenzamide (2-AB) (29), by using a Glyco-profile labeling kit (Sigma Aldrich) and purified via GlycoClean S cartridges(Prozyme), according to the providers’ instructions. Samples were againreduced to dryness in a vacuum centrifuge before they were dissolvedin 200 μL of 50% acetonitrile/water. HPLC-HILIC was performed with theHitachi Lachrom D-7000 HPLC system equipped with a TSKgel Amide-80 column (TOSOH Bioscience). The injected sample volume was 12 μL.

)0-1 )0-2 )0-3 )2-5

)0-1 )0-2 )0-3 )2-3 )2-4 )1-2 )1-3 )0-1 )0-2 )0-3 )0-1

)0-1 )0-2 )0-3 )2-5 )0-2

SE3

TGE3

)0-1

)2-3

)0-1 )0-2 )0-3

+H3N COO- CSCSGNVTSCLCLSVPTDDCTTPCYR NITCPSFSCEK GLLQLTN ATQK EKPCNQTMAGNTLSFLK

SE3

TGE3

)0-3 )0-3

)0-3 )0-3 )0-2 )2-3 )0-2 )0-3

Fig. 4. Site-specific glycan microheterogeneity. Representative TGE- and SE-derived F8-IL9 samples were subjected to nanoLC-HCD-MS/MS glycopeptideanalysis. Peptide sequences upon cleavage by Trypsin, Glu-C endopeptidase, and AspN endopeptidase are shown in the center. Glycosylation sites arenumbered from left to right (sites 1–4).

2004 | www.pnas.org/cgi/doi/10.1073/pnas.1416694112 Venetz et al.

Gradient elution was achieved by using a buffer system consisting of ace-tonitrile (Sigma Aldrich) and 0.1 M ammonium formate (ARCOS Chemicals)(pH 4.5) solution. A gradient of 0.3%/min starting with 24% 0.1 M am-monium formate provided the highest resolution. A dextran calibrationladder (Waters) was run after each sample as external standard to calibratethe system for subsequent conversion of retention times (minutes) intoglucose units (GU).

MALDI-TOF/TOF Analysis for Permethylated Glycans. N-glycans from 50 μg ofpurified F8-IL9 proteins were released by PNGase F (Promega), as describedabove. To verify glycosidic α-2,3–linkage of terminal sialic acids, sampleswere incubated with 20 U of α-2,3-neuraminidase (New England Biolabs) perμg of F8-IL9 for 16 h at 37 °C. The released glycans from all samples wereisolated by C18 Sep-Pak cartridge (Waters) and permethylated as describedby Dell et al. (30). Permethylated glycans samples were mixed 1:1 withdihydroxy-benzonic acid matrix (15 mg/mL in 75% acetonitrile in water with0.1% formic acid) and then spotted onto a MALDI-TOF/TOF MS target plate.Data acquisition was performed manually on a Model 4800 ProteomicsAnalyzer (Applied Biosystems) with a Nd:YAG laser, and 1,000 shots wereaccumulated in the reflectron positive ion mode. MALDI-TOF/TOF massspectrometer was calibrated externally by permethylated N-glycans fromRNase B (Sigma). N-glycan structures ofm/z 2081.1 and 2,362.2 from SE F8-IL9were further confirmed by MALDI-TOF/TOF MS/MS. All other annotationsare based on the current knowledge for the N-glycosylation synthesispathway, and the m/z value of each peak was labeled as monoisotopic massfor spectra acquired in reflectron mode. Data interpretation was processedmanually or with the help of GlycoWorkbench (Version 2.0; ref. 31).

Sample Preparation for Glycopeptides. A total of 50-μg purified F8-IL9 proteinsamples were digested by filter-assisted sample preparation procedure (32)before MS measurement. Briefly, proteins were reduced by 50 mM DTT in50 mM ammonium bicarbonate buffer (pH 8.5) at 37 °C for 1 h, followingby alkylation by 65 mM iodoacetamide at 37 °C in the dark for 1 h. Afterwashing the filter device with ammonium bicarbonate buffer four times,proteins were digested by sequencing-grade modified trypsin (Promega) at aratio of 50:1 at 37 °C overnight. All digested peptides and glycopeptides werecollected by centrifugation and dried with a vacuum centrifuge. Additionally,two-thirds of the samples were further treated with Glu-C endopeptidase(Promega), AspN endopeptidase (Promega), neuraminidase (Calbiochem),

and PNGase F (Roche) individually. All samples were desalted by Zip-Tip C18(Millipore) before nanoLC-MS/MS analysis.

