Proteogenomics Uncovers a Vast Repertoire of SharedTumor-Specific Antigens in Ovarian Cancer A C
Qingchuan Zhao1,2, Jean-Philippe Laverdure1, Jo€el Lanoix1, Chantal Durette1, Caroline Cot�e1, �Eric Bonneil1,C�eline M. Laumont1,2, Patrick Gendron1, Krystel Vincent1, Mathieu Courcelles1, S�ebastien Lemieux1,3,Douglas G. Millar4, Pamela S. Ohashi4,5, Pierre Thibault1,6, and Claude Perreault1,2
ABSTRACT◥
High-grade serous ovarian cancer (HGSC), the principalcause of death from gynecologic malignancies in the world, hasnot significantly benefited from advances in cancer immuno-therapy. Although HGSC infiltration by lymphocytes correlateswith superior survival, the nature of antigens that can elicit anti-HGSC immune responses is unknown. The goal of this studywas to establish the global landscape of HGSC tumor-specificantigens (TSA) using a mass spectrometry pipeline that inter-rogated all reading frames of all genomic regions. In 23 HGSCtumors, we identified 103 TSAs. Classic TSA discovery appro-aches focusing only on mutated exonic sequences would haveuncovered only three of these TSAs. Other mutated TSAsresulted from out-of-frame exonic translation (n ¼ 2) or fromnoncoding sequences (n ¼ 7). One group of TSAs (n ¼ 91)
derived from aberrantly expressed unmutated genomic sequen-ces, which were not expressed in normal tissues. These aberrant-ly expressed TSAs (aeTSA) originated primarily from nonexonicsequences, in particular intronic (29%) and intergenic (22%)sequences. Their expression was regulated at the transcriptionallevel by variations in gene copy number and DNA methylation.Although mutated TSAs were unique to individual tumors,aeTSAs were shared by a large proportion of HGSCs. Takinginto account the frequency of aeTSA expression and HLA allelefrequencies, we calculated that, in Caucasians, the mediannumber of aeTSAs per tumor would be five. We conclude that,in view of their number and the fact that they are shared bymany tumors, aeTSAs may be the most attractive targets forHGSC immunotherapy.
IntroductionOvarian cancer is the principal cause of death from gynecologic
malignancies in the world and is responsible for over 14,000 deaths peryear in the United States (1). High-grade serous ovarian cancer(HGSC) accounts for 70% to 80% of these deaths, and overall survivalhas not changed significantly for several decades (2). The positivecorrelation between the abundance of tumor-infiltrating lymphocytes(TIL) and increased overall survival hints that T cells can recognizebiologically relevant tumor antigens in HGSC (3, 4). Strong evidencesuggests that HGSC TILs adjacent to tumor epithelial cells are actively
engaged in local immune editing. Indeed, in a multimodality study of212 HGSC samples from 38 patients, CD8þ TILs negatively associatewith malignant cell diversity (5). In line with this, and given thetherapeutic efficacy of immune-checkpoint inhibitors in several tumortypes, clinical trials with one or more checkpoint inhibitors arecurrently under way in HGSC. However, initial trials with anti–PD-1 have shown limited activity in HGSC (6, 7). In view of this, there is apressing need to identify the antigens that can elicit therapeutic anti-HGSC immune responses (3, 8). Such antigens could be used asvaccines (� immune-checkpoint inhibitors) or as targets for T-cellreceptor–based approaches (cell therapy, bispecific biologics; ref. 9).
Rapid progress in the field and lack of a standardized nomenclaturecommonly lead to some confusion in the classification of tumorantigens. For the sake of clarity, we therefore used throughout thisarticle a proposed classification of tumor antigens into three discretecategories: tumor-associated antigens (TAA), mutated tumor-specificantigens (mTSA), and aberrantly expressed tumor-specific antigens(aeTSA; ref. 10). TAAs are MHC-associated peptides (MAP) thatshow superior abundance on tumor cells but are nonetheless presenton normal cells and, therefore, may induce central immunetolerance (11–13). mTSAs derive from mutated DNA sequences thatcan be either exonic or nonexonic (10, 14). aeTSAs result fromaberrant expression of unmutated transcripts that are not expressedin any normal somatic cell, including medullary thymic epithelial cells(mTEC), which orchestrate central immune tolerance. Expression ofaeTSAs results from cancer-specific epigenetic changes. In one study,aeTSAs are found to represent the vast majority of TSAs in acutelymphoblastic leukemia and lung cancer (15). Finally, a peculiarantigen family, cancer-germline antigens (CGA), is astride the TAAand aeTSA categories. CGAs are coded by canonical exons normallyexpressed only by germ cells, and their aberrant expression in cancercells is driven mainly by epigenetic alterations. However, some CGAs
1Institute for Research in Immunology and Cancer, Universit�e de Montr�eal,Montreal, Quebec, Canada. 2Department of Medicine, Universit�e de Montr�eal,Montreal, Quebec, Canada. 3Department of Computer Science and OperationsResearch, Universit�e de Montr�eal, Montreal, Quebec, Canada. 4Campbell FamilyInstitute for Breast Cancer Research, Princess Margaret Cancer Centre, Toronto,Ontario, Canada. 5Department of Medical Biophysics and Immunology, Univer-sity of Toronto, Toronto, Ontario, Canada. 6Department of Chemistry, Universit�ede Montr�eal, Montreal, Quebec, Canada.
Note: Supplementary data for this article are available at Cancer ImmunologyResearch Online (http://cancerimmunolres.aacrjournals.org/).
CorrespondingAuthors:ClaudePerreault, Institute for Research in Immunologyand Cancer, Universit�e de Montr�eal, PO Box 6128, Downtown Station, Montr�eal,Qu�ebec H3C 3J7, Canada. Phone: 514-343-6126; Fax: 514-343-5839; E-mail:[email protected]; and Pierre Thibault, Institute for Research inImmunology and Cancer, PO Box 6128, Downtown Station, Montr�eal, Qu�ebecH3C 3J7, Canada. Phone: 514-343-6910; Fax: 514-343-6843; E-mail:[email protected]
are expressed by adult mTECs (16). Accordingly, CGAs expressed inmTECs (or other somatic tissues) are considered as TAAs, and thosenot expressed by any normal tissue (including mTEC) as genuineaeTSAs.
We have an incomplete picture of the HGSC antigenic landscapebecause immunopeptidomic studies of these tumors have been limitedto TAAs andmTSAs coded by canonical exons. Using exome sequenc-ing and NetMHC predictions, three groups have identified severalmTSA candidates which, in some cases, are recognized by TILs and/orperipheral blood lymphocytes (17–19). However, putative TSAs thatare not validated by mass spectrometry (MS) must be considered withcaution because, in other cancer types, predictions based on reverseimmunology are shown to be fraught with exceedingly high falsediscovery rates (10, 20). This is because current algorithms fail to takeinto account the numerous translational and posttranslational eventsthat regulate MAP biogenesis and presentation (21). T-cell reactivityagainst putative tumor antigens can be misleading because of theinherent cross-reactivity of T lymphocytes (22). Nevertheless, a majorlimitation of mTSAs is that they are rarely shared by different tumors.Using a direct proteomic approach, Schuster and colleagues performedhigh-throughputMS analyses in 42 ovarian tumors focusing onMAPscoded by the canonical exonic reading frame of classic unmutatedgenes (13). They found no mTSAs but identified several exonic MAPsthat were not detected in MS analyses of normal tissues. Most of theseMAPs were most likely TAAs, but some could be aeTSAs. With theseconsiderations in mind, the goal of our work in this study was toestablish the global landscape of HGSC mTSAs and aeTSAs coded byall genomic regions, irrespective of their mutational status.
