HLA ligandome analysis identifies the underlyingspecificities of spontaneous antileukemia immuneresponses in chronic lymphocytic leukemia (CLL)Daniel J. Kowalewskia, Heiko Schustera, Linus Backerta,b, Claudia Berlina,c, Stefan Kahna, Lothar Kanzc,Helmut R. Salihc,d, Hans-Georg Rammenseea,e, Stefan Stevanovica,e, and Juliane Sarah Stickelc,1

aInstitute for Cell Biology, Department of Immunology, University of Tübingen, Tübingen 72076, Germany; bApplied Bioinformatics, Center for Bioinformaticsand Department of Computer Science, University of Tübingen, Tübingen 72076, Germany; cDepartment of Hematology and Oncology, University ofTübingen, Tübingen 72076, Germany; dClinical Cooperation Unit Translational Immunology, DKFZ Partner Site Tübingen, German Cancer Consortium (DKTK),Tübingen 72076, Germany; and eDKFZ Partner Site Tübingen, German Cancer Consortium (DKTK), Tübingen 72076, Germany

Edited by Lewis L. Lanier, University of California, San Francisco, CA, and approved December 1, 2014 (received for review August 25, 2014)

The breakthrough development of clinically effective immunecheckpoint inhibitors illustrates the potential of T-cell–based im-munotherapy to effectively treat malignancies. A remaining chal-lenge is to increase and guide the specificities of anticancerimmune responses, e.g., by therapeutic vaccination or by adoptiveT-cell transfer. By analyzing the landscape of naturally presentedHLA class I and II ligands of primary chronic lymphocytic leukemia(CLL), we delineated a novel category of tumor-associated T-cellantigens based on their exclusive and frequent representation inthe HLA ligandome of leukemic cells. These antigens were vali-dated across different stages and mutational subtypes of CLLand found to be robustly represented in HLA ligandomes ofpatients undergoing standard chemo-/immunotherapy. We dem-onstrate specific immune recognition of these antigens exclusivelyin CLL patients, with the frequencies of representation in CLLligandomes correlating with the frequencies of immune recognitionby patient T cells. Moreover, retrospective survival analysis revealedsurvival benefits for patients displaying immune responses to theseantigens. These results directly imply these nonmutant self-peptidesas pathophysiologically relevant tumor antigens and encouragestheir implementation for cancer immunotherapy.

cancer immunotherapy | chronic lymphocytic leukemia | tumor-associatedantigens | therapeutic vaccination | HLA

For a long time peptide-based immunotherapy fell short of itspotential to achieve meaningful responses in the clinical

setting of cancer therapy (1, 2). A disequilibrium between pro-posed tumor-associated antigens on the input side (3) andfunctional vaccines on the output side (4–6) became strikinglyapparent. This might in part have been due to the distorted re-lationship of gene expression and HLA restricted presentation ofthe corresponding gene product (7, 8). In contrast, a recent studyon vaccination with naturally presented tumor-associated pep-tides identified by direct analysis of primary tumor-derived HLAligands reported specific, nontoxic vaccine-induced immuneresponses, which were associated with improved clinical outcome(6). On the other hand, experimental cancer immunotherapyusing adoptive transfer of antigen-specific T cells showed highefficiency but also severe toxicity in some patients due to pres-ence of target antigens on normal tissues (7). Both situationsunderscore the importance and potential of identifying (patho-)physiologically relevant targets by direct differential analysis ofthe entire landscape of HLA-presented peptides, termed theHLA ligandome.Here, we mapped the nonmutant HLA ligandome of chronic

lymphocytic leukemia (CLL) with the aim of developing a CLL-specific multipeptide vaccine. The immunogenicity of CLL, asrevealed in GvL-effects and cases of spontaneous remissions (8–10), as well as favorable immune effector-to-target cell ratiospresent in situations with minimal residual disease (MRD) suggest

that CLL might be effectively targeted by T-cell–based immuno-therapy (11, 12). Furthermore, the highly variable and in somepatients prolonged course of disease indicates an underlyingmechanism of tumor control (13). However, so far only very fewCLL-associated antigens that are able to elicit specific T-cellresponses have been described (14–16).We identified a novel category of ligandome-defined tumor-

associated antigens (LiTAAs; source proteins of HLA ligands ontumor cells), which were frequently and exclusively detected inCLL patients. Specific immune recognition of the correspondingHLA ligands (LiTAPs) was observed exclusively in CLL patients,remarkably showing a direct correlation with the frequency ofHLA restricted presentation. Furthermore, retrospective survivalanalysis points to an association of LiTAP-specific immuneresponses with improved overall survival in CLL patients.

ResultsPrimary CLL Cells Display No Loss or Down-Regulation of HLAExpression. HLA loss or down-regulation in malignancies maypose a major limitation for T-cell–based immunotherapy (17).Therefore, we determined the HLA expression levels on primaryCD19+CD5+ CLL cells compared with autologous CD19+CD5−

B lymphocytes. HLA surface levels were quantified by flowcytometry in a panel of 7 CLL patients, which revealed patient-

Significance

Effective cancer immunotherapy relies on specific immunerecognition of tumor-associated and tumor-specific antigens. Inchronic lymphocytic leukemia (CLL), the highly variable courseof disease implicates an underlying mechanism of tumor con-trol by the immune system. In this study, we directly analyzedthe landscape of naturally presented CLL antigens and identi-fied targets conveying immune protection. These novel anti-gens might be valuable both for patient stratification and forinducing therapeutic antitumor immunity. Taken together, wedemonstrate that CLL is subject to spontaneous T-cell respon-ses targeting nonmutated antigens, which are associated withimproved patient survival and provide novel options for theimmunotherapy of CLL.

Author contributions: D.J.K., H.-G.R., and J.S.S. designed research; D.J.K., H.S., C.B., S.K.,and J.S.S. performed research; D.J.K., L.B., and S.S. contributed new reagents/analytictools; D.J.K., H.S., L.B., L.K., H.R.S., H.-G.R., S.S., and J.S.S. analyzed data; and D.J.K.,L.K., H.R.S., H.-G.R., S.S., and J.S.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. Email: juliane.stickel@med.uni-tuebingen.de.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1416389112/-/DCSupplemental.

E166–E175 | PNAS | Published online December 29, 2014 www.pnas.org/cgi/doi/10.1073/pnas.1416389112

Dow

nloa

ded

by g

uest

on

July

4, 2

020

Dow

nloa

ded

by g

uest

on

July

4, 2

020

Dow

nloa

ded

by g

uest

on

July

4, 2

020

Dow

nloa

ded

by g

uest

on

July

4, 2

020

Dow

nloa

ded

by g

uest

on

July

4, 2

020

individual heterogeneity with total HLA class I molecule countsranging from ∼42,500–288,500 molecules/cell on CLL cells and∼32,000–256,500 molecules per cell on normal B cells. Patient-individual analysis of HLA surface expression in triplicatesrevealed small, albeit significant differences in expression levels(P < 0.01) for 4 of 7 patients (Fig. 1A). HLA-DR expressionranged from ∼29,000–100,500 on CLL cells and ∼19,500–79,500on normal B cells. Minor differences in HLA-DR levels (P <0.01) were detected for 5 of 7 patients. Statistical analysis ofmean surface expression of HLA class I and II on CLL cellscompared with normal B cells showed no significant differences(Fig. 1 C and D).

