Integrative Clinical Genomics of Metastatic Cancer

Authors

Dan R. Robinson

Yi-Mi Wu

Robert J. Lonigro

Pankaj Vats

Erin Cobain

Jessica Everett

Xuhong Cao

Erica Rabban

Chandan Kumar-Sinha

Victoria Raymond

Scott Schuetze

Ajjai Alva

Javed Siddiqui

Rashmi Chugh

Francis Worden

Mark M. Zalupski

Jeffrey Innis

Rajen J. Mody

Scott A. Tomlins

David Lucas

Laurence H. Baker

Nithya Ramnath

Ann F. Schott

Daniel F. Hayes

Joseph Vijai

Kenneth Offit

Elena M. Stoffel

J. Scott Roberts

David C. Smith

Lakshmi P. Kunju

Moshe Talpaz

Marcin Cieslik

Arul M. Chinnaiyan

Doi

10.1038/nature23306

PMID: 28783718 · DOI: 10.1038/nature23306 · Journal: Nature (2017)

TL;DR

Robinson and colleagues performed paired tumor/normal whole-exome and transcriptome sequencing on 500 adult patients with metastatic solid tumors of diverse lineage and biopsy site (the “MET500” cohort) under the Michigan Oncology Sequencing (Mi-Oncoseq) program. They cataloged the somatic landscape (most frequently altered: TP53, CDKN2A, PTEN, PIK3CA, RB1), identified putative pathogenic germline variants in 12.2% of cases (75% in DNA-repair genes), used RNA-seq to call gene fusions and define proliferative vs. EMT-like transcriptional subtypes, and integrated DNA + RNA + TCR-seq to characterize tumor-immune phenotypes relevant to checkpoint immunotherapy.

Cohort & data

  • 500 adult patients with metastatic solid tumors (MET500 cohort) from 537 biopsies / 556 enrolled (93% complete-sequencing success rate); cBioPortal study metastatic_solid_tumors_mich_2017.
  • Cancer types (n=20 represented). Top three: 93 (18.6%) metastatic prostate cancer (PRAD), 91 (18.2%) metastatic breast cancer (BRCA), 42 (8.4%) soft-tissue sarcoma (SARCNOS); also 25 (5%) carcinoma of unknown primary (CUP). Cohort overall is best represented as MIXED lineage.
  • Biopsy sites (>30 organs sampled): 134 liver, 114 lymph node, 46 lung, 42 bone, 32 abdominal mass/ascites/pleural fluid.
  • Demographics: 258 male / 242 female; 460 (92%) Caucasian, 40 (8%) non-Caucasian; median age 59 (range 18–86).
  • Assays: paired tumor/normal whole-exome sequencing (whole-exome-seq) on Agilent SureSelect Human All Exon v4 (mean target coverage 180X tumor / 120X normal; mean tumor content 62%); poly(A)+ and exome-capture transcriptomes (rna-seq) on Illumina HiSeq 2000/2500 (40–50M paired reads); T-cell receptor β CDR3 deep sequencing (tcr-seq) via Adaptive Biotechnologies immunoSEQ.
  • Bioinformatics: alignment via Novoalign to GRCh37/hg19; SNV calling varscan (VarScan2 v2.3.2); indels pindel; annotation annovar; RNA alignment star (STAR_2.4.0g1) with custom CRISP pipeline and CODAC fusion caller.
  • Sequencing data deposited in dbGaP (phs000673.v2.p1); MET500 web portal at met500.path.med.umich.edu.