Materials and MethodsHuman HGSC samples
Tumor fragments of HGSC1-6 and matched normal adjacenttissues of HGSC1-3 were obtained from Tissue Solutions. High-grade serous tumor tissue (OV606) or ascites (OV633 and OV642)were obtained from the PrincessMargaret Cancer Centre BiospecimenProgram (Toronto, ON, Canada) under protocols approved by theUHN Research Ethics Board. All patients gave informed writtenconsent. Snap-frozen samples described above were used for RNAextraction and MHCI-associated peptide isolation (described below).RNA sequencing (RNA-seq) data for OvCa48-114 were downloadedfrom the National Center for Biotechnology Information (NCBI)Sequence Read Archive under project PRJNA398141, converted toFASTQ file using SRA Toolkit (https://github.com/ncbi/sra-tools),and processed like the other samples. MS raw data of the samples inthis cohort were downloaded from ProteomeXchange Consortium viathe PRIDE partner with identifier PXD007635. HLA typing of eachsample was obtained from RNA-seq data using OptiType v1.0 (23)with default parameters. Sample information is presented in Supple-mentary Table S1.
RNA extraction and sequencingFor HGSC1-6, total RNA was isolated using the AllPrep DNA/
RNA/miRNA Universal kit (Qiagen) as recommended by the man-ufacturer. For OV606, OV633, and OV642, total RNA was isolatedusing TRIzol (Invitrogen), and purified using the RNeasy Mini kit(Qiagen) as recommended by the manufacturer. RNA from eachsample was assessed on a 2100 Bioanalyzer (Agilent Genomics) toensure an RNA integrity number >6. cDNA libraries were preparedfrom 4 mg (HGSC1-6) or 500 ng (OV606, OV633, and OV642) of totalRNA using the KAPA Stranded mRNA-Seq Kit (KAPA Biosystems).
Libraries were further amplifiedwithTruseq primers (Illumina) by 7 to10 cycles, and 1 nmol/L library per sample was used for paired-endRNA-seq on the Illumina HiSeq 2000 or the NextSeq 500 platform,which yielded 150 to 300 million reads.
Generation of customized reference databases for MS analysesFor each sample, we generated a customized “global cancer data-
base” by concatenating twomodules, the “canonical cancer proteome”and the “cancer-specific proteome” as previously described (15).Briefly, RNA-seq reads were trimmed by Trimmomatic v0.35 (24)and aligned to the reference human genome version GRCh38.88using STAR v2.5.1b (https://github.com/alexdobin/STAR/releases/tag/2.5.1b). Transcript expression was quantified in transcripts permillion (tpm) with kallisto v0.43.0 (https://pachterlab.github.io/kallisto) using default parameters. A sample-specific exome for eachsample was then built with pyGeno (25) by inserting single-basevariants (quality >20) identified using FreeBayes (https://github.com/ekg/freebayes). Annotated open reading frames (ORF) with tpm>0 were in silico translated from sample-specific exome using pyGenoand added into the canonical cancer proteome in a fasta format.
For a paired-end run, Illumina sequencing generates two FASTQfiles: Read1 and Read2. To generate cancer-specific proteomes,trimmed reads from Read1 FASTQ file were reverse complementedand were used for 33- and 24-nucleotide–long k-mer database gen-eration together with trimmed reads from Read2 FASTQ file. In orderto exclude sequencing errors and to limit database size, sample-specificthresholds ofminimal occurrencewere applied for 33 nucleotides: 7 forHGSC1-3, 8 for OV642, 10 for OV633, 4 for HGSC4 and OV606, 6 forHGSC5, 5 forHGSC6, and 3 forOvCa48-114. Cancer-specific 33-merswere obtained by subtraction of 33-mers expressed in purified humanthymic epithelial cells (TEC) harvested from six human thymi (15),then assembled into longer sequences (contigs) by the kmer_assemblytool fromNEKTAR (https://github.com/iric-soft/nektar). Contigs >34nucleotides long were 3-frame translated and split at internal stopcodons. The resulting subsequences longer than eight amino acidswere included in the relevant cancer-specific proteome.
Isolation of MAPsTumor fragments of HGSC1-6 and OV606 were cut into small
pieces (cubes, �3 mm in size) and 5 mL of ice-cold PBS containingprotein inhibitor cocktail (Sigma, cat. #P8340-5mL) was added. Tis-sueswere first homogenized twice for 20 seconds using anUltra TurraxT25 homogenizer (IKA-Labortechnik) set at speed 20,000 rpm andthen 20 seconds using an Ultra Turrax T8 homogenizer (IKA-Labor-technik) set at speed 25,000 rpm. Then, 550 mL of ice-cold 10� lysisbuffer (5% w/v CHAPS, Sigma, cat. # C9426-5G) was added to eachsample. After a 60-minute incubation with tumbling at 4�C, sampleswere spun at 10,000 � g for 30 minutes at 4�C. Supernatants weretransferred into new tubes containing 1 mg of W6/32 antibody (Bio XCell, cat. #BE0079) covalently cross-linked to 1 mL of protein Amagnetic beads (Pure Proteome, cat. #LSKMAGA10) using dimethyl-pimelidate (26), and MAPs were immunoprecipitated as previouslydescribed (27). MAP extracts were then dried using a Speed-Vac andkept frozen at �20�C until MS analyses.
Liquid chromatography-tandem MS analysesDried peptide extracts were resuspended in 0.2% formic acid and
loaded on a homemade C18 analytical column (15 cm � 150 mm i.d.packed with C18 Jupiter Phenomenex) with a 56-minute (HGSC1-6)or 106-minute (OV606, OV642, andOV633) gradient from 0% to 30%acetonitrile (0.2% formic acid) and a 600 nL/min flow rate on an Easy-
The Global Landscape of Ovarian Cancer–Specific Antigens
AACRJournals.org Cancer Immunol Res; 8(4) April 2020 545
nLC II system. Samples were analyzed with a Q-Exactive HF massspectrometer (Thermo Fisher Scientific) in positive ion mode withNanospray 2 source at 1.6 kV. Each full MS spectrum, acquired with a60,000 resolution, was followed by 20 tandem MS (MS/MS) spectra,where the most abundant multiply charged ions were selected forMS/MS sequencing with a resolution of 30,000, an automatic gaincontrol target of 5 � 104, an injection time of 100 ms, and collisionenergy of 25%. For HGSC4-6, each full MS spectrum, acquired with a60,000 resolution, was followed by 20 MS/MS spectra, where themost abundant multiply charged ions were selected for MS/MS se-quencing with a resolution of 30,000, an automatic gain control targetof 2 � 104, an injection time of 800 ms and collision energy of 25%.
Identification of MAPsAll liquid chromatography (LC)-MS/MS (LC-MS/MS) data were
searched against the relevant global cancer database using PEAKS 8.5or PEAKS X (Bioinformatics Solution Inc.). For peptide identification,tolerance was set at 10 ppm and 0.01 Da for precursor and fragmentions, respectively. For the reanalysis of data from Schuster andcolleagues (13), tolerance was set at 5 ppm and 0.5 Da for precursorand fragment ions, respectively. The occurrences of oxidation (M) anddeamidation (NQ) were set as variable modifications. We used amodified target decoy approach built in PEAKS and applied a sam-ple-specific threshold on the PEAKS score to ensure a false discoveryrate of 5%, calculated as the ratio between the number of decoy hits andthe number of target hits above the score threshold. Applied thresholdson PEAKS score were 9 (HGSC3), 10 (OvCa114, HGSC4-5), 11(HGSC5, OV633, OV642), 12 (HGSC2, OV606, OvCa48, OvCa70,OvCa80, OvCa111), 13 (HGSC1, OvCa64, OvCa105, OvCa99, andOvCa109), 14 (OvCa53, OvCa104), 15 (OvCa84), 16 (OvCa65), and 18(OvCa58). Peptides that passed the threshold were further filteredaccording to the following criteria: peptide length between 8 and 11amino acids, and MHC allele affinity rank ≤2% based on the predic-tion of NetMHC4.0 (28). Peptides that fulfilled the above criteria wereconsidered as MAPs and are reported in Supplementary Table S2.
