henhao
Sign in
Cancer BiologybioRxiv
Preprint

A pan-cancer atlas of therapeutic T cell targets

Guangyuan LiiD, Omar U. Guzmán-BringasiD, Aman Sharma, Maxence O. DellacherieiD, Palak Sekhri, Rachel YaminiD, Dejan Stepec, Maximilien BurqiD, Ioana Clotea, Ethan Tardio, Aswin NatarajaniD, Zachary Harpaz, Xinya Liu, David RequenaiD, Darren Taylor, Beatrix M. UeberheideiD, Michelle KrogsgaardiD, Carolina Colli CruziD, Peter Cimermancic, Mark YarmarkovichiD

Abstract

T-cell-based immunotherapies have revolutionized cancer treatment, yet only a minority of patients are eligible for these approaches, significantly constrained by the limited knowledge of tumor-specific antigens. Here we present ImmunoVerse, a comprehensive map of T cell targets across 21 cancer types, revealing actionable tumor-specific targets in 89% of tumors analyzed. To define the repertoire of actionable T cell targets, we conducted an exhaustive pan-cancer analysis, integrating data from 7,188 RNA-Seq, 1,771 immunopeptidomes from 512 biological samples and 208 single-cell cancer datasets using novel AI methods, and compared these against 17,384 normal samples covering 51 tissues. Our analysis uncovered 62 viable surface protein targets and 28,446 tumor-specific HLA-presented antigens, deriving from 11 distinct molecular events, across 21 tumor types. Among these, we identified 5,928 previously uncharacterized neoantigens, new tumor self-antigens, peptides derived from tumor-specific cryptic ORFs, tumor-associated microbial targets and a novel splicing-derived PMEL peptide (sPMEL) with enhanced abundance and safety compared to the canonical clinical targets. We successfully expanded sPMEL-specific T cells, validating the therapeutic potential of these targets in functional assays. We highlight 153 promising new tumor targets and experimentally validate 19 targets representing six antigen classes. In addition to being the most comprehensive atlas of targets in scope, ImmunoVerse offers the most extensively annotated resource with key parameters for target selection, providing critical insights for therapeutic prioritization and clinical translation. To catalyze therapeutic development, we released our pan-cancer target atlas through an interactive web portal (https://www.immuno-verse.com) and made the accompanying toolkits available to the scientific community. This work redefines the landscape of therapeutic T cell targets and provides a foundational resource to unlock immunotherapy development across multiple cancers previously considered intractable.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
5
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (7 Jul 2025) · CC BY 4.0

ImmunoVerse integrates RNA-Seq and immunopeptidomes to catalog 28,446 tumor-specific HLA-presented antigens across 21 cancers, including a splicing-derived PMEL peptide with T cell expansion support.

Over 50% of MS/MS spectra remain unannotated against the canonical proteome, and most public immunopeptidomes lack matched RNA-Seq, which limits discovery of tumor-specific HLA peptides. ImmunoVerse builds histology-matched search spaces for 11 aberration classes from TCGA, TARGET and GTEx, then searches 1,771 immunopeptidomes.

Key results

  1. ShownA pan-cancer atlas of tumor-specific HLA-presented peptides (pHLAs) spans 11 molecular aberration classes and reports actionable tumor-specific targets in 89% of tumors analyzed.

    28,446 tumor-specific pHLAs across 21 tumor types. Inputs: 7,188 RNA-Seq, 1,771 immunopeptidomes from 512 biological samples, 208 single-cell datasets. Normal reference: 17,384 samples covering 51 tissues. · Abstract; ImmunoVerse: A multimodal approach to establish…; Discussion

  2. ShownRescoring and deep-learning spectral search raised peptide identification over standard MaxQuant searching, so the atlas draws on three search engines combined with netMHCpan filtering.

    Versus the MaxQuant baseline: MS2Rescore raised identifications by 30% on average and Tesorai by 60% on average. Restricted to high-confidence PSMs (Andromeda above 70, Tesorai above 5), Tesorai raised them by 136% on average. · ImmunoVerse: A multimodal approach to establish…

  3. ShownImmunopeptidome-evidenced mutation neoantigens are numerous but overlap little with prediction-only databases, supporting empirical evidence for vaccine and TIL (tumor-infiltrating lymphocyte) target selection.

    6,403 mutation-derived neoantigens with direct evidence; SKCM 2,624, colon 1,117, lung adenocarcinoma 823; missense 87%, frameshift 11.7%. Only 475 TSNAdb predicted neoantigens had evidence in this cohort. · ImmunoVerse reveals numerous…

  4. ShownA splicing-derived PMEL peptide (sPMEL, exon 5 skipping) is presented more abundantly than canonical PMEL epitopes and was not cross-recognized by two canonical-PMEL TCR clones. Expanded T cells responded to it and to PMEL-positive tumor cells in vitro.

    Higher abundance than KTWGQYWQV, ITDQVPFSV and YLEPGPVTA despite lower transcript abundance; two TCR clones did not cross-react; 52.9% antigen-specific cytotoxic T cells after 3 rounds; activation by HLA-A*02:01 PMEL-positive SKMEL5 cells. · Cryptic ORFs and alternative splicing product…; Cross-reactivity TCR experiment; Spliced PMEL peptide TCR functional assays

  5. SuggestedTumor-resident viral and bacterial sequences yield recurrent HLA-presented peptides, including CMV and N. circulans antigens. The authors cannot exclude N. circulans as a sequencing contaminant.