Key findings

  • Higher mutation burden in metastases than in matched primary cohorts. Mean of 119 somatic mutations per patient within targeted regions; for most cancer types mutation burden was significantly elevated relative to TCGA primaries (Fig. 1c), with the largest increase in tumor types that have low primary-stage mutation burden (e.g., prostate, adrenal).
  • Long-tailed driver landscape. Most-altered tumor suppressors: TP53 266 (53.2%), CDKN2A 80 (16%), PTEN 79 (15.8%), RB1 68 (13.6%). Most-altered oncogenes: PIK3CA 67 (13.4%), AR 63 (12.6%), KRAS 51 (10.2%). Tumor suppressors were broadly altered across lineages while oncogenes were more lineage-restricted (e.g., AR in PRAD, GNAS in selected types).
  • Pathogenic germline variants in 12.2% of patients. 63 presumed pathogenic germline mutations (PPGM) in 18 genes affecting 61 individuals (12.2%). Composition: 30 deleterious missense, 8 nonsense, 20 frameshift, 5 deleterious splice. 75% were in DNA-repair genes; the most common were MUTYH (n=10, 16%), BRCA2 (n=9, 14%), CHEK2 (n=9, 14%), BRCA1 (n=5, 8%). Outside DNA repair: APC (n=6, 9.5%), MITF (n=5, 8%), HOXB13 (n=3, 5%).
  • Germline second-hit somatic events. 30/61 (49%) PPGM carriers had a somatic second-allele aberration (including LOH) consistent with biallelic inactivation; remaining cases were carrier-status only.
  • Enrichment vs. ExAC. PPGM odds were significantly elevated vs. 52,790 ExAC controls (excluding TCGA): OR = 3.00 (95% CI 2.28–3.9; P = 1 × 10⁻¹³). Genes individually enriched: BRCA1, BRCA2, APC, CHEK2, MITF, MLH1, NBN, RB1.
  • Pervasive gene fusion landscape. 868 transcriptome libraries from 496 tumors; 12,027 unique fused gene pairs (mean 34/tumor). 199 cases (39.8%) carried at least one putative pathogenic fusion: 138 activating (DNA-binding n=88, protein kinases n=29, signal transducers n=21) and 103 deleterious (canonical tumor suppressors n=59, chromatin modifiers n=35, cell adhesion n=9). Most-fused tumor suppressors: NF1 (n=18), TP53 (n=11), PTEN (n=11), RB1 (n=6).
  • Novel pathogenic fusion partners. Reported 8 novel fusions extending the partner range for clinically targetable kinase fusions, including activating ALK, BRAF, and FGFR fusions with novel partners. New functional-domain fusions include GREB1–NR4A3 in uterine leiomyosarcoma (ULMS), POC5–PRKD1 in polymorphous low-grade adenocarcinoma of the tongue, and CIC–CITED1 in undifferentiated high-grade sarcoma. NOTCH2 fusions split into γ-secretase-sensitive (NOTCH2–SPAG17) and γ-secretase-independent (PARS2–NOTCH2) classes.
  • Two transcriptional metastatic subtypes. Unsupervised clustering of 25 non-redundant meta-signatures across 496 transcriptomes identified four canonical hallmarks (immune response, EMT, proliferation, metabolism) and split metastases into (i) an EMT-like subtype with inflammation signatures and (ii) a proliferative subtype with increased metabolism and stress response; mutual exclusivity between proliferative and EMT gene sets was stronger in metastases than in primary tumors. Subtype assignments were only weakly tied to biopsy site or primary tissue.
  • Dedifferentiation with progression. t-SNE of expression of tissue-specific markers (Nt36) showed metastatic samples were less well-separated and more heterogeneous than primary tumors; only liver biopsies segregated by site. Metastatic samples showed significant tissue-marker dedifferentiation vs. normal/primary.
  • Immune microenvironment is tumor-type-driven. A novel RNA-seq score, MImmScore (derived from 141 ESTIMATE immune-signature genes via inverse-normal-transform), agreed with canonical T-cell expression markers and DNA-based TCR-β CDR3 sequencing. MImmScores correlated negatively with tumor content and positively with stromal infiltration. Cancer types known to be infiltrated in the localized stage (kidney, lung, melanoma) remained infiltrated at metastatic sites; less immunogenic types (breast, prostate) had low MImmScores in both stages.
  • TIL composition heterogeneity. Bulk-transcriptome deconvolution revealed clusters typified by M2:M0 macrophage ratios and CD8+:CD4+ T-cell ratios; M1 macrophages were largely absent while immunosuppressive M2 macrophages were highly prevalent. A small TIL-5 cluster was dominated by cytotoxic CD8+ T-cells. In high-MImmScore index cases, TCR-β CDR3 deep sequencing confirmed expansion of dominant T-cell clones (increased clonality) and lower Treg:cytotoxic ratios.
  • Mutation burden tracks T-cell infiltration. Highly mutated samples were associated with larger numbers of infiltrating T-cells and higher MImmScores.
  • Co-stimulatory ligand/receptor structure. Almost all APC-1 cluster samples expressed CD80/CD86 and almost all Tcell-1 cluster samples expressed CD28, with significant overlap between independent APC and T-cell clusters.
  • Composite immune-activity stratification predicts checkpoint biomarkers. Patients categorized as “fully active” (member of all of TIL-5, APC-1, Tcell-1) or “partially active” had elevated PD-L1, HLA, and granzyme expression and higher mutational burden, plus higher clinical-response scores derived from a published metastatic-melanoma immunotherapy signature.