Glycopeptide Analysis by NanoLC-HCD-MS/MS. Samples were analyzed ona calibrated LTQ-OrbitrapVelosmass spectrometer (ThermoScientific) coupledto an Eksigent-Nano-HPLC system (Eksigent Technologies). Peptides wereresuspended in 2.5% acetonitrile and 0.1% formic acid and loaded on a self-made fritted column (75 μm × 150 mm) packed with reverse-phase C18 ma-terial (AQ, 1.9 μm 200 Å; Bischoff GmbH) and eluted with a flow rate of 300nL/min by a gradient from 3% to 30% of B in 22 min, 50% of B in 25 min, and97% of B in 27 min. One scan cycle comprised a full-scan MS survey spectrum,followed by up to 10 sequential HCD MS/MS on the most intense signalsabove a threshold of 2,000. Full-scan MS spectra (700–2,000 m/z) were ac-quired in the FT-Orbitrap at a resolution of 60,000 at 400 m/z, and HCD MS/MS spectra were recorded in the FT-Orbitrap at a resolution of 15,000 at 400m/z. HCD was performed with a target value of 1e5, and stepped collisionenergy rolling from 35, 40, and 45 V was applied. AGC target values were 5e5for full Fourier transformMS. For all experiments, dynamic exclusion was usedwith one repeat count, 15-s repeat duration, and 60-s exclusion duration.

Database Search and Site-Specific Glycosylation Analysis. MS and MS/MS datawere processed into the Mascot generic format files and searched against theSwissprot database (Version 201408) through Mascot engine (Version 2.4)with the consideration of carbamidomethylation at cysteine and oxidation atmethionine. For PNGase F digestions, deamination was considered as a var-iable modification. The monoisotopic masses of 2+ or more charged peptideswere searched with a peptide tolerance of 10 ppm and a MS/MS tolerance of0.25 Da for fragment ions. Only peptides with a maximum of two missingcleavage sites were allowed in database searches. Positive identification ofdeaminated peptides was performed by using a variety of strict criteria,including manual inspection of spectra. For site-specific glycosylation anal-ysis, all data were interpreted manually. Here, XCalibur (Version 2.2 sp1.48)was used for data analysis.

ACKNOWLEDGMENTS. We thank Dr. Danilo Ritz, Dr. Tim Fugmann, andDr. Vivianne Otto for useful discussions. We give special thanks to Dr. PeterGehrig (Functional Genomics Center Zürich) for help with the mass spec-trometer and to Ruth Alder and Prof. Dr. Irmgard Werner for providingthe HPLC system. This research was supported by the ETH Zürich, the SwissNational Science Foundation, and Philochem AG.

1. Walsh G (2014) Biopharmaceutical benchmarks 2014. Nat Biotechnol 32(10):992–1000.2. Walsh G, Jefferis R (2006) Post-translational modifications in the context of thera-

peutic proteins. Nat Biotechnol 24(10):1241–1252.3. Kaneko Y, Nimmerjahn F, Ravetch JV (2006) Anti-inflammatory activity of immuno-

globulin G resulting from Fc sialylation. Science 313(5787):670–673.4. Ferrara C, et al. (2011) Unique carbohydrate-carbohydrate interactions are required

for high affinity binding between FcgammaRIII and antibodies lacking core fucose.Proc Natl Acad Sci USA 108(31):12669–12674.

5. Sinclair AM, Elliott S (2005) Glycoengineering: The effect of glycosylation on theproperties of therapeutic proteins. J Pharm Sci 94(8):1626–1635.

6. Elliott S, et al. (2003) Enhancement of therapeutic protein in vivo activities throughglycoengineering. Nat Biotechnol 21(4):414–421.

7. Jones AJ, et al. (2007) Selective clearance of glycoforms of a complex glycoproteinpharmaceutical caused by terminal N-acetylglucosamine is similar in humans andcynomolgus monkeys. Glycobiology 17(5):529–540.

8. Wu AM, Senter PD (2005) Arming antibodies: Prospects and challenges for im-munoconjugates. Nat Biotechnol 23(9):1137–1146.

9. Hess C, Venetz D, Neri D (2014) Emerging classes of armed antibody therapeuticsagainst cancer. Medchemcomm 5(4):408–431.

