Comprehensive and Integrated Genomic Characterization of Adult Soft Tissue Sarcomas

Authors

The Cancer Genome Atlas Research Network

Elizabeth G. Demicco

Li Ding

Marc Ladanyi

Alexander J. Lazar

Samuel Singer

Doi

10.1016/j.cell.2017.10.014

PMID: 29100075 · DOI: 10.1016/j.cell.2017.10.014 · Journal: Cell (2017)

TL;DR

The TCGA Sarcoma Analysis Working Group profiled 206 adult soft tissue sarcomas spanning 6 major histologies — leiomyosarcoma (LMS, n=80; 53 STLMS + 27 ULMS), dedifferentiated liposarcoma (DDLPS, n=50), undifferentiated pleomorphic sarcoma (UPS, n=44), myxofibrosarcoma (MFS, n=17), synovial sarcoma (SS, n=10) and malignant peripheral nerve sheath tumor (MPNST, n=5) — using WES/WGS, RNA-seq, miRNA-seq, DNA methylation, SNP6 copy number, and RPPA. Three pan-sarcoma findings emerge: (1) complex-karyotype sarcomas are dominated by somatic copy-number alterations (SCNAs), with very low mutation burden (mean 1.06 mut/Mb) and only 3 significantly mutated genes pan-cohort (TP53, ATRX, RB1); (2) within histologies, integrated SCNA + methylation clustering defines prognostically distinct molecular subtypes (e.g., DDLPS K1/K2 vs K3; STLMS iCluster C1 vs C2); and (3) immune-microenvironment signatures inferred from mRNA and methylation associate with disease-specific survival (DSS), nominating MFS alongside UPS for checkpoint-inhibitor trials.

Cohort & data

Key findings

  • Pan-sarcoma SMGs are rare and tumor-suppressor-dominant. MuSiC identified only 3 significantly mutated genes (FDR<0.05): TP53, ATRX, and RB1. TP53 was mutated in 40/80 LMS (50%); RB1 mutations occurred in LMS, UPS, and MFS. 67% (138/206) carried at least one variant in a known cancer gene, but few hit known hotspots. TextRef PMID:29100075
  • Sarcomas are SCNA-driven with low mutation burden. Mean mutation rate was 1.06/Mb. DDLPS had the highest SCNA frequency of any TCGA tumor type, driven by recurrent 12q13~15 amplification; SS had the fewest SCNAs and mutations.
  • Mutational signatures are clock-like. 90% of mutations attributable to COSMIC5 (53%) and COSMIC1 (37%); APOBEC signatures (COSMIC2/13) modestly elevated in DDLPS and MPNST (p<10⁻⁶, Kruskal-Wallis). Two hypermutators showed COSMIC6 mismatch-repair signature with MSH6 frameshift and low MSH2 expression.
  • DDLPS subtypes by SCNA + methylation predict DSS. SCNA clusters: K1 (JUN-amplified), K2 (TERT-amplified, chromosomally unstable), K3 (6q25.1-amplified). Combined with methylation (Meth1 hypo / Meth2 hyper), partition into K3 vs K1+K2-hypomethylated vs K1+K2-hypermethylated yields 3 groups with significantly different DSS (p=0.001). Meth2 alone HR=4.4, p=0.002 vs Meth1.
  • DDLPS recurrent alterations. MDM2 gain/amp 100%, CDK4 92%, HMGA2 76%, FRS2 96%, NAV3 60%; adipocyte-differentiation suppressors amplified — JUN 42%, DDIT3 32%, PTPRQ 46%, YAP1 16%, CEBPA 24%. Recurrent deletions: ATRX (10% deep / 20% shallow), NF1 (6%/22%), CDKN2A (2%/42%). JUN and PTPRQ amplifications were nearly mutually exclusive (p=0.026).
  • LMS molecular split between ULMS and STLMS. TP53 deletions in 9% (deep) / 60% (shallow) and mutations in 50%; RB1 deletions 14%/78%, mutations 15%; PTEN deletions 13%/68%, mutations 5%. iCluster split LMS by site (ULMS vs STLMS); within STLMS, C1 (hypermethylated, IGF1R, CCNE2, MCM2, FANCI up; RB1 mutations enriched p=0.04; 17p11.2-p12 amplification including MYOCD in 40%) had worse RFS (p=0.0002) and DSS (p=0.008) than C2.
  • AKT-pathway alterations dominate LMS. 84% of ULMS + STLMS C1 vs 44% of STLMS C2 carried PTEN/AKT3/IGF1R/RICTOR/MTOR pathway alterations (p=1e–04).
  • miR-181b-5p is an independent RFS predictor in LMS. Univariate HR 8.03, adjusted p<0.0001; multivariate HR 7.4 (95% CI 3.1–17.8, p=9e–6) controlling for LMS subtype and tumor size. High miR-181b was paradoxically associated with low AKT3 and MTOR expression (p<0.006).
  • UPS and MFS form a single molecular spectrum. Largely indistinguishable across platforms; clustering on myxoid-stroma-associated genes is what segregates them. Recurrent SCNAs across UPS/MFS combined: CCNE1 high-level amplification 10%, VGLL3 11%, YAP1 3%; VGLL3/YAP1 target signature strongly expressed (p=1e–24), implicating the Hippo pathway.
  • SS is genomically distinct. All 10 SS cases carried SS18–SSX1 or SS18–SSX2 fusions; iCluster placed all SS in C4 (high FGFR3, miR-183, PDE4A promoter methylation; partial/complete 3p loss in 5/10).
  • TRIO–TERT fusions identified. Recurrent TRIO–TERT fusions (n=3) plus other TRIO fusions (n=2); TRIO–TERT fusion cases had the highest TERT expression in the cohort.
  • Computational morphometrics ↔︎ genomic complexity. Nuclear pleomorphism score correlated with whole-genome doublings (p=0.003), subclonal genome fraction (p=4e–6), and aneuploidy (p=5e–6).
  • Immune microenvironment is histology-specific and prognostic. UPS/MFS and DDLPS had highest macrophage scores; DDLPS highest CD8; STLMS highest PD-L1 (significantly higher than ULMS, p=4e–5). PD-L1 mRNA correlated with CD274 copy number (r=0.42, adj p=4e–10). NK-cell infiltration was the only signature correlated with DSS across multiple histologies; in UPS/MFS, dendritic-cell signatures predicted improved DSS; in DDLPS, elevated Th2 signature predicted shorter DSS.