Genes & alterations

  • TP53 — most frequently altered tumor suppressor (266/500, 53.2%); also recurrent in fusions (n=11) as a tumor-suppressor breakage event.
  • CDKN2A — second most frequently altered tumor suppressor (80/500, 16%).
  • PTEN — altered in 79/500 (15.8%); also recurrently disrupted by fusions (n=11).
  • RB1 — altered in 68/500 (13.6%); fusion-disrupted in n=6; germline PPGM enrichment.
  • PIK3CA — most frequently altered oncogene (67/500, 13.4%).
  • AR — altered in 63/500 (12.6%); strongly tied to PRAD.
  • KRAS — altered in 51/500 (10.2%).
  • GNAS — lineage-associated oncogene.
  • NF1 — most commonly fused tumor suppressor (n=18).
  • BRCA1, BRCA2 — DNA-repair germline PPGMs significantly enriched vs. ExAC; 5 and 9 PPGM carriers respectively, with implications for PARP inhibitor sensitivity.
  • MUTYH — most common DNA-repair PPGM gene (n=10, 16% of PPGMs); reclassified as a tumor suppressor in this study based on aberration profile.
  • CHEK2, MLH1, NBN — additional DNA-repair PPGMs enriched in MET500 vs. ExAC.
  • APC — non-DNA-repair PPGM in 6 (9.5%) carriers; significantly enriched.
  • MITF — non-DNA-repair PPGM in 5 (8%); enriched vs. ExAC.
  • HOXB13 — non-DNA-repair PPGM in 3 (5%) carriers (prostate-cancer susceptibility allele).
  • ALK, BRAF — activating fusions identified with novel partners, expanding the targetable fusion landscape.
  • GREB1–NR4A3 — novel activating fusion in uterine leiomyosarcoma (ULMS).
  • POC5–PRKD1 — novel fusion in polymorphous low-grade adenocarcinoma of the tongue.
  • CIC–CITED1 — novel fusion in undifferentiated high-grade sarcoma.
  • NOTCH2–SPAG17 (γ-secretase sensitive) and PARS2–NOTCH2 (γ-secretase independent) — two functionally distinct NOTCH fusion classes with differential drug-sensitivity predictions.

Clinical implications

  • Genetic counseling for metastatic patients. Given 12.2% prevalence of presumed pathogenic germline variants — three-fold over ExAC controls — the authors argue metastatic-cancer patients should be considered for genetic counseling and germline testing.
  • PARP-inhibitor opportunity. 75% of germline pathogenic variants were in DNA-repair pathways; HR deficiency suggests PARP inhibitor sensitivity in carriers (BRCA1/BRCA2, etc.).
  • Checkpoint immunotherapy candidacy. Hypermutated tumors may respond to immune checkpoint inhibitors. Composite immune-activity classification (TIL-5 + APC-1 + Tcell-1 cluster membership) correlated with PD-L1, HLA, granzyme expression, mutational burden, and a melanoma-derived clinical-response score; the authors propose RNA-seq-based monitoring as a clinical tool for stratifying immunotherapy candidates.
  • Identifying immune-active patients within “cold” tumor types. High-MImmScore individuals were detectable even within tumor types that have not historically responded to immunotherapy, suggesting per-patient immunogenomic profiling may surface unexpected candidates.
  • Targetable fusions in unexpected lineages. Novel activating ALK, BRAF, and FGFR fusion partners broaden the population eligible for kinase-inhibitor trials beyond canonical fusion contexts.
  • NOTCH-fusion subclassification informs gamma-secretase inhibitor selection. NOTCH2–SPAG17-class fusions are predicted γ-secretase-inhibitor-sensitive, while PARS2–NOTCH2-class fusions are not — guiding therapy choice.
  • Real-time metastatic biopsy preferred over archival primary. Because metastases acquire mutations during evolution and treatment, the authors argue clinical decisions should be informed by current metastatic-tumor sequencing rather than archival primary material.