Identification of TSA candidatesTo identify TSA candidates, eachMAPand its coding sequencewere
queried to relevant cancer and normal (TEC) canonical proteomes or24-nucleotide-long k-mer databases, as previously described (15).MAPs were labeled as TSA candidates in two cases: (i) peptidesequences were detected neither in the normal canonical proteomeof the sample nor in normal k-mers, or (ii) peptides were absent fromboth cancer and normal canonical proteomes and their RNA codingsequence was overexpressed by at least 10-fold in cancer cells com-pared with TECs. When a MAP corresponded to several RNAsequences, it was considered as a TSA candidate only when all of thesequences were consistent with the TSA candidate status. MS/MSspectra of all TSA candidates were manually validated to remove anyfalse identification.
We assigned a genomic location to all TSA candidates by mappingreads containing MAP-coding sequences on the reference genome(GRCh38) using BLAT (https://genome.ucsc.edu/cgi-bin/hgBlat).TSA candidates for which reads matched to hypervariable regions(HLA, Ig, or TCR genes) were excluded. TSA candidates were classifiedas mTSAs if they contained variants in their MAP-coding sequencesthat did not match with known germline polymorphisms [reported inDatabase of Single-Nucleotide Polymorphisms (dbSNP) v149, http://www.ncbi.nlm.nih.gov/SNP/]. Nonmutated candidates were classifiedas aeTSA candidates and subjected to further assessment of theirexpression in normal tissues and organs.
TSA validation with synthetic peptidesTSA identifications were validated by comparing the endogenous
peptide spectrawith synthetic peptide spectra (Supplementary Fig. S1).Synthetic peptides were dissolved in DMSO at 1 nmol/mL and dilutedat 50 fmol/mL in 4% formic acid. The Q-Exactive HF mass spectrom-eter (Thermo Fisher Scientific) was operated at a resolution of 60,000in MS1 scan, where each full MS spectrum was followed by 20MS/MSspectra. The most abundant multiply charged ions were selected forMS/MS sequencing with a resolution of 30,000, an automatic gaincontrol target of 5 � 104, an injection time of 100 ms, and collisionenergy of 25%. For peptides identified in the data set of Schuster andcolleagues (13), a TribridOrbitrap Fusionmass spectrometer (ThermoFisher Scientific) coupled to an Easy nLC1000 was used. Peptides wereseparated on a custom C18 reversed phase column (150 mm i.d. �100 mm, C18 Jupiter, Phenomenex) using a flow rate of 600 nL/minand a linear gradient of 5% to 30%acetonitrile (0.2%FA) in 56minutes.Survey scan (MS1) were acquired with the Orbitrap at a resolvingpower of 120,000 (atm/z 200) over a scan range of 350 to 1200m/zwitha target values of 5� 105 and amaximum injection time of 50ms.MS/MS spectra were acquired in collision-induced dissociation with anormalized collision energy of 30. Up to 20 precursor ions wereaccumulated with a precursor isolation window of 1.3 m/z, anadvanced gain control of 2 � 104 with a maximum injection time of50 ms.
Spectra correlations were computed using a script written in Pythonv3.7.2 using the following steps. The first step reads the list of peptidesand computes expected peptide fragments (b/y ions, singly and doublycharged, water and ammonia losses) with pyteomics v4.0.1 (https://github.com/levitsky/pyteomics). The next step searches reproduciblydetected peptide fragments usingmultiple synthetic peptide scans. Thelist of peptide scans is provided by PEAKS X search result file and MSraw files are read using Thermo MSFileReader Python bindings(pre-release version, https://github.com/mobiusklein/MSFileReader-Python-bindings). Fragment tolerance was set to 0.5 Da for collision-induced dissociation and 0.02 Da for higher-energy collisional disso-ciation scans. Thereafter, root scaled intensities of reproducible frag-ments were correlated between all corresponding endogenous andsynthetic peptide scan pair. The Pearson correlation coefficient,P value, and confidence interval were computed using SciPy v1.2.1(https://www.scipy.org/). Finally, the scan pair of each peptide with thelowest P value was retained to generate a mirror plot using spectru-m_utils v0.2.1 (https://github.com/bittremieux/spectrum_utils).Peptides with poor Pearson correlation scorewere inspectedmanually.This occurred for peptides identified from the data reanalysis ofSchuster and colleagues (13), due to the usage of different massspectrometers.
TSA validation with an isobaric peptide labeling approachWe confirmed eight TSAs fromHGSC1-3 using an isobaric peptide
labeling approach with corresponding specimen (SupplementaryFig. S2). With remaining peptide extracts from HGSC1-3, we usedan isobaric peptide labeling approach in confirmation of 8 expectedTSAs (TASDLNLKV, TEISNSQAA, RTATPLTMK, RTATPLTMKK,SVYMATTLK, SQGFSHSQM, STQMTITTQK, and VTIDTTQTK)in the peptide extracts of the corresponding specimens. Endogenouspeptide extracts were labeled with TMT6-127 (Thermo Fisher Scien-tific) following the manufacturer's protocol, whereas the syntheticpeptides (JPT PeptideTechnologies) were derivatized with TMT6-129and TMT6-131 for spiked levels of 10 fmoles and 200 fmoles, respec-tively. Peptides and corresponding extract were loaded on a home-made C18 analytical column (15 cm � 150 mm i.d. packed with C18
Zhao et al.
Cancer Immunol Res; 8(4) April 2020 CANCER IMMUNOLOGY RESEARCH546
Jupiter Phenomenex) with a 56-minute gradient from 0% to 30%acetonitrile (0.2% formic acid) at 600 nL/min flow rate on a nEasy-LCII system. Samples were analyzed with an Orbitrap Fusion massspectrometer (Thermo Fisher Scientific). Full MS spectra wereacquired with a 120,000 resolution, an automatic gain control of5 � 105 and maximum injection time of 100 ms. Ions were selectedfor fragmentation, based on inclusion lists, using higher-energy col-lisional dissociation with 35% collision energy and an isolation win-dowof 1.6 Th. The automatic gain control target was set to 2� 104witha maximum injection time of 500 ms and a resolution of 50,000.
Computation of hydrophobicity indexFor additional validation, we assessed the correlation between
retention time and hydrophobicity for TSAs (SupplementaryFig. S3). The hydrophobicity values of the putative TSAs and otherMAPs were calculated using an online available tool SSRCalc (https://bio.tools/ssrcalc). Only unmodified peptides were included, and pre-dictions were based on the 100Å C18 column, 0.1% formic acidseparation system. We correlated the observed mean retention timeof a given peptide against the predicted hydrophobicity index, whichcorresponded to the percentage of acetonitrile at which the peptideelutes from the chromatographic system. For samples analyzed on ourLC-MS/MS system,we computed a hydrophobicity index from a linearregression using retention time of our peptide quality control standardsample injected before endogenous and synthetic peptides.
Tissue expression of sequences coding for aeTSA candidatesRNA-seq data for 27 different tissues were downloaded from the
Genotype-Tissue Expression (GTEx) through dbGaP accession num-ber phs000424.v7.p2 on April 16, 2018, and were used to assess theexpression of the coding sequence of aeTSA candidates, as previouslydescribed (15). RNA-seq data from 50 donors were obtained for eachtissue, except for the cervix (n¼ 6), fallopian tube (n¼ 7), bladder (n¼12), kidney (n ¼ 38), and adipose tissue (n ¼ 49; SupplementaryTable S3). For each aeTSA candidate, the number of reads containingthe MAP-coding sequence was normalized to reads per hundredmillion reads sequenced (rphm), log-transformed (log10(rphm þ1)), and averaged across all samples available for each tissue. aeTSAcandidates exhibiting no peripheral expression at rphm >10 in tissuesother than the MHClow tissues (brain cortex, nerve, and testis) wereconsidered as genuine aeTSAs.
