Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma

Authors

Willy Hugo

Jesse M. Zaretsky

Lu Sun

Chunying Song

Blanca Homet Moreno

Siwen Hu-Lieskovan

Beata Berent-Maoz

Jia Pang

Bartosz Chmielowski

Grace Cherry

Elizabeth Seja

Shirley Lomeli

Xiangju Kong

Mark C. Kelley

Jeffrey A. Sosman

Douglas B. Johnson

Antoni Ribas

Roger S. Lo

Doi

10.1016/j.cell.2016.02.065

PMID: 26997480 · DOI: 10.1016/j.cell.2016.02.065 · Journal: Cell (2016)

TL;DR

Hugo et al. profiled pretreatment metastatic melanoma biopsies from 38 patients treated with anti-PD-1 antibodies (pembrolizumab or nivolumab) using whole-exome sequencing and matched RNA-seq on a 28-tumor subset. Overall mutational load did not separate responders from non-responders but did associate with longer survival. Responding tumors were enriched for likely loss-of-function mutations in BRCA2 (28% vs 6%, Fisher P=0.002, OR=6.2). Non-responding tumors converged on a co-regulated transcriptional program — Innate anti-PD-1 Resistance (IPRES) — covering mesenchymal transition, ECM remodeling, angiogenesis, hypoxia, and wound healing, and shared substantial overlap with MAPK-inhibitor-induced signatures. The IPRES program defines a transcriptomic subset present across multiple TCGA cancer types (mel_ucla_2016) (PMID:26997480).

Cohort & data

  • Patients: 38 metastatic melanoma patients treated with anti-PD-1 (pembrolizumab or nivolumab); responding n=21, non-responding n=17 (irRECIST criteria; mixed responses excluded). 14 of 38 had prior MAPK-inhibitor therapy.
  • Specimens: 32 pretreatment tumors, 2 pretreatment tumor-derived cultures, 3 early on-treatment tumors without response, 1 early on-treatment tumor with response, plus patient-matched normals.
  • Assays: Whole-exome sequencing on Illumina HiSeq2000 (2×100 bp; median 140× coverage) for all 38 tumors; RNA-seq on a 28-tumor subset with sufficient RNA quality.
  • Dataset: mel_ucla_2016; transcriptome data deposited as GEO accession GSE78220.
  • External cohorts referenced: TCGA cutaneous melanoma metastatic subset n=282 (skcm_tcga_pub_2015); pre-anti-CTLA-4 cohort n=42 (skcm_dfci_2015, Van Allen et al. 2015); pre-MAPKi cohort n=32 (Hugo et al. 2015); 469-tumor melanoma background mutation-rate cohort (Hodis et al. 2012 + TCGA 2015).
  • Variant calling: Oncotator for SNV/INDEL annotation; copy-number from intersection of Sequenza and VarScan2. HLA typing via ATHLATES; neoepitope prediction via NetMHCpan v2.8 / NetMHCIIpan v3.0. Gene set enrichment via GSVA and GSEA using MSigDB C2/C6/C7 (PMID:26997480).