Limitations & open questions

  • No matched primary–metastasis pairs. The study compares a metastatic cohort to separate primary cohorts (TCGA) rather than matched primary/metastatic biopsies from the same patients, limiting inference about within-patient evolution, resistance, and immune dynamics.
  • Cohort demographics are skewed. 92% Caucasian; 93.6% recruited at a single center (University of Michigan), tempering generalizability across populations and care settings.
  • Single-region biopsies. Each metastasis is sampled at one site/time, missing intratumoral and intermetastatic heterogeneity.
  • CODAC fusion-calling pipeline is custom and unpublished at the time (“manuscript in preparation”), making external benchmarking difficult, although 56/57 randomly selected fusion candidates validated by RT-PCR + Sanger (98.2%).
  • Bulk RNA-seq immune deconvolution. TIL- and APC-cluster definitions rely on bulk-transcriptome deconvolution rather than single-cell measurements; clinical-response correlations leverage a melanoma-derived signature whose generalizability across the MIXED metastatic cohort is untested.
  • MImmScore is correlative, not yet prospectively validated as a treatment-decision tool.
  • The cBioPortal study metastatic_solid_tumors_mich_2017 uses the GRCh37/hg19 reference (paper sequencing data); RNA was processed against GRCh38 with a custom transcript reference (MOTR/Gencode V23).

Citations from this paper used in the wiki

  • “we perform whole exome and transcriptome sequencing of 500 adult patients with metastatic solid tumors of diverse lineage and biopsy site” (Summary).
  • “The most prevalent genes somatically altered in metastatic cancer included TP53, CDKN2A, PTEN, PIK3CA, and RB1” (Summary).
  • “Putative pathogenic germline variants were present in 12.2% of cases of which 75% were related to defects in DNA repair” (Summary).
  • “TP53 (266, 53.2%), CDKN2A (80, 16%), PTEN (79, 15.8%) and RB1 (68, 13.6%), were the most frequently altered tumor suppressors, while the most frequently mutated oncogenes included PIK3CA (67, 13.4%), AR (63, 12.6%) and KRAS (51, 10.2%)” (p. 3, “The landscape of molecular aberrations”).
  • “MUTYH (n=10, 16%), BRCA2 (n=9, 14%), CHEK2 (n=9, 14%), and BRCA1 (n=5, 8%) the most common” PPGMs (p. 4, “Germline variants in metastatic cancer”).
  • “The odds of any PPGM in metastatic cancer significantly exceeded the odds found in the populations comprising ExAC (OR = 3.00, 2.28–3.9, P=1 × 10−13)” (p. 4).
  • “199 cases (39.8%) harbored at least one putative pathogenic fusion with 138 activating fusions and 103 deleterious fusions” (p. 4, “The gene fusion landscape”).
  • “Novel gene fusions with functional domains include GREB1-NR4A3 in uterine leiomyosarcoma, POC5-PRKD1 in polymorphous low-grade adenocarcinoma of the tongue, and CIC-CITED1 in undifferentiated high-grade sarcoma” (p. 4–5).
  • “Notch fusions fall into 2 classes, those predicted to be sensitive to gamma-secretase inhibition (e.g., NOTCH2-SPAG17) and fusions which are independent of gamma-secretase processing (e.g., PARS2-NOTCH2)” (p. 5).
  • “Metastatic tumors fall into two main subtypes: an EMT-like subtype associated with inflammation signatures, and a proliferative subtype associated with increased metabolism and systemic stress” (p. 5, “Transcriptional signatures”).
  • “hypermutated tumors may respond to immune checkpoint inhibitors, while HR deficiency could suggest sensitivity to PARP inhibitors” (p. 7, “Conclusion”).
  • “While this study compares the molecular attributes of a metastatic cancer cohort to those of primary cancer cohorts, it does not utilize matched samples of primary and metastatic biopsies from individual cases” (p. 7–8, “Conclusion”).

This page was processed by crosslinker on 2026-05-15.