Mass Spectrometry (LC-MS/MS) Identifies a Vast Array of NaturallyPresented HLA Class I and II Ligands. Mapping the HLA class Iligandomes of 30 CLL patients, we were able to identify a total of18,844 different peptides representing 7,377 source proteins(Table S1), attaining >95% of maximum attainable coverage(Fig. S1). The numbers of different peptides identified per pa-tient ranged from 345 to 2,497 (mean 1,131). Overall, peptidesrestricted by more than 30 different HLA-A and -B alleles(covering >99% of the Caucasian population; calculatedaccording to ref. 18) were identified in this study. In the healthyvolunteer (HV) cohort of 30 PBMC donors, a total of 17,322unique peptides representing 7,180 different source proteinswere identified (>90% coverage, Table S2). The HLA allotypedistribution in the HV cohort covered 100% of HLA-A and>80% of HLA-B alleles in the CLL patient cohort.Analysis of the HLA class II ligandomes was performed for 20

CLL patients. A total of 5,059 unique peptides representing1,486 source proteins was identified (Table S1). The HLA classII HV cohort of 13 PBMC donors yielded 2,046 different pep-tides representing 756 source proteins (Table S2).

Comparative Profiling of HLA Class I Ligandomes Reveals a DistinctSignature of CLL-Associated Antigens. To identify novel CLL-associated antigens, we compared the HLA ligand source pro-teomes of the CLL and HV cohorts. Overlaps of HLA sourceproteins revealed 2,148 proteins (29.1% of the mapped CLLsource proteome) to be exclusively represented in the HLA

ligandome of CLL (Fig. 2A). With the aim of designing a broadlyapplicable off-the-shelf peptide vaccine, we subsequently prior-itized the selection of potential targets according to the followingcriteria: CLL-exclusivity was defined as paramount criterion,followed by ranking of antigens according to frequency of rep-resentation in CLL ligandomes (Fig. 2B). Our platform high-lighted 49 source proteins (0.7% of the CLL source proteome)represented by 225 different HLA ligands showing CLL-exclu-sive representation in ≥20% of CLL patients (Table S3). Ap-plying the same antigen ranking strategy to HV PBMC exclusiveantigens, a set of 70 ligandome-defined benign tissue-associatedantigens (LiBAAs) and 273 corresponding ligands (LiBAPs)were identified (Table S4) for use as internal control in immuno-logical assays. To estimate the false discovery rate of LiTAA iden-tifications, randomized virtual ligandomes were simulated in silicoand the resultant numbers of false positive LiTAAs were calculatedand plotted for different frequencies of representation (Fig. 2D).For the implemented threshold of ≥20% CLL-exclusive antigenpresentation the false discovery rate was estimated to be 15.3%.To extrapolate the number of original LiTAAs yielded at in-

creased cohort sizes we randomly sampled size-reduced cohortsand defined the resultant LiTAAs (for each whole-numberedfraction). Nonlinear expression analysis locates the plateau valueof LiTAAs at 69.7 proteins (Fig. 2C).Apart from broadly represented CLL-LiTAAs suited for the

design of off-the-shelf vaccines, a second panel of 2,099 CLL-exclusive antigens with representation frequencies <20% wasidentified. These targets lend themselves as repositories for moreindividualized therapeutic approaches (19).

HLA Class I LiTAAs Are Derived from Diverse Pathways and Do NotExhibit Clear Unifying Characteristics Other Than Tumor-AssociatedPresentation. Functional annotation clustering of HLA class ILiTAAs with respect to their biological processes (GO term BPanalysis) using DAVID (20) yielded several small, slightly-enriched clusters of proteins involved in nucleotide metabolism,lymphocyte activation, cell migration, transcriptional regulationand apoptosis (Table S5). These clusters were found to be distinctfrom functional clusters found in HLA class I LiBAAs and HLAclass I common “housekeeping” source proteins. However, the

Fig. 1. HLA surface expression of primary CLL samples. HLA class I (A) and HLA class II (B) expression on CD5+CD19+ CLL cells compared with autologous CD5−

CD19+ B cells in 7 primary CLL samples. Data are expressed as mean ± SD of triplicate experiments. (C) Mean HLA class I and (D) HLA class II expression on CD5+

CD19+ CLL cells compared with autologous CD5−CD19+ B cells (n = 7). *P < 0.01. UPN, uniform patient number.

Kowalewski et al. PNAS | Published online December 29, 2014 | E167

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

July

4, 2

020

small sizes and low enrichment scores of these clusters do notindicate LiTAAs to be associated with specific, characteristic bi-ological processes. KEGG pathway analysis (21) of HLA class ILiTAAs did not identify any overrepresented pathways. To assesswhether LiTAAs show CLL-associated overexpression on themRNA level, we investigated a publicly available dataset (GEO

profiles, dataset GDS3902), which contains gene expression profilesof 12 CLL patient peripheral blood samples compared with pe-ripheral blood from 5 healthy volunteers (22). We found 11 of the49 LiTAAs to be overexpressed in these CLL patients, 5 of 49LiTAAs to be down-regulated, and 33 of 49 LiTAAs not to showany significant differences in gene expression (Table S6).

Fig. 2. Identification of a novel category of tumor-associated antigens by HLA ligandome profiling. (A) Overlap of HLA class I ligand source proteins ofprimary CLL samples (n = 30) and HV PBMC (n = 30). (B) Comparative profiling of HLA class I ligand source proteins based on the frequency of HLA restrictedrepresentation in CLL and HV PBMC ligandomes. Frequencies [%] of CLL patients/HVs positive for HLA restricted presentation of the respective source protein(x axis) are indicated on the y axis. The box on the left highlights the subset of source proteins showing CLL-exclusive representation in >20% of patients(LiTAAs, ligandome-defined tumor-associated antigens). (C) Regression analysis of the number of LiTAAs yielded at different cohort sizes. Reduced-size CLLand HV PBMC cohorts were randomly assembled from the complete respective cohorts (n = 30) and LiTAAs were defined as CLL-exclusive proteins presentedon ≥20% of CLL patients. The process of random cohort assembly and LiTAA definition was repeated 100 times and the resultant mean number of LiTAAs wasplotted for cohort sizes of n = 5, 10, 15, 20 and 25. Regression analysis using an exponential model extrapolated the plateau value for the number of LiTAAsresulting at high values of n to be 69.7. (D) Statistical analysis of the proportion of false positive LiTAA identifications at different representation frequencies.The numbers of original LiTAAs identified based on the analysis of the CLL and HV PBMC cohorts were compared with random virtual LiTAAs. Virtual CLLpatients and HV PBMCs were generated in silico based on random weighted sampling from the entirety of protein identifications in both original cohorts.These randomized virtual ligandomes of defined size (n = 1130 proteins, which is the mean number of protein identifications in all analyzed samples) wereused to define LiTAAs based on simulated cohorts of 30 CLL versus 30 HV PBMC. The process of protein randomization, cohort assembly and LiTAA iden-tification was repeated 1,000 times and the mean value of resultant virtual LiTAAs was calculated and plotted for the different threshold values. The cor-responding false discovery rates for any chosen LiTAA threshold are listed below the x axis. (E) Representation of published CLL-associated antigens in HLAclass I ligandomes. Bars indicate relative representation [%] of respective antigens by HLA class I ligands on CLL and HV PBMC. Dashed lines divide theantigens into three groups according to their degree of CLL-association (CLL exclusive, CLL overrepresented, not overrepresented).

E168 | www.pnas.org/cgi/doi/10.1073/pnas.1416389112 Kowalewski et al.

Dow

nloa

ded

by g

uest

on

July

4, 2

020

Detection of Naturally Presented HLA Class I Ligands Derived fromEstablished CLL-Associated Antigens. Alongside the identificationof novel CLL-associated antigens, a secondary approach focusedon the ranking of the few established CLL-antigens within ourdataset. We identified 28 different HLA ligands representing 8described CLL-associated antigens (Table S7). Notably, onlyfibromodulin (FMOD324–333, RINEFSISSF, HLA-B*15) showedCLL-exclusive representation, ranking at #437 of CLL antigensin our dataset, due to low frequency of representation in the CLLpatient cohort. The remaining 7 antigens showed representationboth on CLL and HV PBMC, thus failing to fulfill the criterionof CLL-exclusivity. However, for CD19, CD20, RHAMM andPRAME, CLL-associated overrepresentation of varying degreeswas detected (Fig. 2E).