Expression of TSA-coding regionsThe RNA expression of the TSA-coding region was quantified from
mapped bam files using the “qCount” function of R package“QuasR” (29) with parameter “orientation ¼ ‘same,’” as counts of allreads that overlapped with TSA-coding region on the TSA-codingstrand. The countswere normalized to reads per hundredmillion readsmapped. In aeTSA expression analyses, we specifically analyzed theunique region to which individual aeTSAs were mapped because thecontext of the surrounding sequence may influence aeTSA expression.
Clinical and genomic data from The Cancer Genome AtlasProcessed and normalized level 3 data with hg38 for HM27
methylation, DNA copy-number variation (CNV), RNA-seq geneexpression (FPKM normalized), as well as the clinical data, weredownloaded from The Cancer Genome Atlas (TCGA) open-accessdatabase using the R package “TCGAbiolinks” (30). Matched RNA-seq and CNV (n ¼ 376) or methylation data (n ¼ 378) from TCGA-OV patients were used for Spearman correlation to assess associationsbetween aeTSA RNA expression and CNV or DNA methylation
changes. Case IDs of included samples were listed in SupplementaryTable S4. To adjust for multiple tests, the Benjamini–Hochbergmethod was applied. Arm-level DNA copy-number alterations weredownloaded from the Broad Institute TCGA Genome Data AnalysisCenter (doi:10.7908/C1P84B9Q). For each tumor, immune cell scoresrepresenting immune cell populations were estimated based on RNA-seq data as described by Danaher and colleagues (31).
Frequency of aeTSA presentationTo estimate the number of aeTSAs presented by individual patients,
we performed a bioinformatic simulation based on two parameters.First, the likelihood of aeTSA expression was based on the proportionof tumors expressing the corresponding RNA in the TCGA-OVcohort. For aeTSA containing a germline SNP, the expression likeli-hood was calculated as follows: (proportion of TCGA tumors expres-sing aeTSA) � (SNP frequency in the given population). The SNPfrequencies were obtained from the Genome Aggregation Database.Second, the HLA allele frequencies were retrieved from USA NationalMarrowDonor Program for the European–American population (n¼1,242,890), African American (n ¼ 416,581), and Chinese (n ¼99,672). Next, we simulated patients’ HLA genotype with six HLAclass I alleles based on the reported frequencies in the given population.Because we assumed that the six HLA alleles were independent events,someHLA loci were homozygous in simulated patients. An aeTSAwasconsidered to be presented in a simulated patient when both the aeTSAand the relevant HLA allele were expressed. Expression of each aeTSAwas considered to be an independent event, except for the overlappingaeTSAs whose expression status was only simulated once for the sameoverlapping group. One million simulated patients and their aeTSApresentation status were generated for each of the three populationsand used to plot the distribution.
Statistical analyses and data visualizationAnalyses and figures were performed using the R v3.5.1 or Python
v2.7.6. The “gplots” package in R was used to generate heat maps ofTSA-coding region expression in tumors. Correlation tests were doneusing the R function “cor.test” with the Spearman method unlessotherwise indicated. Tests involving comparisons of distributions wereperformed with the one-way ANOVA test, and pairwise comparisonsbetween groups were performed by the Wilcoxon rank sum test. Log-rank P values for survival analysis were calculated with the “survival”Rpackage.
Data and materials availabilityMS raw data and associated databases were deposited to the
ProteomeXchange Consortium via the PRIDE (32) partner repositorywith the following data set identifiers: PXD014062 and 10.6019/PXD014062. RNA-seq data were deposited in the NCBI SequenceRead Archive and Gene Expression Omnibus (GEO) under accessioncode GSE131880.
ResultsProteogenomic analyses identify 103 TSAs in 23 HGSCs
To get a systems-level characterization of the TSA landscape, weperformed direct MAP identification with high-throughput MS/MSanalyses (33–35). Current search engines rely on user-defined proteindatabases to match each acquired MS/MS spectrum to a peptidesequence (36). Hence, a peptide in a test sample can only be identifiedby the search engine when its sequence is included in the referencedatabase. Because generic reference protein databases, such as
The Global Landscape of Ovarian Cancer–Specific Antigens
AACRJournals.org Cancer Immunol Res; 8(4) April 2020 547
UniProt, do not contain sample-specific mutations, out-of-frametranslation events, and nonexonic sequences, our quest to captureTSAs coded by all genomic regions required construction of cus-tomized databases containing tumor-specific translation productsfor each tumor. Therefore, we used our previously describedproteogenomic approach (15) to build a customized search databasefrom RNA-seq reads of each sample analyzed. Such customizeddatabases contain two modules: the canonical proteome (in-frametranslation of exons) and the cancer-specific proteome, which is a3-frame translation of cancer-specific RNA sequences after sub-traction of normal RNA sequences (from TECs; Fig. 1). As inprevious studies, we used TECs as a normal control for two reasons:(i) their key role in establishing immune tolerance during thedevelopment of immature T cells (i.e., central tolerance) and (ii)their remarkable ability to promiscuously express more transcriptsthan other types of somatic cells (37). MAPs from nine primaryHGSC samples were obtained by immunoprecipitation of MHC Imolecules, then analyzed by LC-MS/MS (15, 27). We also reana-lyzed the immunopeptidomic data of an additional cohort of 14HGSCs reported by Schuster and colleagues (13), by applying ourproteogenomic approach to the samples for which matched RNA-seq and MS data were available.
TSA candidates that overlap genomic variants absent in dbSNPwere unlikely to represent germline polymorphisms and were there-fore labeled as mTSAs. Such classification yielded 12 mTSAs from the23 samples analyzed (Supplementary Fig. S4; SupplementaryTable S5). None of these mTSAs have been previously reported. Ofthe 12 mTSAs, three resulted from in-frame exonic translation, fourfrom out-of-frame exonic translation, and eight from noncodingsequences. For three tumors, matched normal tissue was available,and RNA-seq analyses confirmed that mTSA variants were not germ-line polymorphisms (Supplementary Fig. S5). For tumors withoutmatched normal tissues, we could not formally exclude that somemTSAs might have corresponded to rare polymorphisms absent indbSNP. The possibility that we may have slightly overestimated thenumber of mTSAs would only reinforce our conclusion that, with lessthan one mTSA per tumor (12 mTSAs/23 tumors), mTSAs are so rarein HGSCs that they do not represent attractive targets. Because classicTSA discovery methods focus strictly on mTSAs resulting from in-frame exonic translation, they would have uncovered only three of theTSAs reported herein.
For the unmutated TSA candidates, we applied stringent criteria toidentify those that were genuine aeTSAs, that is, those whose expres-sion was cancer specific. To this end, we analyzed RNA expression in27 peripheral tissues across the human body. Due to their diversegenomic origins and association with alternative splicing events, wereasoned that quantification of peptide-coding sequence expressionprovided a uniform and more accurate measurement than quantifi-cation of whole gene expression.We thereby excluded fromour aeTSAlist all candidates whose coding RNA were expressed (rphm >10) inany peripheral tissue other than MHClow tissues (brain, nerve, andtestis). Three aeTSAs expressed in brain and nerve, whose applicationshould be case dependent, asMHCIhave been shown to increase underinflammatory conditions (38).When the coding sequence of an aeTSAcandidate contained a germline single polymorphism (reported indbSNP), this candidate was labeled as a valid aeTSA only when theSNP-containing sequence and the reference sequence fulfilled theabove criteria (Supplementary Fig. S6). Overall, 91 aeTSA candidatessatisfied our stringent criteria (Fig. 2; Supplementary Table S5). Five of91 aeTSAs were expressed in the testis, showing that some CGAs areaeTSAs but that most aeTSAs are not CGAs.
