henhaobio
Sign in

Cell ReportsGenomicsPeer-reviewed journal article

Pan-cancer immunogenomic analyses reveal genotype-immunophenotype relationships and predictors of response to checkpoint blockade

Pornpimol CharoentongiD, Francesca FinotelloiD, Mihaela AngelovaiD, Clemens Mayer, Mirjana EfremovaiD, Dietmar RiederiD, Hubert HackliD, Zlatko TrajanoskiiD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

The Cancer Genome Atlas revealed the genomic landscapes of common human cancers. In parallel, immunotherapy with checkpoint blockers is transforming the treatment of advanced cancers. As only a minority of the patients is responsive to checkpoint blockers, the identification of predictive markers and the mechanisms of resistance is a subject of intense research. To facilitate understanding of the tumor-immune cell interactions, we characterized the intratumoral immune landscapes and the cancer antigenomes from 20 solid cancers, and created The Cancer Immunome Atlas (http://tcia.at). Cellular characterization of the immune infiltrates revealed a role of cancer-germline antigens in spontaneous immunity and showed that tumor genotypes determine immunophenotypes and tumor escape mechanisms. Using machine learning we identified determinants of tumor immunogenicity and developed a scoring scheme for the quantification termed immunophenoscore. The immunophenoscore was superior predictor of response to anti-CTLA-4 and anti-PD-1 antibodies in two independent validation cohorts. Our findings and the developed resource may help informing cancer immunotherapy and facilitate the development of precision immune-oncology.

In-depth summary

Claude Sonnet 4.6 · max effort · 8 Nov 2026 · from the full text of version 3 (1 Nov 2016) · CC BY 4.0

Immunogenomic profiling of 8243 TCGA samples across 20 solid cancers yields an immune atlas and a random-forest-derived immunophenoscore stratifying melanoma responders to checkpoint blockade.

Only a minority of patients respond to checkpoint blockers, and earlier TCGA studies either examined few immune cell types or used a two-gene cytolytic score. This work profiles 28 immune subpopulations alongside neoantigens, cancer-germline antigens and tumor heterogeneity to find predictors of immunogenicity.

Key results

  1. SuggestedA score built from random-forest determinants of cytolytic activity (effector cells, suppressor cells, MHC molecules, checkpoints) separated responders from non-responders in two melanoma cohorts and was reported as superior to single checkpoint-gene expression.

    Random forest used 127 parameters and 10,000 trees per cancer; IPS tested on anti-CTLA-4 (Van Allen) and anti-PD-1 (Hugo) cohorts by ROC with bootstrap CIs. AUC values are not printed in the text. IPS was associated with survival in 12 cancers, four of them significant. · Machine learning identifies major determinants…; Immunophenoscore predicts response to…; Identification of determinants of tumor…; Figure legends

  2. SuggestedTumor genotype shaped immunophenotype and escape routes. BRAF-mutated thyroid tumors showed immunosuppressive infiltrates while RAS-mutated ones downregulated MHC and immunomodulatory molecules; melanoma genotypes also differed.

    THCA BRAF vs RAS differed in immune profiles despite comparable neoantigen burden (p>0.05, Wilcoxon). In SKCM four genotypes differed in neoantigen burden (adjusted pairwise p≤0.002, Dunn's tests). · Genotypes of the tumors determine…

  3. ShownCancer-germline antigen expression tracked with effector and memory T cell infiltration and with mutational load, suggesting T cell responses to these antigens as well as to neoantigens.

    Significant CGA-T cell associations ranged from 3 in STAD to 47 in LIHC; CAGE1 hit 16 of 19 cancers; CGA score vs mutational load r=0.4, p<2.2e−16; 5775 unique CGA peptides predicted to bind HLA. · CGAs are associated with infiltration of CD4+…

  4. ShownThe neoantigen landscape is large but sparsely shared between patients and mostly from passenger genes, arguing against off-the-shelf neoantigen vaccines apart from a few recurrent driver peptides.

    933,954 expressed neoantigens (911,548 unique) from 893,960 mutations; only 24 unique neoantigens shared in at least 5% of patients; 7.6% from driver genes; on average 56% clonal. · Neoantigen landscape is diverse and sparse; DISCUSSION

  5. ShownInfiltrate composition was linked to survival and to mutational load and stage: effector memory CD8+ T cells predicted good prognosis in most cancers, MDSCs and Tregs bad, with a shift to suppression at later stages.

    Tem CD8+ good prognosis in 15 of 19 cancers; MDSCs bad in 13 of 19; Tregs bad in 10 of 19. High-load tumors enriched in activated and Tem T cells; neoantigen load rose slightly by stage (adjusted p<0.032). · Cellular characterization of immune infiltrates…; Mutational load and tissue context determine…; Neoantigen landscape is diverse and sparse

Methods, data and reproducibility

Single-sample GSEA with 782 immune-cell genes estimated 28 TIL subpopulations from TCGA RNA-seq, combined with neoantigen prediction, CGA scoring and ABSOLUTE heterogeneity. A random forest on cytolytic activity gave determinants for the immunophenoscore, tested by ROC in melanoma cohorts.

  • TCGA RNA-seq expression profiles for 20 solid cancers (colon and rectal merged as CRC) · human · n = 8243 samples
  • TCGA clinical information, SNP arrays and paired Affymetrix microarrays · human · n = clinical n=9151; SNP arrays n=3377; microarrays n=550
  • Compendium of sorted immune cell expression profiles used to define immune-related genes (1625, then 782 selected) · n = 37 studies, 366 microarrays
  • Melanoma cohort treated with anti-CTLA-4 antibody (Van Allen et al.), exome and RNA-seq · human
  • Melanoma cohort treated with anti-PD-1 antibody (Hugo et al.), exome and RNA-seq · human
Methods & tools
ssGSEA / GSEA, CIBERSORT, NetMHCpan 2.8 and 3.0, OptiType, ABSOLUTE, HAPSEG, R randomForest, ROCR, DESeq2, Oncotator, ClueGO, R survival
Software & environment
R packages randomForest, ROCR, DESeq2, survival, PMCMR; NetMHCpan 2.8 and 3.0; Oncotator re-annotation; R-based CIBERSORT with default settings; TCIA built on a MEAN stack (MongoDB, ExpressJS, AngularJS, NodeJS).
Experimental conditions
Analysis covered 1 PBytes of primary and processed data, condensed to 50 Gbytes of structured immunogenomic data; no hardware or runtime is stated.
Reproducibility gaps
HLA allele data are available only to researchers with access to TCGA primary sequence data; most package versions and compute hardware are not stated.
Limitations (5)
  • CIBERSORT was optimized for microarray data and lacked several subpopulations (activated and memory CD8+ T cells, MDSCs), so only GSEA results are presented; RNA-seq-to-microarray transformation may underperform for wider-range samples. · stated by the authors
  • Limited data are available for validation, so further studies are needed to confirm the immunophenoscore; the mechanisms behind genotype-immunophenotype links are speculative. · stated by the authors
  • OV (hg18 only) and STAD (SNP arrays unprocessed) were excluded from the clonal versus subclonal neoantigen analysis. · stated by the authors
  • Predictive validation rests on two melanoma cohorts, and AUCs and confidence intervals are not stated in the text, so the size of the advantage over checkpoint-gene expression cannot be checked here. · noted in this summary
  • Cytolytic activity (GZMA and PRF1 mean) serves as the proxy for immunogenicity, so the determinants and score inherit that proxy's biases; associations with survival and genotype are correlational. · noted in this summary

Read this if you want a pan-cancer resource linking tumor-infiltrating immune cells, neoantigens and genotypes, or a gene-expression score for stratifying checkpoint-blockade response.

Citations
5,042
FWCI
67.76
Field percentile
top 1%

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
THISGuo '23Sears '24Keddar '26Chang '23Roelands '19Mukherjee '25
68% semantic matchCancer Biology
Y. A. Guo, T. Kulshrestha, M. M. Chang+7 more·bioRxiv·Jul 30, 2023