Comparative Ligandome Profiling Identifies LiTAAs Shared AmongDifferent Disease Stages and Risk Strata. To assess the applica-bility of the novel targets across different stages of disease, weperformed subset-specific ligandome profiling comparingpatients in disease stages Binet A (n = 9), B (n = 7), and C (n =14). Overlap analysis of the 2,148 CLL-exclusive source proteinsrevealed that 550 (25.6%) of them were shared among at leasttwo stages, with a core group of 137 proteins (6.1%) representedin patients of all three disease stages (Fig. 3A). Of note, 45 of 49(91.8%) of LiTAAs (Table S3) belong to the core group ofshared source proteins represented in all three subsets. Heatmapanalysis of the representation frequencies of all 49 LiTAAsacross Binet stages A, B, and C is shown in Fig. 3B.Another focus was placed on determining the representation

of LiTAAs in the subsets of high-risk patients carrying the 17p13deletion (del17p, n = 5) compared with patients without thisgenetic aberration (no del17p, n = 25). We found 77.7% of theidentified LiTAAs to be represented in both subsets (Fig. 3C).Together, these data support the devised strategy of cohort-comprising analysis of HLA ligandomes for selection of broadlyapplicable targets.

Functional Characterization of HLA Class I LiTAPs Reveals CLL-Associated Immunoreactivity. To evaluate the immunogenicityand specificity of our HLA class I LiTAPs, we performed 12-drecall IFNγ ELISPOT assays. A panel of 15 LiTAPs (6 A*02,3 A*03, and 6 B*07 LiTAPs) was implemented for stimulation ofHLA-matched PBMC obtained from CLL patients and healthy

volunteers (Fig. 4A). We observed IFNγ secretion for 14 of 15(93.3%) of tested LiTAPs in CLL patients (Fig. 4 A and C andFig. S2 C and F), but not in healthy controls (0 of 10, Fig. 4B andFig. S2 B and E). These findings were confirmed exemplarily forPA*03

3 (DMXL11271–1279 SSSGLHPPK) by tetramer staining ofCD8+ T cells and intracellular cytokine staining for IFNγ andTNFα (Fig. S3 A and B). ELISPOT assays using HLA-matchedbenign tissue-derived LiBAPs were performed to control for theCLL-specificity of the observed LiTAP-directed immune recog-nition in CLL patients. We tested a panel of 9 LiBAPs (3 A*02, 3A*03, 3*B*07, Table S4) and observed no significant IFNγ se-cretion in any of the tested CLL patients [0 of 7 A*03 (Fig. 3D),0 of 10 A*02 and 0 of 5 B*07 (Fig. S2 A and D)].

The Frequencies of HLA Restricted Presentation and ImmuneRecognition Show a Direct Correlation. For the 14 of 15 LiTAPsshowing immune recognition in one or more patients, we cal-culated the allele-adjusted frequencies of HLA restricted pre-sentation (as detected by LC-MS/MS) and the frequencies ofimmunoreactivity (as detected by ELISPOT) in CLL patients.Strikingly, a linear correlation of these two parameters was ob-served (Pearson’s r = 0.75, R2 = 0.56, Fig. 4E). For six patients,we were able to perform both, HLA ligandome analysis andfunctional characterization in ELISPOT assays. Notably, allthree patients that were positive for presentation of specific LiTAPsin LC-MS/MS analysis also showed immune recognition of theseLiTAPs in ELISPOT assays. On the other hand, we did observeimmune reactions to LiTAPs in five patients, which were negativefor the respective peptides in ligandome analysis (Table S8).

HLA Class II Ligandome Analysis Identifies CD4+ T-Cell Epitopes forSynergistic Vaccine Design. Because of the important indirect anddirect roles of CD4+ T cells in anticancer immune responses(23), an optimal vaccine require the inclusion of HLA class IIepitopes. We performed overlap analysis of CLL and HV PBMCligandomes and identified 937 proteins (63.0% of the identifiedCLL source proteins) to be exclusively represented in theligandomes of CLL patients (Fig. 5A). Applying the sameantigen-ranking strategy as described for HLA class I, we iden-tified 73 HLA class II LiTAAs represented by 460 correspondingLiTAPs (Fig. 5B and Table S9). Functional characterization ofa panel of 7 HLA class II LiTAPs (Fig. 5C) in IFNγ ELISPOTassays revealed significant IFNγ secretion for 6 of 7 (85.7%)

Fig. 3. LiTAAs are robustly represented across different CLL disease stages and risk strata. (A) Source protein overlaps of CLL samples from different stages ofdisease [Binet A (n = 9), Binet B (n = 7), Binet C (n = 14)]. (B) Heatmap analysis of the representation frequencies [%] of LiTAAs across different disease stages(Binet A–C, as in A). (C) Heatmap analysis of LiTAA representation [%] on primary CLL samples with del17p (n = 5) and without del17p (n = 25). CLL, chroniclymphocytic leukemia; HV, healthy volunteer.

Kowalewski et al. PNAS | Published online December 29, 2014 | E169

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

July

4, 2

020

LiTAPs in CLL patients (Fig. 5E), but not in healthy controls (0of 10, Fig. 5D). Next, we performed combined analysis of HLAclass I and II ligandomes to identify shared, synergistic targets.Overlap analysis of CLL-exclusive source proteins revealed132 proteins to be represented both in HLA class I and IIligandomes (Fig. 5F). Heatmap analysis identified 2 proteinsdisplaying representation frequencies ≥20% in both ligandomes[B4GALT1 (26.7% class I/30.0% class II), HLA-DMA (20.0%class I/20% class II), Fig. 5G]. Strikingly, one of the class ILiTAPs (HLA-DMA206–214, HEIDRYTAI, B*18) was revealedto be completely embedded in the corresponding HLA class IILiTAP (VTHEIDRYTAIAY).

HLA Class I and Class II Ligand Source Proteins Show Distinct SubcellularDistribution Patterns. Bioinformatic analysis based on gene ontology(GO) term clustering implementing DAVID (20) revealed distinctclusters of cellular compartments for HLA ligand source proteinsrepresented in HLA class I and II ligandomes (Table S10). Themost distinct clusters identified were a large group of nuclear lumenderived source proteins presented on HLA class I and a highlyenriched cluster of vesicle-derived proteins presented on HLA classII. The clusters for HLA class I- and II-shared source proteinsgenerally showed lower levels of enrichment, with the most prom-inent cluster comprising proteins from the lysosome/endosomepathway, followed by membrane-bound and extracellular proteins,

Fig. 4. LiTAAs are specifically recognized by CLL patient immune responses. (A) HLA class I LiTAAs and corresponding LiTAPs (3 HLA-A*03, 5 HLA-A*02,6 HLA-B*07) functionally evaluated in IFNγ ELISPOT assays. Absolute numbers and frequencies of peptide-specific immune recognition by CLL patient PBMCare summarized in the right hand column. (B) Example of A*03 LiTAPs evaluated in ELISPOT using HV PBMC as a control. An EBV epitope mix containing fivefrequently recognized peptides [BRLF109–117 YVLDHLIVV (A*02), EBNA3471–479 RLRAEAQVK (A*03), EBNA3247–255 RPPIFIRRL (B*07), BZLF1190–197 RAKFKQLL(B*08), EBNA6162–171 AEGGVGWRHW (B*44)] was used as positive control, HIV GAG18–26 A*03 peptide KIRLRPGGK served as negative control. (C) Example ofELISPOT assays using HLA-A*03 LiTAPs (n = 3) on PBMC of three different CLL patients. Results are shown for immunoreactive LiTAPs. EBV epitope mix servedas positive control, HIV GAG18–26 A*03 peptide as negative control. (D) Example of HLA-A*03 benign tissue-derived LiBAPs (n = 3) tested on CLL patient PBMC asinternal control for the target selection strategy. EBV epitopemix served as positive control, HIV GAG18–26 A*03 peptide as negative control. (E) Scatterplot of the allele-adjusted frequencies of LiTAP presentation in CLL ligandomes (as detected by MS) and the corresponding allele-adjusted frequencies of immune recognition by CLLpatient PBMC in IFNγ ELISPOT. Data points are shown for the 14 of 15 LiTAPs showing immune recognition. HV, healthy volunteer; LiBAP, ligandome-defined benigntissue-associated peptide; LiTAP, ligandome-defined tumor-associated peptide; MS, mass spectrometry; neg., negative; pos., positive; UPN, uniform patient number.