Each TSA was assigned a genomic location. When multiple loca-tions were possible, the one with the highest occurrence of matchingRNA reads was selected. Features of all TSAs are reported in Supple-mentary Table S5. It is formally possible that our stringent approachmay have underestimated the total number of aeTSAs resulting fromatypical translation [50untranslated region (UTR), 30UTR, intergenic,frameshift]. Indeed, although we knew the reading frame used togenerate MAPs in tumors when their coding RNA was expressed insome normal tissue, we could not infer which reading frame wastranslated. Such aeTSAs candidates were therefore excluded in order toavoid inclusion of false positives in our TSA list.
Most HGSC TSAs are unmutated MAPs resulting fromnoncanonical translation
With an average of 2,200 unique MAPs identified per sample, wefound a total of 103 unique TSAs (Fig. 3A; Supplementary Table S2).Each TSA was validated with its corresponding synthetic peptide(Supplementary Figs. S1 and S2). The retention characteristics of ourTSAs were consistent with the distribution of MAPs and were cor-related with their hydrophobicity indexes, supporting their correctidentification (Supplementary Fig. S3). The number of TSAs identified
Figure 1.
Schematic workflow of the TSA identification pipeline used in this study. HGSC samples were processed for immunoprecipitation (IP) and RNA-seq. Peptidesequences were identified using MS analyses that identified MAPs by searching for matches in customized individual global cancer databases built from RNA-seqdata. pMHC, peptide–MHC I complex; SNV, single-nucleotide variant.
Zhao et al.
Cancer Immunol Res; 8(4) April 2020 CANCER IMMUNOLOGY RESEARCH548
per sample significantly correlated with the number of MAPs(Fig. 3B), and a correlation between the number of MAPs per HLAallele and tumor sample size was seen (Pearson correlation coefficient¼ 0.31; Supplementary Fig. S7). This is consistent with the notion thattumor sample size is a limiting factor inMS analyses (27). In principle,TSAs that originate from mutated noncoding sequences could bedesignated as bothmTSAs and aeTSAs. Arbitrarily, we decided to labelthem as mTSAs. The rationale was that irrespective of their genomicorigin (exonic or not), mTSAs are expected to be “private TSAs,” thatis, not to be shared by a large proportion of tumors. In contrast,unmutated aeTSAs can theoretically be shared by a significant pro-portion of HGSCs.
The features of TSAs identified in samples initially processed by usor by Schuster and colleagues (13) were similar (Fig. 3C). Thissuggested that our proteogenomic approach could be applied toRNA-seq and MS data, in general, and was not ostensibly affectedby interlaboratory variability. In both cohorts, about 89% of TSAswereunmutated, and the majority of TSAs resulted from noncanonicaltranslation, primarily from noncoding regions, and to a lesser extentfrom out-of-frame exonic translation (Fig. 3C). Two features ofaeTSAs were observed: (i) 79% derived from noncoding sequences,in particular intronic (29%) and intergenic (22%) sequences and (ii)91% were novel MAPs (Fig. 3D; Supplementary Table S5). Previouslyreported MAPs derived from in-frame exonic translation, except for
Figure 2.
Expression in normal tissues of RNAs coding for aeTSA candidates. Heat map showing the average RNA expression of aeTSA coding sequences in 27 peripheraltissues, with color intensity corresponding to the expression level in each tissue (mean log-transformed rphm). Bold boxes indicate tissues/organs with above-threshold RNA expression (mean rphm >10). Numbers beside each peptide sequence show the number of tissues with above-threshold expression of thecorresponding RNA. Red arrowheads point to the aeTSAs expressed in testis. Orange dots denote the aeTSAs expressed in brain and nerve. ncRNA, noncoding RNA;UTR, untranslated region.
The Global Landscape of Ovarian Cancer–Specific Antigens
AACRJournals.org Cancer Immunol Res; 8(4) April 2020 549
one that matched to a processed transcript (biotype annotated byEnsembl database) whose corresponding protein isoform is includedin the UniProt database (13, 39–44).
Expression of aeTSA-coding transcripts in ovarian cancersamples
We then asked whether cancer-specific expression of aeTSA-coding transcripts resulted from random transcriptional noise orfrom recurrent transcriptional aberrations. To address this, weanalyzed the RNA expression of genomic regions coding for our91 aeTSAs in samples from this study and from the ovarian cancercohort from TCGA. Regions coding for aeTSAs were expressed in asubstantial proportion of ovarian cancers: 71 (78%) were expressedin at least 10% of samples and 16 (18%) were expressed in at least80% of samples (Fig. 4; Supplementary Table S5). These commonlyexpressed regions have high potential to generate shared TSAsbetween patients. We conclude that expression of this set of 91
aeTSA-coding transcripts in HGSC is not a rare or random eventbut rather a common feature of HGSCs.
Genomic correlates of aeTSA expressionTo understand the mechanisms of aeTSA expression, we took
advantage of multiomic data from the data set TCGA-OV to explorethe relationship between aeTSA RNA expression and local genetic orepigenetic aberrations. Correlations were tested between focal DNAcopy-number changes, DNA methylation on the gene promoterregions, and the RNA expression for each aeTSA when applicable(Fig. 5A). When an aeTSA derived from a genomic region that waspart of a gene (exon, intron, or UTR), we also analyzed the correlationbetween expression of the relevant gene and aeTSA expression. In thelatter situation, we observed a conspicuous correlation between geneand aeTSA expression (Fig. 5A). This suggested that for aeTSAswhosecoding region was in a gene, regulation of aeTSA expression generallyaffected the whole gene. Changes in DNA copy number showed a
Figure 3.
Most TSAs derive from unmutated nonexonic sequences.A,Number of MAPs (left) and TSAs (right) identified in each sample from our study and Schuster et al. (13).Dashed line indicates the mean number of MAPs identified per sample. B, Scatter plots showing the Pearson correlation between the number of MAPs and TSAsidentified per sample. Linear regression line is shown. The gray area represents the 95% confidence interval. C, Bar graph showing the origin of TSAs identified in ourcohort and the reanalyzed data set of Schuster et al. (13). Shades of blue depict the number of TSAs resulting from in-frame exonic translation (coding-in), out-of-frame exonic translation (coding-out), or nonexonic translation (noncoding).D,Pie chart showing the translational reading frame (inner pie), detailed genomic origin(middle pie) of aeTSAs, and their report status (outer circle).
Zhao et al.
Cancer Immunol Res; 8(4) April 2020 CANCER IMMUNOLOGY RESEARCH550
positive correlation with RNA expression of aeTSAs. This was the casefor both within-gene aeTSAs (left) and out-of-gene aeTSAs (antisenseand intergenic, right). This suggested that DNA copy-number altera-tions had a substantial effect on aeTSA expression, especially for thewithin-gene aeTSAs expressed in a larger proportion of tumors
(Supplementary Fig. S8A). When we examined the chromosomaldistribution of aeTSA-coding regions, we found that several chromo-some arms, frequently amplified in HGSC, yielded many aeTSAs(Fig. 5B). For example, the long arm of chromosome 3, which iscommonly amplified in ovarian cancer (45), was the source of eight
Figure 4.
Expression of aeTSA-coding regionsacross ovarian cancer samples. Heatmap shows the RNA expression forthe region coding each aeTSA in 9samples reported in this study (left)and 378 samples from TCGA-OV(right), with blue intensity showingRNA expression in reads mapped inregion per million reads. aeTSAswere ordered according to the pro-portion of tumors in which they wereexpressed.
Figure 5.
Copy-number changes correlate with expression of several aeTSAs. A, Heat map shows the Spearman correlation between aeTSA RNA expression and DNA copynumber, promotermethylation, or gene expression for aeTSAs locatedwithin genes (left) or outside of genes (right). Data not available are shownas light gray.B,Thenumber of aeTSAs identified for each chromosome arm (top) with arm-level amplification score (bottom). Asterisks indicate amplification considered to besignificant (GISTIC Q value < 0.25; ref. 45). Ampl., amplification.
