A Comprehensive Pan-Cancer Molecular Study of Gynecologic and Breast Cancers

Abstract

We analyzed molecular data on 2,579 tumors from The Cancer Genome Atlas (TCGA) of four gynecological types plus breast. Our aims were to identify shared and unique molecular features, clinically significant subtypes, and potential therapeutic targets. We found 61 somatic copy-number alterations (SCNAs) and 46 significantly mutated genes (SMGs). Eleven SCNAs and 11 SMGs had not been identified in previous TCGA studies of the individual tumor types. We found functionally significant estrogen receptor-regulated long non-coding RNAs (lncRNAs) and gene/lncRNA interaction networks. Pathway analysis identified subtypes with high leukocyte infiltration, raising potential implications for immunotherapy. Using 16 key molecular features, we identified five prognostic subtypes and developed a decision tree that classified patients into the subtypes based on just six features that are assessable in clinical laboratories.

Keywords: TCGA; The Cancer Genome Atlas; breast cancer; cervical cancer; gynecologic cancer; omics; ovarian cancer; pan-gynecologic; uterine cancer; uterine carcinosarcoma.

Figure 1. Genomic features that distinguish Pan-Gyn from other tumor types

(A) Heat map showing the frequencies of mutations (green) in 23 genes across all 33 TCGA tumor types and frequencies of amplifications (red) in 23 genes across all 33 TCGA tumor types. (B) Amplification (red) and deletion (blue) q values from GISTIC2.0 for SCNA peaks of significant copy number gain and loss plotted for Pan-Gyn vs. non-Gyn cohorts. Genes named are the suspected targets of amplification or deletion, if identifiable. Otherwise, peaks are labeled with the nearest cytoband’s designation. Peaks found in only one cohort were assigned values of NS (not significant) in the other cohort. See also Figure S1 and Table S1.

Figure 2. Landscape of mutations in Pan-Gyn tumor types

(A) Mutation profiles of 2,029 Pan-Gyn samples (columns) in which at least one somatic mutation occurred in at least one of the 46 significantly mutated genes (SMGs). (Top) Mutation burdens per sample, divided into synonymous and non-synonymous mutation types. (Middle) Types of mutations in each of the 46 SMGs per sample. (Bottom) Covariate bars showing the mutation cluster, genomic alterations in six genes from the DNA damage response pathway, and tumor type for each sample. (B) Clustered heat map showing correlations between 10 of our mutation signatures (rows labeled S1 to S10) and 30 COSMIC signatures (columns). (C) Clustered heat map of the mutation signatures (rows) present in each sample (columns) showing ten clusters. The dendrogram is color-coded by predominant COSMIC signature. See also Figure S2, and Tables S2-5.

Figure 3. Clustered heat map of significantly recurring SCNAs as determined by GISTIC2.0 analysis across Pan-Gyn cancers

The heat map shows SCNAs in tumor samples (columns) plotted by chromosomal location (rows). Red and blue indicate amplifications and deletions, respectively. See also Figure S3 and Tables S4-5.

Figure 4. mRNA expression clusters and their association with overall survival

(A) Unsupervised hierarchical clustering of previously reported cancer genes identifies nine mRNA-based subtypes/clusters. Clinical and molecular features are indicated by the annotation bars above the heat map. (B) Overall survival for each of the gene expression clusters (chi-squared test p value < 0.0001, adjusted for differences in tumor type survival rates). (C) Overall survival for endometrial cancer (UCEC) patients in the gene expression clusters (log rank test p value < 0.0001). (D) Differential expression of ESR1, AR, SOX2, and CDH1 in different clusters (Kruskal-Wallis test p values < 0.0001 for all 4 genes). The bars represent mean expression of the gene (log₂ scale) in each cluster, together with the upper or lower 95% confidence interval (whiskers above or below the bars, respectively). See also Tables S4-5.

Figure 5. lncRNA clusters and gene/lncRNA interaction networks

(A) Clustered heat map based on expression of cancer lncRNA regulators. The rows have 1,986 lncRNAs, whereas the columns have 1,597 samples. L1-L6 indicate the six clusters and their association with protein clusters is shown (p value < 0.05, Fisher’s Exact test). (B) Schematic illustration of dual-layer ER-ceRNA regulation of BRCA1. ER transcriptionally regulates both BRCA1 and non-coding TUG1 in ER-positive breast cancer. Those RNAs subsequently compete for miRNA binding. (C) ER modulates the TERC-DKC1 complex and its transcriptional activity. Estradiol (E2)-activated ER binds to cis-regulatory DNA regions of both DKC1 and TERC and regulates their activity. Further, ER binds to regulatory regions of DKC1-regulated lncRNAs (listed on the right) and modulates their expression. (D) Gene/lncRNA interaction networks in the overall Pan-Gyn lncRNA cohort and each of the four individual disease types. The nodes represent genes (green) or lncRNAs (burgundy), whereas each edge represents statistically significant Pearson’s correlation coefficient between the connected nodes. See also Figure S4-5, and Tables S4-5.

Figure 6. Pathways-based clusters

(A) Consensus Clustered heat map based on PARADIGM integrated pathway levels (IPLs). Selected pathway features with characteristic patterns of inferred activation across clusters are labeled on the rows. Samples are in columns. (B) Constituent pathways with differential single-sample gene set enrichment analysis (ssGSEA) scores across PARADIGM clusters. A comparison of ssGSEA scores of constituent pathways integrated by the PARADIGM algorithm identified 263 differentially enriched pathways across clusters. Samples are arranged in the same order as (A) and differentially expressed pathways are arranged based on unsupervised clustering of their ssGSEA scores. Dominant themes within sub-groupings of differential pathways across PARADIGM clusters are labeled. Examples of immune-related pathways include IL12, IL23, IL27, IFNG, STAT and T-cell receptor signaling pathways. Proliferation and DNA damage repair related pathways include FOXM1, PLK2, Cyclins, MYC, E2F, ATM, ATR, BARD1, and Fanconi anemia pathways. See also Tables S4-5.

Figure 7. Cross-tumor type Pan-Gyn subtypes with prognostic significance

(A) Clustered heat map of 16 features across 1,956 Pan-Gyn samples. Cluster 2 is split further into four subclusters, 2A-2D. Purple rectangles highlight HER2⁺ samples that have high immune infiltration scores; black rectangles highlight HER2⁺ samples with low immune infiltration scores. (B) Cross-tabulation showing the distribution of Pan-Gyn tumor types across the five clusters. (C) Kaplan-Meier curves showing differences in overall survival among the five clusters (with 5-and 10-year survival rates shown). Before adjusting for tumor type differences in overall survival rates, the log rank test p value < 0.0001, and after adjusting for tumor type differences, p value = 0.0006 (chi-squared test). (D) Decision tree that predicts clusters using just six of the 16 features. The predicted clusters are shown in a covariate bar in the heat map in A. (E) Kaplan-Meier curves showing differences in overall survival among the five decision tree-based predicted clusters (with 5- and 10-year survival rates shown). Log rank test p values are less than 0.0001, before (log rank test) and after (chi-squared test) adjusting for tumor type differences in overall survival rates. See also Figure S6 and Tables S4-5.