E170 | www.pnas.org/cgi/doi/10.1073/pnas.1416389112 Kowalewski et al.

Dow

nloa

ded

by g

uest

on

July

4, 2

020

hinting at cross-presentation on HLA class I as an underlyingmechanism. Of note, for HLA class II, two clusters of cell-endogenous source proteins derived from the cytosolic ribosomeand the nuclear pore were identified, indicating underlyingmechanisms of autophagy and cross-presentation.

Longitudinal Analysis of CLL Patient Ligandomes Under DifferentTherapeutic Regimens. A rational setting to apply peptide-basedimmunotherapy is maintenance therapy and eradication of MRD.

As a consequence, peptide vaccination in CLL would mainly beperformed after standard chemo- or immunotherapy. Therefore,we analyzed HLA expression and performed ligandome profilingacross different time points of CLL patients undergoing differenttherapeutic regimens.We quantified HLA class I and II surface expression in 4

patients undergoing rituximab treatment (Rt0h, Rt24h) and in 1patient receiving alemtuzumab (At0h, At72h, At7d, Fig. S4 A–D).HLA surface expression showed patient-individual heterogeneity

Fig. 5. Identification of additional/synergistic HLA class II LiTAAs and LiTAPs. (A) Overlap of HLA class II ligand source proteins of primary CLL samples (n = 20)and HV PBMC (n = 13). (B) Comparative profiling of HLA class II ligand source proteins based on the frequency of HLA restricted representation in CLL and HVPBMC ligandomes. Frequencies [%] of CLL patients/HVs positive for HLA restricted presentation of the respective source protein (x axis) are indicated on the yaxis. The box on the left highlights the subset of source proteins showing CLL-exclusive representation in >20% of patients (LiTAAs, ligandome-definedtumor-associated antigens). (C) HLA class II LiTAAs and corresponding LiTAPs (n = 6) functionally evaluated in IFNγ ELISPOT assays. Absolute numbers andfrequencies of peptide-specific immune recognition by CLL patient PBMC are summarized in the right column. (D) Example of HLA class II LiTAPs evaluated inELISPOT using HV PBMC as a control. PHA was used as positive control. FLNA1669–1683 HLA-DR peptide (ETVITVDTKAAGKGK) served as negative control.(E) Example of ELISPOT assays using HLA class II LiTAPs (n = 6) on PBMC of three different CLL patients. Results are shown for immunoreactive LiTAPs. PHA wasused as positive control, FLNA1669–1683 HLA-DR peptide served as negative control. (F) Overlap analysis of CLL-exclusive HLA class I and HLA class II ligandsource proteins for shared/synergistic vaccine targets. (G) Heatmap analysis of the 132 shared HLA class I/II LiTAAs identified in D. The two source proteinsshowing representation in ≥20% of both, HLA class I and II CLL patient ligandomes are specified.

Kowalewski et al. PNAS | Published online December 29, 2014 | E171

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

July

4, 2

020

with no significant changes in mean HLA class I (Rt0h = 50,500,Rt24h = 48,000; At0h = 42,500, At7d = 61,500) and HLA class II(Rt0h = 36,500, Rt24h = 27,500; At0h = 47,000, At7d = 55,500)expression over the course of either therapeutic regimen.Longitudinal HLA class I ligandome profiling was performed

in single patients undergoing rituximab-bendamustin, alemtuzu-mab and ofatumumab treatment, respectively (Fig. 6 A–C).Differential presentation (defined as ≥twofold change with P ≤0.05 after Bonferroni adjustment for multiple testing) was ob-served for 0.73% (n = 14) of HLA class I ligands under rituximab-bendamustin treatment, for 7.4% (n = 182) of ligands underalemtuzumab treatment and for 6.5% (n = 98) of ligands underofatumumab treatment. Overall, 6 LiTAPs representing 6 of thetotal of 32 LiTAAs (18.8%) detectable in these three patientswere revealed to be differentially presented over the course oftherapy. Of note, 5 of 6 (83.3%) of these LiTAPs showed signif-icant up-regulation posttherapy.

Immune Responses Against LiTAPs Are Associated with ImprovedOverall Survival of CLL Patients. As a last step, we investigatedthe prognostic relevance of LiTAP-specific immune responses.We performed retrospective survival analysis of 45 CLL patientsanalyzed by ELISPOT assay (Table S11). We dichotomizedpatients into groups with T-cell responses specific for 0–1LiTAPs (n = 32) versus >1 LiTAP (n = 13) according to previousresults in RCC patients demonstrating significantly higher levels

of disease control in patients showing responses to multipleantigens (6). We found that 9 of 32 (28.1%) of patients in thelow-responding cohort, versus 0 of 13 of patients in the high-responding cohort had died. A strong indication for prolongedoverall survival in the high-responding cohort compared with thelow responders based on survival time from study enrollment (P <0.05, Fig. 7A) as well as survival time from diagnosis was observed(P = 0.0695, Fig. 7B). This analysis does not consider differencesin disease stage, cytogenetics or preceeding treatments.

DiscussionWith T-cell–based immunotherapy proclaimed as the scientificbreakthrough of 2013 (24), the research field finally revealed itsinherent potential for highly effective cancer therapy (25–27).Remaining challenges lie in guiding the specificity and increasingthe frequency of antitumor immune responses and expanding thespectrum of targetable entities (26). A rational and promisingapproach to achieve this goal is (multi) peptide vaccination (28,29). For this purpose the identification of immunologically rel-evant, tumor-associated antigens is indispensable and enabled bythe direct analysis of naturally presented HLA ligands. Here, wecomprehensively mapped the landscape of naturally presentedHLA ligands in primary CLL samples and assessed interpatientindividuality as well as treatment-induced changes in the HLAligandome composition. We implemented a new strategy, whichdefines a novel category of tumor-associated antigens strictly

Fig. 6. Longitudinal HLA class I ligandome analysis of CLL patients undergoing chemo- or immunotherapy. Volcano plots of the relative abundance of HLAligands in the class I ligandomes of patients after treatment compared with their respective abundance before therapy (ratio post- and pretherapy). Dashedlines indicate the thresholds for differential peptide presentation (defined as ≥twofold ratio with P < 0.05 after Bonferroni correction), with up-regulatedligands in the upper right and down-regulated ligands in the upper left areas. Frequencies and absolute numbers of differentially presented ligands arespecified in the respective quadrants. LiTAPs showing differential presentation over the course of therapy are marked in red and their sequences arespecified. (A) Analysis of a CLL patient ligandome before therapy, 48 h after treatment with rituximab (375 mg/m2), and 24 h after treatment withbendamustin (90 mg/m2). One of 25 (4.0%) of detectable LiTAPs showed differential presentation above the specified thresholds. (B) Analysis of a CLLpatient ligandome before therapy and after the first 7 d of treatment with alemtuzumab (three doses of alemtuzumab, 10 mg, 20 mg, and 30 mg on day 1, 3,and 5; ligandome analysis on day 7). Three of 24 (12.5%) of detectable LiTAPs showed differential presentation above threshold. (C) Analysis of a CLL patientligandome before therapy and 24h after treatment with 300 mg of atumumab. Two of 10 (20.0%) of detectable LiTAPs showed differential presentationabove threshold.