Genes & alterations

  • TP53 — pan-sarcoma SMG; mutated in 50% of LMS (40/80); deep deletions in 9% LMS, 16% UPS, 12% MFS.
  • ATRX — pan-sarcoma SMG; deletions/mutations in ~30% of DDLPS; loss associated with telomere lengthening in UPS/MFS (p=0.013); proposed correlative biomarker for CDK4 inhibitor trials in DDLPS.
  • RB1 — pan-sarcoma SMG; deep deletions in 14% LMS, 16% UPS, 24% MFS; enriched mutations in STLMS iCluster C1 (p=0.04).
  • MDM2, CDK4, HMGA2, FRS2, NAV3 — defining 12q13~15 amplifications of DDLPS (100%, 92%, 76%, 96%, 60%).
  • JUN, DDIT3, PTPRQ, YAP1, CEBPA — adipocyte-differentiation-inhibitor amplifications in DDLPS (42%/32%/46%/16%/24%); JUN amplification defines poor-prognosis K1 cluster.
  • CDKN2A — deep deletions in 8% LMS, 20% UPS, 18% MFS; shallow in 42% DDLPS.
  • NF1, NF2, PRKDC — truncating mutations (n=3, 1, 4 respectively); PRKDC implicated in telomere stabilization / double-strand-break repair.
  • TERT — amplification defines DDLPS K2 cluster; recurrent TRIO–TERT fusions (n=3) drive highest TERT expression.
  • SS18, SSX1, SSX2 — t(X;18) fusions present in 100% of SS.
  • PTEN, AKT3, MTOR, IGF1R, RICTOR — PI3K/AKT/MTOR pathway alterations in 84% of ULMS+STLMS C1 vs 44% of STLMS C2; PTEN deletions 81% (combined deep+shallow), mutations 5%; nominates dual PI3K/MTOR or TORC1/TORC2 inhibitors.
  • MYOCD — high-level amplification at 17p11.2-p12 in 40% of STLMS C1 (smooth-muscle differentiation TF).
  • CCNE1, VGLL3, YAP1 — focal amplifications in 10%/11%/3% of UPS/MFS; Hippo-pathway TEAD-cofactor target signature strongly expressed.
  • CCNE2, MCM2, FANCI — overexpressed in poor-prognosis STLMS C1 (cell cycle / DNA replication / DNA repair).
  • FGFR3 — high expression characteristic of SS iCluster C4 (p=7e–20).
  • CD274 (PD-L1) — copy-number variation correlates with mRNA (r=0.42); STLMS has highest PD-L1 expression among sarcoma types.
  • ESR1 — target genes hypomethylated in ULMS (vs STLMS), supporting hormonal-axis differences.
  • MSH6, MSH2 — frameshift / low expression respectively in the two cohort hypermutators showing COSMIC6 MMR signature.
  • MYLK, MYH11 — myogenic-differentiation markers highly expressed in LMS-dominant iCluster C1 (p<5e–39).