10. Morell AG, Gregoriadis G, Scheinberg IH, Hickman J, Ashwell G (1971) The role ofsialic acid in determining the survival of glycoproteins in the circulation. J Biol Chem246(5):1461–1467.

11. Lee SJ, et al. (2002) Mannose receptor-mediated regulation of serum glycoproteinhomeostasis. Science 295(5561):1898–1901.

12. Ghaderi D, Taylor RE, Padler-Karavani V, Diaz S, Varki A (2010) Implications of thepresence of N-glycolylneuraminic acid in recombinant therapeutic glycoproteins. NatBiotechnol 28(8):863–867.

13. Villa A, et al. (2008) A high-affinity human monoclonal antibody specific to the al-ternatively spliced EDA domain of fibronectin efficiently targets tumor neo-vascula-ture in vivo. Int J Cancer 122(11):2405–2413.

14. Rybak JN, Roesli C, Kaspar M, Villa A, Neri D (2007) The extra-domain A of fibronectinis a vascular marker of solid tumors and metastases. Cancer Res 67(22):10948–10957.

15. Perrino E, et al. (2014) Curative properties of noninternalizing antibody-drug con-jugates based on maytansinoids. Cancer Res 74(9):2569–2578.

16. Purwar R, et al. (2012) Robust tumor immunity to melanoma mediated by interleukin-9-producing T cells. Nat Med 18(8):1248–1253.

17. Lu Y, et al. (2014) Tumor-specific IL-9-producing CD8+ Tc9 cells are superior effectorthan type-I cytotoxic Tc1 cells for adoptive immunotherapy of cancers. Proc Natl AcadSci USA 111(6):2265–2270.

18. Rajendra Y, Kiseljak D, Baldi L, Hacker DL, Wurm FM (2011) A simple high-yieldingprocess for transient gene expression in CHO cells. J Biotechnol 153(1-2):22–26.

19. Gutbrodt KL, Casi G, Neri D (2014) Antibody-based delivery of IL2 and cytotoxicseradicates tumors in immunocompetent mice. Mol Cancer Ther 13(7):1772–1776.

20. Graff CP, Wittrup KD (2003) Theoretical analysis of antibody targeting of tumorspheroids: Importance of dosage for penetration, and affinity for retention. CancerRes 63(6):1288–1296.

21. Hughes RC (2001) Galectins as modulators of cell adhesion. Biochimie 83(7):667–676.22. Liu FT, Rabinovich GA (2005) Galectins as modulators of tumour progression. Nat Rev

Cancer 5(1):29–41.23. Carter PJ (2006) Potent antibody therapeutics by design. Nat Rev Immunol 6(5):

343–357.24. Schiestl M, et al. (2011) Acceptable changes in quality attributes of glycosylated

biopharmaceuticals. Nat Biotechnol 29(4):310–312.25. Brik A, Ficht S, Wong CH (2006) Strategies for the preparation of homogenous gly-

coproteins. Curr Opin Chem Biol 10(6):638–644.26. Schwarz F, et al. (2010) A combined method for producing homogeneous glyco-

proteins with eukaryotic N-glycosylation. Nat Chem Biol 6(4):264–266.27. Meuris L, et al. (2014) GlycoDelete engineering of mammalian cells simplifies N-gly-

cosylation of recombinant proteins. Nat Biotechnol 32(5):485–489.28. Anumula KR (1995) Rapid quantitative determination of sialic acids in glycoproteins

by high-performance liquid chromatography with a sensitive fluorescence detection.Anal Biochem 230(1):24–30.

29. Bigge JC, et al. (1995) Nonselective and efficient fluorescent labeling of glycans using2-amino benzamide and anthranilic acid. Anal Biochem 230(2):229–238.

30. Dell A, et al. (1994) Mass spectrometry of carbohydrate-containing biopolymers.Methods Enzymol 230:108–132.

31. Ceroni A, et al. (2008) GlycoWorkbench: A tool for the computer-assisted annotationof mass spectra of glycans. J Proteome Res 7(4):1650–1659.

32. Wi�sniewski JR, Zougman A, Nagaraj N, Mann M (2009) Universal sample preparationmethod for proteome analysis. Nat Methods 6(5):359–362.

33. Varki A, et al. (2009) Symbol nomenclature for glycan representation. Proteomics9(24):5398–5399.

Venetz et al. PNAS | February 17, 2015 | vol. 112 | no. 7 | 2005

APP

LIED

BIOLO

GICAL

SCIENCE

S