The Global Landscape of Ovarian Cancer–Specific Antigens
AACRJournals.org Cancer Immunol Res; 8(4) April 2020 551
aeTSAs. As one of the top amplified regions, MECOM located at3q26.2 (45) generated three overlapping exonic out-of-frame aeTSAs(Supplementary Table S5). However, amplification of chromosomearms was not always necessary (e.g., 15q) nor sufficient (e.g., 8q) togenerate aeTSAs (Fig. 5B; Supplementary Fig. S8B).
Due to the technology used by TCGA for analysis of DNA meth-ylation (HM27 arrays), methylation data were unavailable for the out-of-gene aeTSAs and the promoters of some aeTSA source genes. Ouranalysis of promoter methylation was therefore limited to a subset of17 aeTSAs. Still, for six aeTSAs, we observed a significant correlationbetween DNA methylation and aeTSA expression (Fig. 5A). Thecorrelation was negative in five cases and positive in one case. Thisis consistent with the notion that promoter demethylation frequentlyresults with enhanced transcription. The two genes showing thehighest negative correlation were MAGEC1 (r ¼ �0.53, Padj ¼ 1.6� 10�26) and MAGEA4 (r ¼ �0.51, Padj ¼ 6.7 � 10�25; dark bluebars, Fig. 5A). Genes of the MAGE family are CGAs that are over-expressed in several cancer types, including HGSC (3). Overall, weconclude that aeTSA expression is regulated, at least in part, at thetranscriptional level by variations in gene copy number and DNAmethylation.
The expression of three aeTSAs correlates with improvedsurvival
Next, we sought to evaluate whether some aeTSAs might elicitspontaneous protective immune responses. Addressing this question is
complicated by the fact that expression of aeTSAs at the peptide levelrequires, in addition to expression of aeTSA RNA, the presence of therelevant HLA allotype. We therefore allocated patients from TCGAinto four subgroups based on the expression (or not) of individualaeTSA RNA and the presence (or not) of the relevant HLA allotype.Presentation of three aeTSAs correlated with a more favorable clinicaloutcome (Fig. 6A–C). The polymorphism of HLA alleles considerablyreduced the size of each group, and therefore the statistical power ofthis analysis. Accordingly, the log-rank P value for the three aeTSAsranged from 0.013 to 0.076 (Fig. 6A–C). Nonetheless, two observa-tions suggested that these correlations were biologically meaningful.First, the “protective effect” of these aeTSAs appeared to be HLArestricted. In patients expressing aeTSA RNA, survival was superiorwhen they also expressed the relevant HLA allele. Second, expressionof the RTHQMNTFQR aeTSA and its relevant HLA allotype showed apositive correlation with tumor infiltration by T cells and cytotoxic Tcells (Fig. 6D and E).
What is the median number of aeTSAs presented by individualtumors?
With the list of 91 aeTSAs, we estimated to what extent our studymay benefit TSA-targeted immunotherapy. We randomly simulatedthe presentation status for 91 aeTSAs in one million patients. Toestimate HLA allele frequencies, we used the three largest data setsfrom the USA National Marrow Donor Program: European Amer-icans, African Americans, and Chinese (46). Six HLA alleles and
Figure 6.
Presentation of three aeTSAsmay elicit spontaneous antitumor immune responses.A–C, Kaplan–Meier curves depicting survival of four groups of patients in TCGA-OV: ED, expression of aeTSA-coding RNA, relevant HLA allotype absent; ND, no expression of aeTSA-coding RNA, relevant HLA allotype absent; EP, expression ofaeTSA-coding RNA, relevant HLA allotype present; and NP, no expression of aeTSA-coding RNA, relevant HLA allotype present. Color shades represent the 95%confidence interval for the associated curve. Log-rank P values are indicated. D and E, The abundance of T cells and cytotoxic T cells in tumors from the four groupspresented in C. The bottom and top of the boxes represent the lower and upper quartiles. The thick line indicates themedian value of each group. Outliers are shownas points outside the boxes. Statistical difference between the four groups was analyzed using the one-way ANOVA test.
Zhao et al.
Cancer Immunol Res; 8(4) April 2020 CANCER IMMUNOLOGY RESEARCH552
aeTSA expression status were randomly generated independentlyusing the allele frequencies in the given population, the expressionproportion in TCGA-OV tumors, and the SNP frequencies whenapplicable. The number of aeTSAs per individual tumorwas calculatedas the sum of expressedHLA-aeTSA pairs. Based on these simulations,we found that at least one aeTSA could be found in 98% EuropeanCaucasians, 74% African Americans, and 77% Chinese (Fig. 7). Themedian number of aeTSAs per tumorwasfive in EuropeanCaucasians,two in African Americans, and four in Chinese. Differences amongthese populations resulted from variations in HLA allele frequenciesand the fact that our tumor samples were mainly from EuropeanCaucasians. We suspect that our calculations underestimated thenumber of aeTSAs per tumor, mainly for three reasons. First, we didnot take into account that more than 50% of MAPs bind two or moreHLA allotypes, often across supertypes or even loci (47). Second,genomic regions that code for a given MAP frequently generateoverlapping MAPs presented by different HLA allotypes (21). Third,for the five aeTSAs that include nonsynonymous SNPs listed indbSNP, we assumed that only the SNP variant generating MAPs inour samples was valid and that the other SNP variant did not generateMAPs. We adopted this cautious strategy because changes in a singleamino acid are sufficient to abrogate MAP presentation (48). Weconclude that vaccines, including the current set of 91 aeTSAs, wouldcover practically all Caucasians with HGSC. Proteogenomic analysesof HGSC samples obtained from other groups should be performed toattain the same level of population coverage.
DiscussionThe main conclusion from this study is that expanding the search
space to all reading frames of all genomic regions has a transformativeimpact on the discovery of HGSC TSAs. Out of 103 TSAs reportedherein, only three would have been picked up by strategies focusingsolely on mutated exonic sequences. Limited numbers of mTSAs alsoresulted from out-of-frame exonic translation (n ¼ 2) or fromnoncoding sequences (n ¼ 7). In addition to their relative scarcity,mTSAs were unique to individual tumors.We therefore believe that, atleast for HGSCs, mTSAs are not particularly attractive targets. Incontrast, aeTSAs, which derived primarily from noncoding sequences,are considerably more attractive because they were more numerous
(91 of our 103 TSAs) and were shared by a substantial proportion ofHGSC tumors.
aeTSAs are not expressed in normal tissues, and their expression inHGSC was regulated at the transcriptional level by variations insomatic copy-number alterations and DNA methylation. In HGSC,the bulk of genetic changes is not somatic point mutations, but rathersomatic copy-number alterations (49). These copy-number alterationsdrive cancer through defects in DNA repair, losses of tumor suppres-sors (such as TP53), and amplification of oncogenes (45, 49). Becausesomatic DNA copy-number alterations are common in various cancertypes, it will be interesting to explore the impact of DNA copy numberon aeTSA expression in other cancers. DNA hypomethylation alsocontributed to the aberrant expression of six aeTSAs, including twomembers of the MAGE family that are silenced by methylation innormal cells (3). However, because most aeTSAs derive from allegedlynoncoding regions, it will be necessary to combine whole-genomemethylation analyses with proteogenomic studies of HGSCs in orderto evaluate more comprehensively the relation between DNA meth-ylation and aeTSA expression. Other regulatory mechanisms couldalso be relevant to aeTSA biogenesis. In particular, recurrent alterna-tive splicing events have been identified in various cancers (50) andcould be instrumental to the appearance of intronic aeTSAs. Finally,the potential impact of trans factors, such as miRNAs and transcrip-tion factors, on aeTSAs has yet to be explored. Our observation thataeTSA expression was a common feature of HGSCs should provideimpetus to perform such analyses.