E172 | www.pnas.org/cgi/doi/10.1073/pnas.1416389112 Kowalewski et al.

Dow

nloa

ded

by g

uest

on

July

4, 2

020

based on their exclusive and frequent representation in CLLligandomes. This approach identified a panel of 49 HLA class ILiTAAs showing broad and frequent representation across dif-ferent Binet stages and mutational backgrounds as well as robustpresentation under chemo-/immunotherapy, thus representingsuitable candidates for broadly applicable off-the-shelf peptidevaccines, also after standard treatment of CLL. Reverse bio-informatics analysis of LiTAAs, such as functional annotationclustering and gene expression analysis did not lead to theidentification of comprehensive unifying characteristics whichwould enable the categorization or even prediction of LiTAAs.This result indicates the unique character of the HLA ligandomeand underscores the importance of defining T-cell antigens bydirect differential ligandome profiling. The relatively high falsediscovery rate (15.3%) at the implemented threshold for LiTAAdefinition is considered an acceptable tradeoff between assaystringency and discovery rates, as the profiling approach wasused as a screening step to identify candidates for subsequentimmunological validation.Recognition of tumor-associated peptides by the immune

system is a fundamental requirement for effective T-cell–basedimmunotherapy (30). Recent studies showing promising resultsfocused on mutated neoantigens, which, in theory, constitutetumor-specific and highly immunogenic targets (23, 31, 32).Here, we observed spontaneous T-cell responses targeting non-mutant epitopes in CLL patients. We validated these immuneresponses to be CLL-associated and directed strictly against CLLantigens (LiTAPs), thus indicating tumor-dependent priming ofT cells specific for nonmutated tumor epitopes in vivo in CLLpatients. This immunoreactivity against “self” in the context ofcancer has long been established as a component of anticancerimmune responses (33) and a range of nonmutant tumor-associated antigens (TAAs) have been described (3). However,recently it became evident that in solid tumors, only a smallfraction of the tumor antigens described so far - including mu-tated and nonmutated epitopes - are frequently and effectivelytargeted by T-cell responses (34), even in the context of immunecheckpoint modulation (35). By corollary, a large proportion ofthe specificities of tumor-reactive lymphocytes remain un-defined. Surprisingly, in the study at hand we fairly frequentlyobserved spontaneous T-cell responses targeting the novelantigen class of LiTAAs in leukemia patients. We also foundthat some of the established TAAs, when evaluated directlyand strictly on the HLA ligandome level, show lacking degreesof tumor association and frequencies of presentation. Similarobservations were made in our previous study of AML anti-gens (36). Together, these results suggest that direct, HLAligandome-centric approaches might be indispensable for theidentification of immunologically relevant, but otherwise in-conspicuous, tumor antigens and might help explain thelacking immunogenicity observed for some of the previouslyproposed TAAs.

Interestingly, for immunoreactive LiTAPs the direct correla-tion between frequencies of epitope detection in CLL ligandomesand frequencies of LiTAP-specific immune recognition by CLLpatient PBMC might indicate that antigen presentation on cancercells is the major prerequisite/limiting factor for the priming ofLiTAP-specific immune responses. The, presence of LiTAP-spe-cific T cells in patients could be explained by the absence ofnegative thymic selection for these antigens due to tumor-specificalterations in antigen presentation not paralleled in the thymus.Further investigations aiming to elucidate the naïve T-cell reper-toire including T-cell receptor affinities and the cellular processesleading to the exclusive appearance of antigen on cancer cells inLiTAP-specific immune responses are ongoing.As CD4+ T cells play important indirect and direct roles in

antitumor immunity (23, 37–42), we further applied our profilingapproach to HLA class II ligandomes, identifying a panel of 73additional class II LiTAAs. Notably, one of the HLA class IILiTAPs was found to contain a complete, embedded HLA class ILiTAP, which points to a striking new option for synergisticvaccine design as discussed in recent publications (43, 44). Suchnaturally presented embedded HLA ligands might representoptimal vaccine candidates that are recognized by both, CD4+

and CD8+ T cells. Interestingly, the clusters of proteins pre-sented on both HLA class I and II were found to be derived fromcellular compartments typically associated with the HLA class IIpresentation, thus indicating cross-presentation on HLA class Ias the mechanistic basis of this dual representation.To investigate the clinical relevance of anti-LiTAP immune

responses in CLL patients, we performed retrospective survivalanalysis and observed a survival benefit for patients showingimmune reactions to multiple of the tested LiTAPs. This resultindicates the involvement of these immune responses in diseasecontrol in CLL patients and directly implies LiTAAs as patho-physiologically relevant tumor antigens in leukemia. In conse-quence, these malignancies might be targeted effectively byimmunotherapy despite their reported low number of somaticmutations (45), which are currently viewed as the most highlyactive targets for T-cell–based immunotherapy. The LiTAPs wefind on tumor cells based on tumor-specific antigen processingmodalities may well be of similar immunogenicity as mutatedantigens, if our hypothesis of the absence of such processingmodalities in the thymus is correct.In conclusion, the HLA ligandome-centric approach might

prove an invaluable tool for antigen discovery and re-evaluationas well as for patient stratification. In CLL, it enabled theidentification of a panel of novel immunogenic nonmutatedepitopes amenable for clinical application and provided a newperspective on the immunogenicity of this malignancy.

Materials and MethodsPatients and Blood Samples. For HLA ligandome analysis, PBMC from CLLpatients (>80% CLL cell frequency) as well as PBMC from healthy volunteers

Fig. 7. Retrospective survival analysis of CLL patients (n = 45) with respect to their immune recognition of LiTAPs. Overall survival of subjects evaluated forLiTAP-specific immune responses grouped as follows: black, CLL patients showing immune responses to >1 LiTAPs (n = 13). Red, CLL patients showing immunereactions to 0–1 LiTAPs (n = 32). (A) Follow-up of patient survival from time of study enrollment. (B) Follow-up of patient survival from time of diagnosis.

Kowalewski et al. PNAS | Published online December 29, 2014 | E173

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

July

4, 2

020

(HVs) were isolated by density gradient centrifugation. Informed consentwas obtained in accordance with the Helsinki protocol. This study was per-formed according to the guidelines of the local ethics committee (373/2011BO2). Patient characteristics are provided in Table S1. HLA typing wascarried out by the Department of Hematology and Oncology, University ofTübingen. Samples were stored at −80 °C until further use.

Quantification of HLA Surface Expression. To allow for comparison with healthyautologous B lymphocytes, quantification of HLA surface expression wasperformed in patient samples containing at least 0.5% CD5−CD19+ normalB cells. HLA surface expression was analyzed using the QIFIKIT quantitativeflow cytometric assay (Dako) according to manufacturer’s instructions. Inbrief, triplicates of each sample were stained with the pan-HLA class I specificmonoclonal antibody (mAb) W6/32, HLA-DR specific mAb L243 (both pro-duced in house) or IgG isotype control (BioLegend), respectively. Surfacemarker staining was carried out with directly labeled CD3 (BD), CD5 (BD), andCD19 (BD) antibodies. 7-AAD (BioLegend) was added as viability marker im-mediately before flow cytometric analysis on a FACSCanto Analyzer (BD).