    187 CMV-derived antigens across six tumor types; N. circulans in 98% of ovarian RNA-Seq samples with 555 unique peptides from 518 genes; 14 peptides also in esophageal cancer; none of the 14 found in eight neuroblastoma cell lines lacking N. circulans reads. · Tumor resident pathogens provide an unexpectedly…; N. circulans Alignment Validation

Methods, data and reproducibility

Tumor-specific events are called from TCGA/TARGET RNA-Seq against GTEx normals (cryptic ORFs, lncRNAs and pseudogenes come from the nuORF database), translated into histology-matched peptide search spaces, searched in public immunopeptidomes with MaxQuant, MS2Rescore and Tesorai, then filtered by netMHCpan 4.1 binding.

  • Tumor bulk RNA-Seq from TCGA and TARGET across 21 adult and pediatric tumor types · human · n = 7,188 RNA-Seq samples · phs000178.v11; phs000218.v26
  • GTEx normal tissue RNA-Seq used to exclude non-tumor-specific events (51 tissues) · human (GTEx) · n = up to 17,384 normal samples · phs000424.v6.p1
  • Public tumor immunopeptidomes curated from PRIDE and MassIVE · human · n = 1,771 immunopeptidomes from 512 biological samples
  • Normal tissue immunopeptidomes from the HLA Ligand Atlas for safety screening · human · PXD019643
  • Tumor single-cell RNA-Seq from CancerSCEM for target homogeneity (14 cancer types) · human · n = 208 scRNA-Seq datasets overall; 14 cancer types used for homogeneity
  • Ribo-Seq of neuroblastoma patient-derived xenograft models to test cryptic ORF translation · human (patient-derived xenografts) · n = 8 models
  • One high-TMB melanoma patient (Hartwig cohort) WGS and RNA-Seq, searched against public melanoma immunopeptidomes · human · n = 1 patient; 267 melanoma immunopeptidomes
  • Synthetic peptide MS/MS comparison and refolding for candidate targets across antigen classes · n = Twenty peptide candidates
Methods & tools
MaxQuant 2.4.9.0, MS2Rescore, Tesorai Search, netMHCpan 4.1, Kallisto 0.44.0, STAR 2-pass, StringTie2 2.1.3, Kraken2 (PlusPF database), STAR-Fusion 1.13.0, AlphaFold2 2.3.0, RibORF 2.0, Scanpy
Code
https://github.com/frankligy/pan_cancer_intracellular_antigen_atlashttps://frankligy.github.io/pan_cancer_intracellular_antigen_atlas/notes/deployed_pipeline_latest.pdf
Data & accessions
https://www.immuno-verse.com/https://ngdc.cncb.ac.cn/cancerscem/
Software & environment
Tool versions stated include MaxQuant 2.4.9.0, Kallisto 0.44.0, StringTie2 2.1.3, STAR-Fusion 1.13.0 (docker container), AlphaFold2 2.3.0, RibORF 2.0 and netMHCpan 4.1; the pipeline is hosted on the Cancer Genomics Cloud.
Experimental conditions
The personalized vaccine workflow is reported to take under four hours; no hardware specification is given.
Reproducibility gaps
Raw TCGA, TARGET and GTEx RNA-Seq require dbGaP approval. The authors note filed patents on the pipeline and targets. Per-experiment compute and donor numbers for T cell assays are not reported.
Limitations (6)
  • Immunopeptidome availability varies by cancer type: stomach (2 samples) and cholangiocarcinoma (5 samples) have sparse data and low HLA coverage (58%, 26%), so antigen yields there are low. · stated by the authors
  • The atlas relies on bulk sequencing, so stromal or immune contamination can affect signals and intratumoral heterogeneity is not captured; search space is defined per histology, not per subtype. · stated by the authors
  • N. circulans may be a contaminant introduced during sequencing rather than a resident pathogen, and the tumor-resident microbiome remains contentious. · stated by the authors
  • Circular RNAs, RNA editing, proteasomal splicing, PTMs and alternative codon-derived antigens were not analyzed, and the analysis is limited to HLA class I. · stated by the authors
  • Searches use a 5% peptide-level FDR for MaxQuant and MS2Rescore results; this raises false-positive risk, mitigated only by netMHCpan filtering and spot checks with synthetic peptides. · noted in this summary
  • Functional T cell evidence for sPMEL is shown for one antigen with few reported replicates or donors, so generalization of the 19 experimentally validated targets to therapeutic efficacy is untested. · noted in this summary

Who should read:Read this if you build or prioritize T cell, CAR-T, TCR or neoantigen vaccine targets and want empirically supported pan-cancer immunopeptidome evidence.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISZheng '26Bricard '26Madan '21Shraim '25Tanis '26Qin '25
68% semantic matchBioinformatics
Tao Zheng, Zhi Ran, Jing Zeng+10 more·bioRxiv·Oct 16, 2026