Clinical implications

  • Refined risk stratification in DDLPS. Combined SCNA + methylation subtype (K3 vs K1+K2 / Meth1 vs Meth2) stratifies DSS (p=0.001) and could inform adjuvant-therapy intensity in future trials. Hypermethylated K1+K2 has worst outcome and lowest immature-DC infiltration (p=0.004).
  • Refined risk stratification in STLMS. iCluster C1 vs C2 splits STLMS into worse-RFS (p=0.0002) and worse-DSS (p=0.008) subtypes; C1 features MYOCD amplification, hypermethylation, and PI3K/AKT activation.
  • CDK4-inhibitor biomarker hypothesis. ATRX deletions in ~30% of DDLPS may represent a correlative biomarker for response to CDK4 inhibitors (citing Kovatcheva et al., 2015).
  • MTOR-pathway therapy in LMS. Authors note prior efficacy of everolimus and temsirolimus in LMS but flag the limitation of compensatory AKT activation; recommend evaluating dual PI3K/MTOR inhibitors and TORC1/TORC2 inhibitors given pervasive PTEN/AKT3/MTOR/IGF1R/RICTOR alterations.
  • Hippo-pathway therapy in UPS/MFS. A subset of UPS/MFS may be Hippo-driven via VGLL3/YAP1 amplification; emerging Hippo-pathway inhibitors are nominated.
  • JUN as therapeutic target in DDLPS. JUN amplification (defining poor-prognosis K1 cluster) is proposed as a tractable target as JUN inhibitors mature.
  • Checkpoint inhibition. Authors cite the SARC028 trial (Burgess et al., 2017) showing 40% response to a PD-1 inhibitor in UPS, and propose that MFS — molecularly indistinguishable from UPS — should be added to checkpoint-inhibitor trials such as those evaluating pembrolizumab. Sarcoma-type-specific differential expression of B7-H3, TGFB1, and TIM3 may also guide checkpoint-target selection.
  • Histopathology revision. Molecular data argue MFS and UPS are not distinct entities but a continuum, supporting common systemic-treatment approaches.

Limitations & open questions

  • Hypothesis-generating outcome analyses. Cases were not consecutively accrued; all survival/subtype associations require confirmation in independent cohorts (the authors explicitly flag this).
  • MPNST and SS underpowered. Only 5 MPNST and 10 SS — insufficient for the deep prognostic and methylation-subtype analyses performed for DDLPS, LMS, and UPS/MFS.
  • Treatment-naïve only. Patients with prior chemotherapy or radiotherapy were excluded, leaving the genomics of post-treatment / recurrent / metastatic sarcoma uncharacterized; the well-differentiated liposarcoma component of DDLPS was also excluded due to nucleic-acid-yield constraints.
  • MFS/UPS similarity may reflect sampling. The substantial fraction of “nonclassic” MFS samples (11/17, 65%) — including 5 high-grade epithelioid MFS — may have biased the MFS arm toward UPS-like profiles.
  • miR-181b-5p mechanism unclear. High miR-181b-5p is associated with low expression of its predicted PI3K targets (AKT3, MTOR) — opposite to the mechanism proposed in vascular smooth muscle (Li et al., 2015), leaving the operative pathway in LMS open.
  • Immune-signature validation partial. External validation in 113 sarcomas (Lesluyes et al., 2016) succeeded for LMS, MFS, and SS but was inconclusive for DDLPS and UPS where median scores clustered near zero.
  • No prospective treatment data. Therapeutic recommendations (dual PI3K/MTOR, Hippo, JUN inhibitors, MFS checkpoint expansion) are nominated but unvalidated in this study.

Citations from this paper used in the wiki

  • “We studied 206 sarcomas with diagnoses confirmed by expert pathology review: 80 LMS (53 STLMS and 27 ULMS), 50 DDLPS, 44 UPS, 17 MFS, 10 SS, and 5 MPNST” (p. 5).
  • “This identified only 3 SMGs: TP53, ATRX, and RB1” (p. 5).
  • “The overall somatic mutation burden in these 206 sarcomas was low (average 1.06 per Mb)” (p. 5).
  • “MDM2 amplification was present in all DDLPS by definition…CDK4 amplification in 86% and CDKN2A deep deletion in 2%” (p. 5).
  • “Of the mutations, 90% were attributable to COSMIC5 (53%) and COSMIC1 (37%)” (p. 6).
  • “Meth2 had more genome doublings (p=0.002) and lower leukocyte fraction (p=0.0007), and correlated with worse DSS (HR=4.4; p=0.002)” (p. 6).
  • “miR-181b-5p (univariate HR 8.03; adjusted p<0.0001)…independent predictor of RFS (HR 7.4, 95% confidence interval 3.1–17.8, p=9e-6)” (p. 7).
  • “46/55 (84%) of ULMS and STLMS iCluster C1 tumors contained alterations in the AKT pathway compared to 11/25 (44%) of STLMS iCluster C2 (p=1e–04)” (p. 7).
  • “high-level amplification of CCNE1 in 10%, VGLL3 in 11%, and YAP1 in 3%” in UPS/MFS (p. 8).
  • “the promising results of the SARC028 trial of a PD-1 inhibitor, in which 40% of UPS cases showed responses…these immunotherapy agents should be specifically explored in MFS as well” (p. 9).

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