Isolation of HLA Ligands from Primary Samples. HLA class I and II moleculeswere isolated using standard immunoaffinity purification as described (46, 47)using the pan-HLA class I specific mAb W6/32 and the pan-HLA class II specificmAb Tü39, respectively.

Analysis of HLA Ligands by LC-MS/MS. Peptide samples were separatedby reversed-phase liquid chromatography (nanoUHPLC, UltiMate 3000RSLCnano, Dionex) and subsequently analyzed in an on-line coupled LTQOrbitrap XL hybrid mass spectrometer (ThermoFisher). Samples were ana-lyzed in five technical replicates. Sample volumes of 5 μL (sample shares of20%) were injected onto a 75 μm × 2 cm trapping column (Acclaim PepMapRSLC, Dionex) at 4 μL/min for 5.75 min. Peptide separation was subsequentlyperformed at 50 °C and a flow rate of 175 nL/min on a 50 μm × 50 cmseparation column (Acclaim PepMap RSLC, Dionex) applying a gradientranging from 2.4 to 32.0% of ACN over the course of 140 min. Elutingpeptides were ionized by nanospray ionization and analyzed in the massspectrometer implementing a top 5 CID (collision induced dissociation)method generating fragment spectra for the 5 most abundant precursorions in the survey scans. Resolution was set to 60,000. For HLA class I ligands,the mass range was limited to 400–650 m/z with charge states 2+ and 3+selected for fragmentation. For HLA class II, a mass range of 300–1,500 m/zwas analyzed with charge states ≥2 selected for fragmentation.

Database Search and Spectral Annotation. For data processing, the softwareProteome Discoverer (v1.3, ThermoFisher) was used to integrate the searchresults of the Mascot search engine (Mascot 2.2.04, Matrix Science) againstthe human proteome as comprised in the Swiss-Prot database (www.uniprot.org, release: September 27th, 2013; 20,279 reviewed proteinsequences contained). The search combined data of technical replicates andwas not restricted by enzymatic specificity. Precursor mass tolerance was setto 5 ppm, and fragment mass tolerance was set to 0.5 Da. Oxidized methi-onine was allowed as a dynamic modification. False discovery rates (FDR)were determined by the Percolator algorithm (48) based on processingagainst a decoy database consisting of the shuffled target database. FDRwas set at a target value of q ≤ 0.05 (5% FDR). Peptide-spectrum matches(PSMs) with B ≤ 0.05 were filtered according to additional, orthogonalparameters, to ensure spectral quality and validity. Mascot scores were fil-tered to ≥20. For HLA class I, peptide lengths were limited to 8–12 aa oflength. For HLA class II, peptides were limited to 12–25 aa of length. Proteingrouping was disabled, allowing for multiple annotations of peptides (e.g.,conserved sequences mapping into multiple proteins). For quality control,yield thresholds of ≥300 unique HLA class I ligands and ≥100 unique HLAclass II ligands per sample were applied. HLA annotation was performedusing SYFPEITHI (www.syfpeithi.de) or an extended in-house database.

Longitudinal Analysis of CLL Patient Ligandomes Over the Course of Therapy.For label-free quantification (LFQ) of the relative HLA ligand abundancesover the course of therapy, the injected peptide amounts of paired sampleswere normalized and LC-MS/MS analysis was performed in five technicalreplicates for each sample.

In brief, relative amounts of substance of paired samples were calculatedfrom average precursor ion intensities determined in dose-finding massspectrometry runs and adjusted accordingly by dilution. Relative quantifica-tion of HLA ligands was performed by calculating the area under the curve ofthe corresponding precursor extracted ion chromatograms (XIC) using Pro-teomeDiscoverer 1.3. The ratios of themean areas of the individual peptides inthe five LFQ-MS runs of each sample were calculated and two-tailed t testswere performed using an in-house Matlab script (v8.2, Mathworks).

Peptide Synthesis. The automated peptide synthesizer EPS221 (Abimed) wasused to synthesize peptides using the 9-fluorenylmethyl-oxycarbonyl/tert-butyl (Fmoc/tBu) strategy as described (49). Synthetic peptides were used forvalidation of LC-MS/MS identifications as well as for functional experiments.

Amplification of Peptide-Specific T Cells. PBMC from CLL patients and HVswere cultured in IMDM medium (Gibco) supplemented with 10% humanserum (PHS), 100 mM β-mercaptoethanol (Roth) and 1% penicillin/streptomycin(PAA). For CD8+ T-cell stimulation, PBMC were thawed and pulsedwith 1 μg/mLper peptide. Peptide-pulsed PBMC (5-6 × 106 cells per mL) were culturedat 37 °C and 5% CO2 for 12 d. On day 0 and day 1, 5 ng/mL IL-4 (R&D Systems)and 5 ng/mL IL-7 (Promokine) were added to the culture medium. On days 3,5, 7, and 9, 2 ng/mL IL-2 (R&D Systems) were added to the culture medium.Peptide-stimulated PBMC/CD8+ T cells were characterized by ELISPOT assaysor tetramer staining on day 12 and by intracellular cytokine staining on day13, respectively. For tetramer staining, CD8+ T cells were enriched by positiveselection using magnetic cell sorting (Miltenyi Biotec) and stimulated asdescribed above. For CD4+ T-cell stimulation, culture was performed as de-scribed for CD8+ T cells with 2 modifications: Pulsing was carried out with10 μg/mL of HLA class II peptide, and no IL-4 and IL-7 was added.

IFNγ ELISPOT Assay. IFNγ ELISPOT assays were carried out as described (50). Inbrief, 96-well nitrocellulose plates (Millipore) were coated with 1 mg/mLIFNγ mAb (Mabtech) and incubated over night at 4 °C. Plates were blockedwith 10% (vol/vol) human serum for 2 h at 37 °C. 5 × 105 cells per well ofprestimulated PBMC were pulsed with 1 μg/mL (HLA class I) or 2.5 μg/mL(HLA class II) peptide and incubated for 24–26 h. Readout was performedaccording to manufacturer’s instructions. Spots were counted using anImmunoSpot S5 analyzer (CTL). T-cell responses were considered to bepositive when >10 spots per well were counted and the mean spot countper well was at least threefold higher than the mean number of spots inthe negative control wells [according to the cancer immunoguidingprogram (CIP) guidelines (51)].

Intracellular Cytokine and Tetramer Staining. The frequency and functionalityof peptide-specific CD8+ T cells was analyzed by intracellular IFNγ and TNFαstaining as described (50, 52). PBMC were pulsed with 10 μg/mL of individualpeptide and incubated in the presence of 10 μg/mL Brefeldin A (SigmaAldrich) and 10 μg/mL GolgiStop (BD Bioscience) for 6–8 h. Cells were labeledusing Cytofix/Cytoperm (BD), CD8-PECy7 (Beckman Coulter), CD4-APC (BDBioscience), TNFα-PE (Beckman Coulter), and IFNγ-FITC (BD). Samples wereanalyzed on a FACS Canto II.

The frequency of peptide-specific CD8+ T cells was determined by stainingwith anti-CD8 and HLA:peptide-tetramer-PE as described (53).

Software and Statistical Analysis. Flow cytometric data analysis was per-formed using FlowJo 7.2 (Treestar). Heatmap analysis of antigen frequenciesas well as volcano plot analysis of longitudinal HLA ligandome data wereperformed using Matlab (v8.2, Mathworks). In-house R and Python scriptswere used for the generation of virtual ligandomes and definition of virtualLiTAAs in the analysis of LiTAA false discovery rates and for the LiTAA-plateauregression analysis. GraphPad Prism 6.0 (GraphPad Software) was used forstatistical analysis. Statistical analysis of HLA surface expression was based onunpaired t tests. Comparative survival analysis was performed based ona log-rank (Mantel-Cox) test.