Perhaps the nextmost critical question to address in future studies isthe immunogenicity of individual aeTSAs. Our bioinformatic analysesof TCGA-OV provided hints that some aeTSAs might elicit sponta-neous protective immune responses. However, the question of immu-nogenicity has to be addressed experimentally by studying responses ofblood and tumor-infiltrating lymphocytes to aeTSAs. Testing theimmunogenicity of 91 antigens is far from being trivial but is certainlydoable using state-of-the-art high-throughput pipelines (51). Wespeculate that a high proportion of our aeTSAs will prove to beimmunogenic. Indeed, when we previously performed comprehensive(ex vivo and in vivo) immunogenicity analyses of aeTSAs discovered inmice with the method used in the present article, we found that theywere all immunogenic (15). In contrast to what is seen with “predictedMAPs” (i.e., MAPs not validated byMS analyses), a study has revealed
Figure 7.
Estimated frequency of aeTSAs presented by individual HGSCs in three different populations. Onemillion simulated patientswere generated for each population. AnaeTSA was considered to be present in a simulated patient when its RNA was expressed and the relevant HLA allotype was present. Frequency of aeTSA RNAexpressionwasbased onTCGA-OVRNA-seqdata (as inFig. 4). HLAallotype frequencieswere obtained fromUSANationalMarrowDonor Program. Reddashed linesindicate the median number of aeTSAs per tumor in each population.
The Global Landscape of Ovarian Cancer–Specific Antigens
AACRJournals.org Cancer Immunol Res; 8(4) April 2020 553
that more than 80% of MS-validated mouse MAPs are immuno-genic (52). Finally, our bioinformatic predictions suggest that thegroup of aeTSAs reported herein would be sufficient to cover HGSCin almost all Caucasians. We intend to extend our proteogenomicanalyses to HGSC samples obtained from other populations withdifferent HLA allele frequencies. Many epitopes can be included in avaccine. Hence, for HGSC treatment, it might be possible togenerate aeTSA-based off-the-shelf vaccines. In order to increasethe level of “precision,” different vaccines could be created thatcontain aeTSAs presented by one or a few HLA allotypes (e.g., HLA-A�02:01 aeTSAs).
Disclosure of Potential Conflicts of InterestQ. Zhao and C. Perreault have ownership interest in a patent application filed by
Universit�e de Montr�eal. No potential conflicts of interest were disclosed by the otherauthors.
Authors’ ContributionsConception and design: Q. Zhao, P. Thibault, C. PerreaultDevelopment of methodology: J.-P. Laverdure, J. Lanoix, C.M. Laumont,K. Vincent, P. ThibaultAcquisition of data (provided animals, acquired and managed patients, providedfacilities, etc.): Q. Zhao, J. Lanoix, C. Durette, C. Cot�e, P. Gendron, K. Vincent,D.G. Millar, P.S. Ohashi
Analysis and interpretation of data (e.g., statistical analysis, biostatistics,
computational analysis): Q. Zhao, J.-P. Laverdure, �E. Bonneil, P. Gendron,M. Courcelles, S. Lemieux, P. ThibaultWriting, review, and/or revision of the manuscript: Q. Zhao, J.-P. Laverdure,J. Lanoix, C. Cot�e, C.M. Laumont, P. Gendron, K. Vincent,M. Courcelles, D.G.Millar,P.S. Ohashi, P. Thibault, C. PerreaultAdministrative, technical, or material support (i.e., reporting or organizing data,constructing databases): J.-P. Laverdure, P. GendronStudy supervision: S. Lemieux, P. Thibault, C. Perreault
AcknowledgmentsWe thank J. Huber, R. Lambert, and J.-D. Larouche for technical support and
thoughtful suggestions. We also thank staff of the GTEx Consortium and TCGAProgram for sharing RNA-seq data from normal human tissues and ovariancancers, respectively. This work was supported by grants from the Terry FoxTranslational Cancer Research Program for The Immunotherapy Network(iTNT)/Targeting Ovarian Cancer (to P.S. Ohashi and C. Perreault), and GenomeQuebec's Center for Advanced Proteomics and Chemogenomic Analyses (toP. Thibault).
The costs of publication of this article were defrayed in part by the payment of pagecharges. This article must therefore be hereby marked advertisement in accordancewith 18 U.S.C. Section 1734 solely to indicate this fact.
Received July 12, 2019; revised October 3, 2019; accepted February 7, 2020;published first February 11, 2020.
2014;384:1376–88.2. Bowtell DD, Bohm S, Ahmed AA, Aspuria PJ, Bast RC Jr, Beral V, et al.
Rethinking ovarian cancer II: reducingmortality fromhigh-grade serous ovariancancer. Nat Rev Cancer 2015;15:668–79.
3. Want MY, Lugade AA, Battaglia S, Odunsi K. Nature of tumour rejectionantigens in ovarian cancer. Immunology 2018;155:202–10.
4. Yang SYC, Lheureux S, Karakasis K, Burnier JV, Bruce JP, Clouthier DL,et al. Landscape of genomic alterations in high-grade serous ovariancancer from exceptional long- and short-term survivors. Genome Med2018;10:81.
5. ZhangAW,McPherson A,Milne K, Kroeger DR, Hamilton PT,Miranda A, et al.Interfaces of malignant and immunologic clonal dynamics in ovarian cancer.Cell 2018;173:1755–69.
6. Hamanishi J, Mandai M, Ikeda T, Minami M, Kawaguchi A, Murayama T, et al.Safety and antitumor activity of anti-PD-1 antibody, nivolumab, in patients withplatinum-resistant ovarian cancer. J Clin Oncol 2015;33:4015–22.
7. Hamanishi J, Mandai M, Konishi I. Immune checkpoint inhibition in ovariancancer. Int Immunol 2016;28:339–48.
8. Rodriguez-Garcia A, Minutolo NG, Robinson JM, Powell DJ. T-cell targetantigens across major gynecologic cancers. Gynecol Oncol 2017;145:426–35.
9. Riley RS, June CH, Langer R, Mitchell MJ. Delivery technologies for cancerimmunotherapy. Nat Rev Drug Discov 2019;18:175–96.
10. Ehx GE, Perreault C. Discovery and characterization of actionable tumorantigens. Genome Med 2019;11:1–3.
11. Millar DG, Ohashi PS. Central tolerance: what you see is what you don't get!Nat Immunol 2016;17:115–6.
12. Haen SP, Rammensee HG. The repertoire of human tumor-associated epitopes–identification and selection of antigens and their application in clinical trials.Curr Opin Immunol 2013;25:277–83.
13. Schuster H, Peper JK, Bosmuller HC, Rohle K, Backert L, Bilich T, et al. Theimmunopeptidomic landscape of ovarian carcinomas. Proc Natl Acad Sci U S A2017;114:E9942–E51.
14. Laumont CM, Daouda T, Laverdure JP, Bonneil E, Caron-Lizotte O, HardyMP, et al. Global proteogenomic analysis of human MHC class I-associatedpeptides derived from non-canonical reading frames. Nat Commun 2016;7:10238.
15. Laumont CM,Vincent K,Hesnard L, Audemard E, Bonneil E, Laverdure JP, et al.Noncoding regions are the main source of targetable tumor-specific antigens.Sci Transl Med 2018;10. pii: aau5516.
16. Gotter J, Brors B, Hergenhahn M, Kyewski B. Medullary epithelialcells of the human thymus express a highly diverse selection of tissue-specific genes colocalized in chromosomal clusters. J Exp Med 2004;199:155–66.
17. Bobisse S, Genolet R, Roberti A, Tanyi JL, Racle J, Stevenson BJ, et al. Sensitiveand frequent identification of high avidity neo-epitope specificCD8 (þ) T cells inimmunotherapy-naive ovarian cancer. Nat Commun 2018;9:1092.
18. DenigerDC, PasettoA, Robbins PF,Gartner JJ, Prickett TD, Paria BC, et al. T-cellresponses to TP53 "hotspot" mutations and unique neoantigens expressed byhuman ovarian cancers. Clin Cancer Res 2018;24:5562–73.