Key findings

  • Mutational load and response: Median non-synonymous SNVs were higher in responding (495) vs non-responding (281) pretreatment tumors but the difference was not significant (P=0.30, Mann-Whitney). HLA class I (231 vs 156, P=0.41) and class II (130 vs 95, P=0.36) predicted neoepitope loads similarly trended higher without statistical significance (PMID:26997480).
  • Mutational load and survival: Patients in the top tertile of mutational load had significantly improved overall survival vs the bottom tertile (log-rank P significant in Figure 1A). Among non-responding/responding subgroups split at within-group median load, responders with low loads still significantly outlived non-responders with high loads (PMID:26997480).
  • BRCA2 enrichment in responders: nsSNVs in BRCA2 occurred in 6 of 21 responders (28%) vs 1 of 17 non-responders (6%); against a 6% melanoma background rate (28/469 tumors), Fisher P=0.002, odds ratio=6.2. Mutations distributed across NPM1-interacting, POLH-interacting, and FANCD2-interacting (helical) domains, consistent with loss-of-function. BRCA2-mutant tumors had significantly higher overall mutational loads than BRCA2-wildtype tumors in this cohort and in two independent melanoma cohorts (PMID:26997480).
  • No predictive copy-number alterations: Genome-wide CNV analysis identified no recurrent alterations exclusive to either responder group (PMID:26997480).
  • Differentially expressed genes: 693 genes met two-fold median difference at nominal Mann-Whitney p≤0.1; FDR p≤0.25 left only ALDH1L2 and MFAP2 (up in non-responders) and CDH1 (up in responders). Up-regulated in non-responders: mesenchymal-transition genes (AXL, ROR2, WNT5A, LOXL2, TWIST2, TAGLN, FAP); immunosuppressive cytokines (IL10, VEGFA, VEGFC); monocyte/macrophage chemoattractants (CCL2, CCL7, CCL8, CCL13); and down-regulation of CDH1 (E-cadherin) (PMID:26997480).
  • No significant differences in checkpoint/effector genes: CD8A/CD8B, PDCD1 (PD-1), CD274 (PD-L1), PDCD1LG2 (PD-L2), CTLA4, LAG3, IFNG, GZMA, and PRF1 did not differ between responders and non-responders. HLA class I gene expression (HLA-A, HLA-B, HLA-C) trended higher in responders but not significantly (PMID:26997480).
  • PTEN: Only one homozygous PTEN deletion (in the non-responding group); no significant overall PTEN expression difference between groups (insufficient power) (PMID:26997480).
  • IPRES signature: A set of 26 co-enriched transcriptomic signatures (mesenchymal transition, angiogenesis, hypoxia, wound healing) were enriched en bloc in 9 of 13 non-responding vs 1 of 15 responding pretreatment tumors. Signatures included MAPK-inhibitor-induced programs derived from melanoma cell lines and patient tumors (PMID:26997480).
  • GO enrichment: Genes up in non-responders enriched for cell adhesion, ECM organization, wound healing, and angiogenesis (FDR-significant); no GO terms were enriched among responder-up genes (PMID:26997480).
  • Cross-cohort validation: IPRES co-enrichment was over-represented in non-responders (Fisher P=0.013, OR=4.6) and under-represented in responders (P=0.04, OR=0.15) within cohort 1. IPRES was NOT differentially distributed between pre-anti-CTLA-4 responders and non-responders in cohort 2 (skcm_dfci_2015), suggesting innate resistance mechanisms differ between anti-PD-1 and anti-CTLA-4 (PMID:26997480).
  • IPRES across cancers: IPRES co-enrichment defines a transcriptomic subset across TCGA pancreatic adenocarcinoma (majority of tumors), lung adenocarcinoma, colon adenocarcinoma, and clear-cell renal cell carcinoma, in addition to melanoma. Within melanoma, only 6 of 69 primary cutaneous tumors showed IPRES co-enrichment vs 90 of 282 metastatic TCGA tumors (skcm_tcga_pub_2015; P=3.9e-5, OR=0.2), consistent with IPRES marking metastasis-associated mesenchymal programs (PMID:26997480).

Genes & alterations

  • BRCA2 — Recurrent likely loss-of-function nsSNVs across NPM1-interacting, POLH-interacting, and FANCD2-interacting helical domains; 28% of anti-PD-1 responders vs 6% of non-responders (Fisher P=0.002, OR=6.2 vs background); BRCA2-mutant melanomas have higher overall mutational loads than wild-type (PMID:26997480).
  • AXL, ROR2, WNT5A, LOXL2, TWIST2, TAGLN, FAP — Mesenchymal-transition transcripts up-regulated in non-responding pretreatment tumors; co-enriched within IPRES signature (PMID:26997480).
  • IL10, VEGFA, VEGFC — Immunosuppressive/angiogenic cytokines up-regulated in non-responders; VEGFA and CCL2 link to a published mouse model of innate anti-PD-1 resistance (Peng et al. 2015) (PMID:26997480).
  • CCL2, CCL7, CCL8, CCL13 — Monocyte/macrophage chemoattractants up-regulated in non-responders (PMID:26997480).
  • CDH1 — E-cadherin down-regulated in non-responders (one of three FDR-significant DEGs at q≤0.25) (PMID:26997480).
  • PTEN — One case of homozygous deletion (non-responder); cohort underpowered to confirm Peng et al. 2015 PTEN-loss / immunotherapy-resistance association (PMID:26997480).
  • PDCD1, CD274, PDCD1LG2, CTLA4, LAG3, IFNG, GZMA, PRF1 — No significant differential expression between responders vs non-responders in this whole-tumor transcriptome cohort (PMID:26997480).
  • BRAF — Status not the primary focus; cited indirectly as a contextual factor (a separate cohort of V600BRAF-mutant melanomas under MAPKi therapy used as cross-validation cohort 3) (PMID:26997480).