ACKNOWLEDGMENTS. We thank Patricia Hrstic, Beate Pömmerl, ClaudiaFalkenburger, Katharina Graf, and Nicole Zuschke for excellent technicalsupport. This work was supported by the Deutsche Forschungsgesellschaft(DFG, SFB 685), the Fortüne Program of the University of Tübingen (Fortüne2032-0-0), and the German Cancer Consortium (DKTK).

1. Rosenberg SA, Yang JC, Restifo NP (2004) Cancer immunotherapy: Moving beyondcurrent vaccines. Nat Med 10(9):909–915.

2. Mellman I, Coukos G, Dranoff G (2011) Cancer immunotherapy comes of age. Nature480(7378):480–489.

3. Vigneron N, Stroobant V, Van den Eynde BJ, van der Bruggen P (2013) Database of Tcell-defined human tumor antigens: The 2013 update. Cancer Immun 13:15.

4. Schwartzentruber DJ, et al. (2011) gp100 peptide vaccine and interleukin-2 in patientswith advanced melanoma. N Engl J Med 364(22):2119–2127.

E174 | www.pnas.org/cgi/doi/10.1073/pnas.1416389112 Kowalewski et al.

Dow

nloa

ded

by g

uest

on

July

4, 2

020

5. Mittendorf EA, et al. (2012) Clinical trial results of the HER-2/neu (E75) vaccine toprevent breast cancer recurrence in high-risk patients: From US Military Cancer In-stitute Clinical Trials Group Study I-01 and I-02. Cancer 118(10):2594–2602.

6. Walter S, et al. (2012) Multipeptide immune response to cancer vaccine IMA901 aftersingle-dose cyclophosphamide associates with longer patient survival. Nat Med 18(8):1254–1261.

7. Morgan RA, et al. (2013) Cancer regression and neurological toxicity following anti-MAGE-A3 TCR gene therapy. J Immunother 36(2):133–151.

8. Ritgen M, et al. (2004) Graft-versus-leukemia activity may overcome therapeutic re-sistance of chronic lymphocytic leukemia with unmutated immunoglobulin variableheavy-chain gene status: Implications of minimal residual disease measurement withquantitative PCR. Blood 104(8):2600–2602.

9. Gribben JG, et al. (2005) Autologous and allogeneic stem cell transplantations forpoor-risk chronic lymphocytic leukemia. Blood 106(13):4389–4396.

10. Russell NH, et al. (2005) Donor lymphocyte infusions can result in sustained remissionsin patients with residual or relapsed lymphoid malignancy following allogeneichaemopoietic stem cell transplantation. Bone Marrow Transplant 36(5):437–441.

11. Hallek M (2013) Signaling the end of chronic lymphocytic leukemia: New frontlinetreatment strategies. Blood 122(23):3723–3734.

12. Burkhardt UE, et al. (2013) Autologous CLL cell vaccination early after transplantinduces leukemia-specific T cells. J Clin Invest 123(9):3756–3765.

13. Giannopoulos K, et al. (2010) Peptide vaccination elicits leukemia-associated antigen-specific cytotoxic CD8+ T-cell responses in patients with chronic lymphocytic leuke-mia. Leukemia 24(4):798–805.

14. Giannopoulos K, Schmitt M (2006) Targets and strategies for T-cell based vaccines inpatients with B-cell chronic lymphocytic leukemia. Leuk Lymphoma 47(10):2028–2036.

15. Giannopoulos K, et al. (2006) Expression of RHAMM/CD168 and other tumor-associ-ated antigens in patients with B-cell chronic lymphocytic leukemia. Int J Oncol 29(1):95–103.

16. Cobbold M, et al. (2013) MHC class I-associated phosphopeptides are the targets ofmemory-like immunity in leukemia. Sci Transl Med 5(203):ra125.

17. Campoli M, Ferrone S (2008) HLA antigen changes in malignant cells: Epigeneticmechanisms and biologic significance. Oncogene 27(45):5869–5885.

18. Schipper RF, van Els CA, D’Amaro J, Oudshoorn M (1996) Minimal phenotype panels.A method for achieving maximum population coverage with a minimum of HLAantigens. Hum Immunol 51(2):95–98.

19. Rammensee HG, Weinschenk T, Gouttefangeas C, Stevanovi�c S (2002) Towardspatient-specific tumor antigen selection for vaccination. Immunol Rev 188:164–176.

20. Huang W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis oflarge gene lists using DAVID bioinformatics resources. Nat Protoc 4(1):44–57.

21. Kanehisa M, Goto S (2000) KEGG: Kyoto encyclopedia of genes and genomes. NucleicAcids Res 28(1):27–30.

22. Vargova K, et al. (2011) MYB transcriptionally regulates the miR-155 host gene inchronic lymphocytic leukemia. Blood 117(14):3816–3825.

23. Tran E, et al. (2014) Cancer immunotherapy based on mutation-specific CD4+ T cells ina patient with epithelial cancer. Science 344(6184):641–645.

24. Couzin-Frankel J (2013) Breakthrough of the year 2013. Cancer immunotherapy.Science 342(6165):1432–1433.

25. Wolchok JD, et al. (2013) Nivolumab plus ipilimumab in advanced melanoma. N Engl JMed 369(2):122–133.

26. Topalian SL, et al. (2012) Safety, activity, and immune correlates of anti-PD-1 antibodyin cancer. N Engl J Med 366(26):2443–2454.

27. Hodi FS, et al. (2010) Improved survival with ipilimumab in patients with metastaticmelanoma. N Engl J Med 363(8):711–723.

28. Riley JL (2013) Combination checkpoint blockade—taking melanoma immunotherapyto the next level. N Engl J Med 369(2):187–189.

29. Perez SA, et al. (2010) A new era in anticancer peptide vaccines. Cancer 116(9):2071–2080.

30. Cheever MA, et al. (2009) The prioritization of cancer antigens: A national cancerinstitute pilot project for the acceleration of translational research. Clin Cancer Res15(17):5323–5337.

31. van Rooij N, et al. (2013) Tumor exome analysis reveals neoantigen-specific T-cellreactivity in an ipilimumab-responsive melanoma. J Clin Oncol 31(32):e439–e442.

32. Robbins PF, et al. (2013) Mining exomic sequencing data to identify mutated antigensrecognized by adoptively transferred tumor-reactive T cells. Nat Med 19(6):747–752.

33. van der Bruggen P, et al. (1991) A gene encoding an antigen recognized by cytolytic Tlymphocytes on a human melanoma. Science 254(5038):1643–1647.

34. Andersen RS, et al. (2012) Dissection of T-cell antigen specificity in human melanoma.Cancer Res 72(7):1642–1650.

35. Kvistborg P, et al. (2012) TIL therapy broadens the tumor-reactive CD8(+) T cellcompartment in melanoma patients. OncoImmunology 1(4):409–418.

36. Berlin C, et al. (2014) Mapping the HLA ligandome landscape of acute myeloid leu-kemia: A targeted approach toward peptide-based immunotherapy. Leukemia,10.1038/leu.2014.233.

37. Schoenberger SP, Toes RE, van der Voort EI, Offringa R, Melief CJ (1998) T-cell help forcytotoxic T lymphocytes is mediated by CD40-CD40L interactions. Nature 393(6684):480–483.

38. Janssen EM, et al. (2003) CD4+ T cells are required for secondary expansion andmemory in CD8+ T lymphocytes. Nature 421(6925):852–856.

39. Mumberg D, et al. (1999) CD4(+) T cells eliminate MHC class II-negative cancer cells invivo by indirect effects of IFN-gamma. Proc Natl Acad Sci USA 96(15):8633–8638.