19. Liu S, Matsuzaki J, Wei L, Tsuji T, Battaglia S, Hu Q, et al. Efficient identificationof neoantigen-specific T-cell responses in advanced human ovarian cancer.J Immunother Cancer 2019;7:1–17.
20. L€offlerMW,Mohr C, Bichmann L, Freudenmann LK,WalzerM, Schroeder CM,et al. Multi-omics discovery of exome-derived neoantigens in hepatocellularcarcinoma. Genome Med 2019;11:1–16.
21. Pearson H, Daouda T, Granados DP, Durette C, Bonneil E, Courcelles M, et al.MHC class I-associated peptides derive from selective regions of the humangenome. J Clin Invest 2016;126:4690–701.
22. Popovic J, Li LP, Kloetzel PM, LeisegangM, UckertW, Blankenstein T. The onlyproposed T-cell epitope derived from the TEL-AML1 translocation is notnaturally processed. Blood 2011;118:946–54.
23. Szolek A, Schubert B, Mohr C, SturmM, FeldhahnM, Kohlbacher O. OptiType:precision HLA typing from next-generation sequencing data. Bioinformatics2014;30:3310–6.
24. Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illuminasequence data. Bioinformatics 2014;30:2114–20.
25. Daouda T, Perreault C, Lemieux S. pyGeno: a Python package for precisionmedicine and proteogenomics. F1000Res 2016;5:381.
26. Lamoliatte F, McManus FP, MaarifiG, Chelbi-Alix MK, Thibault P. Uncoveringthe SUMOylation and ubiquitylation crosstalk in human cells using sequentialpeptide immunopurification. Nat Commun 2017;8:14109.
27. Lanoix J, Durette C, Courcelles M, Cossette E, Comtois-Marotte S, Hardy MP,et al. Comparison of the MHC I immunopeptidome repertoir of B-cell lympho-blasts using two isolation methods. Proteomics 2018;18:e1700251.
28. Andreatta M, Nielsen M. Gapped sequence alignment using artificial neuralnetworks: application to the MHC class I system. Bioinformatics 2016;32:511–7.
29. Gaidatzis D, Lerch A, Hahne F, Stadler MB. QuasR: quantification and anno-tation of short reads in R. Bioinformatics 2015;31:1130–2.
Zhao et al.
Cancer Immunol Res; 8(4) April 2020 CANCER IMMUNOLOGY RESEARCH554
30. Colaprico A, Silva TC, Olsen C, Garofano L, Cava C, Garolini D, et al.TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGAdata. Nucleic Acids Res 2016;44:e71.
31. Danaher P, Warren S, Dennis L, D'Amico L, White A, Disis ML, et al. Geneexpressionmarkers of tumor infiltrating leukocytes. J Immunother Cancer 2017;5:18.
32. Perez-Riverol Y, Csordas A, Bai J, Bernal-Llinares M, Hewapathirana S,Kundu DJ, et al. The PRIDE database and related tools and resources in2019: improving support for quantification data. Nucleic Acids Res 2019;47:D442–50.
33. ShaoW, Pedrioli PGA,WolskiW, Scurtescu C, Schmid E, Vizcaino JA, et al. TheSysteMHC Atlas project. Nucleic Acids Res 2018;46:D1237–47.
34. Gfeller D, Bassani-Sternberg M. Predicting antigen presentation-what could welearn from a million peptides? Front Immunol 2018;9:1716.
35. Villani AC, Sarkizova S, Hacohen N. Systems immunology: learning the rules ofthe immune system. Annu Rev Immunol 2018;36:813–42.
36. Caron E, Kowalewski DJ, Chiek Koh C, Sturm T, Schuster H, Aebersold R.Analysis of major histocompatibility complex (MHC) immunopeptidomesusing mass spectrometry. Mol Cell Proteomics 2015;14:3105–17.
37. Sansom SN, Shikama-Dorn N, Zhanybekova S, Nusspaumer G, Macaulay IC,Deadman ME, et al. Population and single-cell genomics reveal the Airedependency, relief from Polycomb silencing, and distribution of self-antigenexpression in thymic epithelia. Genome Res 2014;24:1918–31.
38. Jongsma MLM, Guarda G, Spaapen RM. The regulatory network behind MHCclass I expression. Mol Immunol 2019;113:16–21.
39. Bassani-Sternberg M, Pletscher-Frankild S, Jensen LJ, Mann M.Mass spectrom-etry of human leukocyte antigen class I peptidomes reveals strong effects ofprotein abundance and turnover on antigen presentation. Mol Cell Proteomics2015;14:658–73.
40. Bassani-Sternberg M, Braunlein E, Klar R, Engleitner T, Sinitcyn P, AudehmS, et al. Direct identification of clinically relevant neoepitopes presented onnative human melanoma tissue by mass spectrometry. Nat Commun 2016;7:13404.
41. Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, Makarov V, Havel JJ, et al.Cancer immunology. Mutational landscape determines sensitivity to PD-1blockade in non-small cell lung cancer. Science 2015;348:124–8.
42. Gloger A, Ritz D, Fugmann T, Neri D. Mass spectrometric analysis of the HLAclass I peptidome ofmelanoma cell lines as a promising tool for the identificationof putative tumor-associated HLA epitopes. Cancer Immunol Immunother2016;65:1377–93.
43. UniProt Consortium. UniProt: a worldwide hub of protein knowledge.Nucleic Acids Res 2019;47:D506–15.
44. Admon A, Bassani-Sternberg M, inventors; Technion Research and Develop-ment Foundation Ltd., assignee. Soluble HLA complexes for use in diseasediagnosis. United States patent US 20120077696A1. 2012 Mar 29.
45. Cancer Genome Atlas Research Network. Integrated genomic analyses ofovarian carcinoma. Nature 2011;474:609–15.
46. MaiersM, Gragert L, KlitzW.High-resolutionHLA alleles and haplotypes in theUnited States population. Hum Immunol 2007;68:779–88.
47. Rao X, Hoof I, Costa AI, van Baarle D, Kesmir C. HLA class I allele promiscuityrevisited. Immunogenetics 2011;63:691–701.
48. Granados DP, Sriranganadane D, Daouda T, Zieger A, Laumont CM, Caron-Lizotte O, et al. Impact of genomic polymorphisms on the repertoire of humanMHC class I-associated peptides. Nat Commun 2014;5:3600.
49. Delaney JR, Patel CB, Willis KM, Haghighiabyaneh M, Axelrod J, Tancioni I,et al. Haploinsufficiency networks identify targetable patterns of allelic deficiencyin low mutation ovarian cancer. Nat Commun 2017;8:14423.
50. Kahles A, Lehmann KV, Toussaint NC, Huser M, Stark SG, Sachsenberg T, et al.Comprehensive analysis of alternative splicing across tumors from 8,705patients. Cancer Cell 2018;34:211–24.
51. Ali M, Foldvari Z, Giannakopoulou E, Boschen ML, Stronen E, Yang W, et al.Induction of neoantigen-reactive T cells from healthy donors. Nat Protoc 2019;14:1926–43.
52. Croft NP, Smith SA, Pickering J, Sidney J, Peters B, Faridi P, et al. Most viralpeptides displayed by class I MHC on infected cells are immunogenic. Proc NatlAcad Sci U S A 2019;116:3112–7.
AACRJournals.org Cancer Immunol Res; 8(4) April 2020 555
The Global Landscape of Ovarian Cancer–Specific Antigens
2020;8:544-555. Published OnlineFirst February 11, 2020.Cancer Immunol Res Qingchuan Zhao, Jean-Philippe Laverdure, Joël Lanoix, et al. Tumor-Specific Antigens in Ovarian CancerProteogenomics Uncovers a Vast Repertoire of Shared
Updated version
10.1158/2326-6066.CIR-19-0541doi:
Access the most recent version of this article at:
