Comprehensive and Integrated Genomic Characterization of Adult Soft Tissue Sarcomas

Abstract

Sarcomas are a broad family of mesenchymal malignancies exhibiting remarkable histologic diversity. We describe the multi-platform molecular landscape of 206 adult soft tissue sarcomas representing 6 major types. Along with novel insights into the biology of individual sarcoma types, we report three overarching findings: (1) unlike most epithelial malignancies, these sarcomas (excepting synovial sarcoma) are characterized predominantly by copy-number changes, with low mutational loads and only a few genes (TP53, ATRX, RB1) highly recurrently mutated across sarcoma types; (2) within sarcoma types, genomic and regulomic diversity of driver pathways defines molecular subtypes associated with patient outcome; and (3) the immune microenvironment, inferred from DNA methylation and mRNA profiles, associates with outcome and may inform clinical trials of immune checkpoint inhibitors. Overall, this large-scale analysis reveals previously unappreciated sarcoma-type-specific changes in copy number, methylation, RNA, and protein, providing insights into refining sarcoma therapy and relationships to other cancer types.

Keywords: DNA methylation; The Cancer Genome Atlas; dedifferentiated liposarcoma; genomics; immune infiltration; leiomyosarcoma; molecular subtype; myxofibrosarcoma; pleomorphism; undifferentiated pleomorphic sarcoma.

Figure 1. Landscape of Genomic Alterations in 206 Sarcomas

(A) Integrated plot of clinical and molecular features for all samples, ordered by sarcoma type. From top to bottom panels indicate: frequency of mutations per Mb; mutational signatures, indicating type of substitution; patient sex; sarcoma grade; number of whole genome doublings; Number of unbalanced genomic segments; tumor site; sarcoma type; cluster from iCluster analysis; significantly mutated genes, defined by false discovery rate (FDR) of <0.05 as computed by MuSiC2; TRIO or SS18-SSX gene fusions; frequent focal somatic copy number alterations including gains (pink), amplification (red), shallow deletion (pale blue) or deep deletion (dark blue). The key to the color coding of sarcomas and mutation types is at the bottom. See also Figure S1 and Table S1. (B) Median numbers of unbalanced copy number segments vs nonsynonymous somatic mutations in each TCGA cohort. Sarcoma types are color-coded.

Figure 2. Mutational Landscape of Sarcomas

(A) Mutations in significantly mutated genes in sarcoma and selected known oncogenes and tumor suppressors. Only genes with recurrent or truncating mutations are shown. (B) Mutation and indel profiles for TP53, ATRX, RB1, NF1, and PRKCD, color-coded by sarcoma type. Splice site mutations are indicated as involving the donor site (exon number + nucleotide position of mutation, e.g. e3+1) or acceptor site (exon – nucleotide position of mutation). See also Figure S2 and Table S2.

Figure 3. Mutational Signatures, Genomic Complexity, and Integrated Analysis in Sarcoma

(A) Top, signature activities (number of mutations) and bottom, normalized signature activities, projected onto 3 mutational processes, COSMIC1, COSMIC5, and APOBEC (COSMIC2 and 13). Tumors are ordered by overall mutation frequency; not shown are the 2 hypermutated samples (AB32 and A9HT). (B) Left, activities of COSMIC1, COSMIC5, and APOBEC signatures by sarcoma type. Right, normalized signature activity. (C) Variance in nuclear area according to the number of genome doublings in each tumor. (D) Representative nuclear area analyses for sarcomas with whole genome doublings of 0, 1, and 2. See also Figure S2E, F. (E) Unsupervised iCluster analysis, which integrated DNA copy number, DNA methylation, and expression of mRNA and miRNA. Color coding of tumor characteristics is at the bottom. Cluster C1 comprised 64 LMS and 1 UPS, including 10 low-grade LMS, and was relatively hypermethylated. Cluster C2 and C3 comprised 49/50 DDLPS and 35 other sarcomas. C4 comprised all 10 SS and one MPNST, and C5 comprised a mix of high-grade sarcomas, with the majority (34/56) being UPS/MFS. See also Figures S3 and S4 and Table S3.

Figure 4. Dedifferentiated Liposarcoma (DDLPS)

(A) Recurrent focal copy-number alterations in the 50 DDLPS samples by GISTIC 2.0 analysis. Green line indicates the significance threshold (FDR 0.25) for focally amplified and deleted regions. See also Table S5. (B) Alterations of genes involved in inhibition of adipose differentiation. The frequency of copy-number alterations in DDLPS is shown for each of the 3 SCNA clusters, and the heatmap shows gene expression. (C) Methylation clusters from unsupervised consensus clustering of DNA methylation data in DDLPS. Within methylation clusters, samples are ordered by SCNA cluster and genome doubling. (D) DSS in clusters defined by copy number and DNA methylation. See also figure S4.

Figure 5. Leiomyosarcoma (LMS)

(A) Recurrent focal copy-number alterations in the 80 LMS samples by GISTIC 2.0 analysis. Green line indicates the significance threshold (FDR 0.25) for focally amplified and deleted regions. (B) Molecular landscape of LMS. ULMS was enriched for tumors in iCluster C1, mRNA C2, methylation C3, and SCNA C3 (characterized by genomic instability). STLMS was enriched for the other 2 SCNA clusters: C2 (characterized by chromosome 17p11~12 gains) and SCNA C1 (genomically quiet). FDR values next to gene mutations were computed by MuSiC2. See also Table S6 and Figure S6. (C) iCluster analysis of STLMS, demonstrating hypomethylation of C2 relative to C1. Heat maps display the most variable distinguishing factors between clusters. See also Table S6. (D) Kaplan-Meier analysis of STLMS iCluster C1 vs C2. Median DSS was 6.7 years for C1 and was not reached for C2. (E) Recurrent AKT pathway alterations in LMS. Top, pathway diagrams and percentage of alterations (mutation, SCNA, and/or relative change in mRNA level) in ULMS and STLMS iClusterC1 and C2. Bottom: specific alterations for each gene.

Figure 6. Undifferentiated Pleomorphic Sarcoma (UPS) and Myxofibrosarcoma (MFS)

(A) Integrated molecular profile of MFS and UPS, showing clusters from unsupervised analyses and recurrent gene mutations. FDRs next to gene mutations were computed by MuSiC2. (B) Molecular classification of UPS/MFS by myxoid stromal content of frozen tumor sample. Unsupervised clustering was performed on genes differentially expressed (q<0.05) between groups defined by extent of myxoid stroma (none, 1–49% of the tissue, ≥50% of the tissue). “Classic” cases of MFS (n=6) and UPS (n=20) on frozen material are indicated. See also Figure S6E. (C) SCNAs in MFS and UPS. VGLL3 amplification and RB1 deletion are shown at the top. (D) Hippo pathway activation. The boxplots show YAP1 and VGLL3 target gene expression signature (Helias-Rodzewicz et al., 2010). (E) Multivariable miRNA prognostic classifier for DSS. We performed a penalized regression analysis using all miRNAs and tumor size in the 54 UPS/MFS samples with outcome data. The samples were split into high and low groups based on model score, minimizing the log-rank p-value. P value shown is corrected for multiple testing. See also Figure S6F.

Figure 7. Specific Types of Immune Infiltration Show Associations with Survival Outcomes

(A) Clusters identified by unsupervised clustering of the 2038 most variably expressed genes across 206 samples. Heat map shows expression; the gray wedge marks 203 genes with immune-related and inflammatory-related GO terms. The bar graph (right) shows the Benjamini-Hochberg adjusted P values for enrichment for the specific ontologies listed, as defined by the DAVID algorithm. (B) Unsupervised cluster analysis of tumors by calculated immune infiltration scores. The analysis defines a subset of DDLPS, LMS, MFS and UPS with high immune infiltrates (right). (C) Selected Kaplan-Meier curves for DSS by histology and immune class. The graphs show the patients in the top third vs bottom third for the immune scores indicated. (D) Significant DSS associations (p<0.05) for high immune score by histology. See also Figure S7.