Clinical implications

  • Mutational load is prognostic, not predictive of response: High mutational load predicts longer survival on anti-PD-1 but does not reliably distinguish responders from non-responders. Low-load tumors can respond and high-load tumors can fail (PMID:26997480).
  • BRCA2 status as a candidate response biomarker: The 6× enrichment of BRCA2 LOF mutations among anti-PD-1 responders in melanoma supports prospective evaluation of HR-deficiency markers as predictors of immunotherapy benefit (PMID:26997480).
  • IPRES as a candidate resistance biomarker: The transcriptional IPRES program is enriched in innately resistant tumors and detectable across multiple cancer types, suggesting tractable patient stratification by RNA-seq and a target for combination strategies (e.g., anti-angiogenics, EMT modulators, or other interventions targeting wound-healing/ECM programs) (PMID:26997480).
  • MAPKi cross-resistance hypothesis: Because MAPK-inhibitor treatment of melanoma induces transcriptional programs that overlap with IPRES, prior or concurrent MAPKi therapy may compromise subsequent anti-PD-1 response — relevant for sequencing of pembrolizumab / nivolumab with vemurafenib, dabrafenib, or trametinib in BRAF-mutant disease (PMID:26997480).
  • Mechanism distinct from anti-CTLA-4 resistance: IPRES did not distinguish anti-CTLA-4 responders from non-responders in the Van Allen cohort (skcm_dfci_2015), arguing against using a single transcriptomic resistance signature for both checkpoint axes (PMID:26997480).

Limitations & open questions

  • Small cohort (n=38 WES, n=28 RNA-seq): Underpowered for many subgroup comparisons; authors had to bypass multiple-hypothesis correction in single-gene differential-expression analysis to retain enough genes for GO enrichment (PMID:26997480).
  • No recurrent neoepitope or experimentally validated neoantigen: The Snyder et al. tetrapeptide signature for anti-CTLA-4 response did not transfer to anti-PD-1 in this cohort or in the independent Van Allen cohort (PMID:26997480).
  • PTEN-loss / immunotherapy-resistance link could not be tested: Only a single PTEN-deleted case in the cohort (PMID:26997480).
  • Whole-tumor transcriptome may mask immune-cell-restricted signals: Authors note their negative result on interferon-γ signature differs from the Ribas et al. nanostring panel result, possibly because bulk RNA-seq dilutes immune-cell-specific signal (PMID:26997480).
  • Mechanism by which BRCA2 loss promotes anti-PD-1 sensitivity is unresolved: Hypotheses include specific HRD-derived mutational signatures producing more immunogenic neoepitopes vs DNA-damage-induced cell death and immune priming — neither directly tested (PMID:26997480).
  • IPRES is correlative, not causal: The signature predicts non-response but the paper does not demonstrate that pharmacologically suppressing IPRES restores anti-PD-1 sensitivity; authors call this out as the next step (PMID:26997480).
  • Validation in independent prospective cohorts is required before IPRES can be used clinically (PMID:26997480).

Citations from this paper used in the wiki

  • “BRCA2 harbored nsSNVs in six of 21 responding tumors (28%) but only one of 17 non-responding tumors (6%)… Fisher P=0.002, odds ratio=6.2.” (Results, p. 4)
  • “responding tumors are specifically enriched for mutations in the DNA repair gene BRCA2” (Summary)
  • “Innately resistant tumors display a transcriptional signature (referred to as the IPRES or Innate anti-PD-1 Resistance) indicating concurrent upexpression of genes involved in the regulation of mesenchymal transition, cell adhesion, ECM remodeling, angiogenesis and wound-healing.” (Summary)
  • “MAPK-targeted therapy (MAPKi) induces similar signatures in melanoma, suggesting that a non-genomic form of MAPKi resistance mediates cross-resistance to anti-PD-1 therapy.” (Summary)
  • “a group of 26 transcriptomic signatures were co-enriched en bloc in 9 of 13 non-responding versus 1 of 15 responding pre-anti-PD-1 tumors” (Results, p. 6)
  • “this IPRES-enriched transcriptomic subset was over-represented among the anti-PD-1 non-responding pretreatment tumors (Fisher P=0.013, odds ratio=4.6) and under-represented among the responding pretreatment tumors (Fisher P=0.04, odds ratio=0.15)” (Results, p. 7)
  • “median nsSNVs responding=495 and non-responding=281, P=0.30, Mann-Whitney” (Results, p. 3)
  • “the GEO accession number for the transcriptome data is GSE78220” (Accession Number)

This page was processed by crosslinker on 2026-05-14.