40. Greiner J, et al. (2012) Mutated regions of nucleophosmin 1 elicit both CD4(+) andCD8(+) T-cell responses in patients with acute myeloid leukemia. Blood 120(6):1282–1289.

41. Perez-Diez A, et al. (2007) CD4 cells can be more efficient at tumor rejection than CD8cells. Blood 109(12):5346–5354.

42. Sun JC, Williams MA, Bevan MJ (2004) CD4+ T cells are required for the maintenance,not programming, of memory CD8+ T cells after acute infection. Nat Immunol 5(9):927–933.

43. Tomita Y, et al. (2013) Identification of promiscuous KIF20A long peptides bearingboth CD4+ and CD8+ T-cell epitopes: KIF20A-specific CD4+ T-cell immunity in patientswith malignant tumor. Clin Cancer Res 19(16):4508–4520.

44. Melief CJ (2013) “License to kill” reflects joint action of CD4 and CD8 T cells. ClinCancer Res 19(16):4295–4296.

45. Vogelstein B, et al. (2013) Cancer genome landscapes. Science 339(6127):1546–1558.46. Falk K, Rötzschke O, Stevanovi�c S, Jung G, Rammensee HG (1991) Allele-specific motifs

revealed by sequencing of self-peptides eluted from MHC molecules. Nature351(6324):290–296.

47. Kowalewski DJ, Stevanovi�c S (2013) Biochemical large-scale identification of MHCclass I ligands. Methods Mol Biol 960:145–157.

48. Käll L, Canterbury JD, Weston J, Noble WS, MacCoss MJ (2007) Semi-supervisedlearning for peptide identification from shotgun proteomics datasets. Nat Methods4(11):923–925.

49. Sturm T, et al. (2013) Mouse urinary peptides provide a molecular basis for genotypediscrimination by nasal sensory neurons. Nat Commun 4:1616.

50. Widenmeyer M, et al. (2012) Promiscuous survivin peptide induces robust CD4+ T-cellresponses in the majority of vaccinated cancer patients. Int J Cancer 131(1):140–149.

51. Britten CM, et al. (2008) The CIMT-monitoring panel: A two-step approach to har-monize the enumeration of antigen-specific CD8+ T lymphocytes by structural andfunctional assays. Cancer Immunol Immunother 57(3):289–302.

52. Neumann A, et al. (2013) Identification of HLA ligands and T-cell epitopes for im-munotherapy of lung cancer. Cancer Immunol Immunother 62(9):1485–1497.

53. Rudolf D, et al. (2008) Potent costimulation of human CD8 T cells by anti-4-1BB and anti-CD28 on synthetic artificial antigen presenting cells. Cancer Immunol Immunother 57(2):175–183.

Kowalewski et al. PNAS | Published online December 29, 2014 | E175

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

July

4, 2

020

Correction

IMMUNOLOGY AND INFLAMMATIONCorrection for “HLA ligandome analysis identifies the un-derlying specificities of spontaneous antileukemia immune re-sponses in chronic lymphocytic leukemia (CLL),” by DanielJ. Kowalewski, Heiko Schuster, Linus Backert, Claudia Berlin,Stefan Kahn, Lothar Kanz, Helmut R. Salih, Hans-GeorgRammensee, Stefan Stevanovic, and Juliane Sarah Stickel, whichappeared in issue 2, January 13, 2015, of Proc Natl Acad Sci USA(112:E166–E175; first published December 29, 2014; 10.1073/pnas.1416389112).The authors note that Figs. 4 and 5 appeared incorrectly. The

errors pertain only to the depiction of the exemplary ELISPOTwells and the UPNs and do not affect the quantitative resultsdisplayed in the bar graphs nor any of the results stated anddiscussed in the paper. The corrected figures and their legendsappear below.

E6254–E6256 | PNAS | November 10, 2015 | vol. 112 | no. 45 www.pnas.org

Fig. 4. LiTAAs are specifically recognized by CLL patient immune responses. (A) HLA class I LiTAAs and corresponding LiTAPs (3 HLA-A*03, 5 HLA-A*02,6 HLA-B*07) functionally evaluated in IFNγ ELISPOT assays. Absolute numbers and frequencies of peptide-specific immune recognition by CLL patient PBMCare summarized in the right hand column. (B) Example of A*03 LiTAPs evaluated in ELISPOT using HV PBMC as a control. An EBV epitope mix containing fivefrequently recognized peptides [BRLF109–117 YVLDHLIVV (A*02), EBNA3471–479 RLRAEAQVK (A*03), EBNA3247–255 RPPIFIRRL (B*07), BZLF1190–197 RAKFKQLL(B*08), EBNA6162–171 AEGGVGWRHW (B*44)] was used as positive control, HIV GAG18–26 A*03 peptide KIRLRPGGK served as negative control. (C) Example ofELISPOT assays using HLA-A*03 LiTAPs (n = 3) on PBMC of three different CLL patients. Results are shown for immunoreactive LiTAPs. EBV epitope mix servedas positive control, HIV GAG18–26 A*03 peptide as negative control. (D) Example of HLA-A*03 benign tissue-derived LiBAPs (n = 3) tested on CLL patient PBMCas internal control for the target selection strategy. EBV epitope mix served as positive control, HIV GAG18–26 A*03 peptide as negative control. (E) Scatterplotof the allele-adjusted frequencies of LiTAP presentation in CLL ligandomes (as detected by MS) and the corresponding allele-adjusted frequencies of immunerecognition by CLL patient PBMC in IFNγ ELISPOT. Data points are shown for the 14 of 15 LiTAPs showing immune recognition. HV, healthy volunteer; LiBAP,ligandome-defined benign tissue-associated peptide; LiTAP, ligandome-defined tumor-associated peptide; MS, mass spectrometry; neg., negative; pos.,positive; UPN, uniform patient number.

PNAS | November 10, 2015 | vol. 112 | no. 45 | E6255

CORR

ECTION

www.pnas.org/cgi/doi/10.1073/pnas.1519135112

Fig. 5. Identification of additional/synergistic HLA class II LiTAAs and LiTAPs. (A) Overlap of HLA class II ligand source proteins of primary CLL samples (n = 20) and HVPBMC (n = 13). (B) Comparative profiling of HLA class II ligand source proteins based on the frequency of HLA restricted representation in CLL and HV PBMCligandomes. Frequencies [%] of CLL patients/HVs positive for HLA restricted presentation of the respective source protein (x axis) are indicated on the y axis. The box onthe left highlights the subset of source proteins showing CLL-exclusive representation in >20% of patients (LiTAAs, ligandome-defined tumor-associated antigens).(C) HLA class II LiTAAs and corresponding LiTAPs (n = 6) functionally evaluated in IFNγ ELISPOT assays. Absolute numbers and frequencies of peptide-specific immunerecognition by CLL patient PBMC are summarized in the right column. (D) Example of HLA class II LiTAPs evaluated in ELISPOT using HV PBMC as a control. PHA wasused as positive control. FLNA1669–1683 HLA-DR peptide (ETVITVDTKAAGKGK) served as negative control. (E) Example of ELISPOT assays using HLA class II LiTAPs (n = 6)on PBMC of three different CLL patients. Results are shown for immunoreactive LiTAPs. PHA was used as positive control, FLNA1669–1683 HLA-DR peptide served asnegative control. (F) Overlap analysis of CLL-exclusive HLA class I and HLA class II ligand source proteins for shared/synergistic vaccine targets. (G) Heatmap analysis of the132 shared HLA class I/II LiTAAs identified inD. The two source proteins showing representation in ≥20% of both, HLA class I and II CLL patient ligandomes are specified.

E6256 | www.pnas.org