Multi-omics analysis identifies therapeutic vulnerabilities in triple-negative breast cancer subtypes

Abstract

Triple-negative breast cancer (TNBC) is a collection of biologically diverse cancers characterized by distinct transcriptional patterns, biology, and immune composition. TNBCs subtypes include two basal-like (BL1, BL2), a mesenchymal (M) and a luminal androgen receptor (LAR) subtype. Through a comprehensive analysis of mutation, copy number, transcriptomic, epigenetic, proteomic, and phospho-proteomic patterns we describe the genomic landscape of TNBC subtypes. Mesenchymal subtype tumors display high mutation loads, genomic instability, absence of immune cells, low PD-L1 expression, decreased global DNA methylation, and transcriptional repression of antigen presentation genes. We demonstrate that major histocompatibility complex I (MHC-I) is transcriptionally suppressed by H3K27me3 modifications by the polycomb repressor complex 2 (PRC2). Pharmacological inhibition of PRC2 subunits EZH2 or EED restores MHC-I expression and enhances chemotherapy efficacy in murine tumor models, providing a rationale for using PRC2 inhibitors in PD-L1 negative mesenchymal tumors. Subtype-specific differences in immune cell composition and differential genetic/pharmacological vulnerabilities suggest additional treatment strategies for TNBC.

Fig. 1. Identification of TNBC subtype features through integrative genomic analyses.

a Summary of datasets and workflow used in this study. b Heatmap shows TNBC samples from TGCA stratified by subtype correlation strength and annotated for k-means group, PAM50 subtype, age, positive lymph nodes, and tumor microenvironment (TIME) classification. Gene expression heatmaps show immune cell abundance (ESTIMATE), scRNA deconvolution of normal mammary cells and immune cell lineages, relative RNA expression for immune markers, and antigen presentation and immune checkpoint genes. Mutation and copy number alterations are displayed for individual tumors and stratified by pathway. * indicates significant (p < 0.05 two-sided Fisher’s exact test, raw p-values in source file) differences in mutation/CNA in one subtype (colored) compared to all others. c Representative H&E images showing TIME classification of TCGA into fully inflamed (FI), stromal-restricted (SR), margin-restricted (MR), or immune desert (ID). These images have no scale bar because they were obtained from the TCGA Digital Slide Archive. d Barplot shows TIME quantification of images by TNBC subtype. See also Supplementary Fig. 1, and Supplementary Data 1, 2, and 3.

Fig. 2. Gene expression and phosphoproteomic analyses identify unique subtype-specific targetable pathways.

a Heatmap shows differentially expressed proteins (p < 0.05 and Log2 FC >1) and significantly enriched pathways (Reactome) by subtype for TNBC tumors in CPTAC. Balloon plots show integrated pathway analysis of RNA, protein, and phosphoprotein log fold change values for genes/proteins from CPTAC TNBC tumors demonstrating subtype-specific differences in b DNA repair/cell cycle, c PI3K/mTOR, d EGFR/MAPK, and e antigen presentation pathways. FDR represents the false discovery rate that the normalized enrichment score represents a false-positive finding.

Fig. 3. In silico analyses of datasets from genetic and pharmacologic screens identifies subtype-specific vulnerabilities in TNBC.

a Datasets and workflow for identifying subtype-specific genetic dependencies and pharmacologic sensitivities. TNBC cell lines identified from datasets and subtype correlation of all cell line models with values to yield a single subtype T-value for each drug sensitivity/genetic dependency. b Subtype-specific genetic dependencies from Broad DepMap whole-genome RNAi and CRISPR screen. Overlapping genetic dependencies and drug sensitivities from both screens are highlighted in the heatmap by pathway. Heatmaps show significant subtype-specific pharmacological dependencies determined with a modified T-test corrected for multiple hypothesis testing (T-value, FDR <0.1) in c TNBC cell lines screened in the Genomics of Drug Sensitivity in Cancer or d PDX explant drug sensitivity colored by similar pathway and subtype. See also Supplementary Data 6.

Fig. 4. TNBC subtypes differ in global methylation patterns and mesenchymal TNBCs show specific methylation for EZH2 targets and antigen presentation genes.

a Heatmap shows all differentially (FDR <0.05 and β-value>0.1) methylated CpGs by subtype. b Graph shows the number of differentially) hypomethylated and hypermethylated CpGs by TNBC subtype. c Barplots show the percentage of hypomethylated (blue) and hypermethylated (yellow) CpGs within each chromosome by TNBC subtype. d Pie charts show the distribution of differentially methylated CpG sites in each subtype by genomic location. e Starburst plots show gene expression and DNA methylation <3kb from promoter regions of corresponding genes. Significantly (FDR <0.05) hypo- and hyper-methylated promoter regions are colored in blue and yellow respectively, with genes of interest labeled in the plot. f Gene set enrichment analysis of genes negatively correlated with hypomethylated (blue) or hypermethylated (yellow) promoter methylation probes.

Fig. 5. Inhibition of polycomb repressive complex 2 restores MHC-I expression in mesenchymal TNBC models.

a Immunoblot shows levels of MHC-I (HLA-A/B/C) across TNBC cell lines. Blots are representative of two independent experiments. b Representative images show immunohistochemistry for MHC-I expression on individual cells in TNBC cell lines. c Histogram plots show the distribution of membrane-bound MHC-I protein (+) in TNBC models compared to unstained controls (−). Images are representative of three independent cores from a tissue microarray. d Plots show relative viability of TNBC cell lines treated with increasing concentrations (1.25, 2.5, 5, 10, and 20 μM) of tazemetostat, CPI-1205, or MAK-683. Error bars represent the standard deviation of three independent experiments. e Diagram shows experimental workflow and f heatmaps show differential gene expression of M-subtype TNBC cell line models after 5 days of a single 10 μM treatment with inhibitors of the PRC2 complex, tazemetostat (EZH2 inhibitor), CPI-1205 (EZH2 inhibitor) or MAK-683 (EED inhibitor). g Gene ontology analysis (Hallmark) of the gene in the union (Supplementary Fig. 9e) between CAL51, CAL-120, and BT549 cell lines that were significantly (FDR p-value <0.05, FC>2) upregulated between treatment with PRC2 inhibitors (n = 9) compared to DMSO treatment (n = 3). Differential genes identified by modified T-test corrected for multiple hypothesis testing. h Heatmap shows an expression of MHC-I and MHC-II genes in M-subtype TNBC cell lines treated with control (DMSO) or PRC2 inhibitors. i Immunoblots show H3K27me3 and MHC-I protein expression at 1, 3, 5, 7 days after a single 10 μM treatment with either tazemetostat, CPI-1205, or MAK-683. Immunoblots are representative of two experiments. j Histograms show the distribution of cell-surface MHC-I protein expression 5 days after a 10 μM treatment with the indicated PRC2 inhibitors. Error bars were determined from three independent experiments.

Fig. 6. EZH2 inhibition decreases global H3K27me3 and repressive marks at MHC-I locus.

a Profile plot and heatmap for H3K27me3 ChIP-seq signal for differential peak, H3K27me3 targets (BENPORATH_ES_WITH_H3K27ME3), or EZH2 targets (NUYTTEN_EZH2_TARGETS_UP) centered on transcriptional-start sites (TSS) for BT549, CAL-120, and CAL-51 cells treated for 4 days with either DMSO or 1 μM tazemetostat. Sequencing reads were normalized to reads per genomic content. b Scatterplots show differential RNA expression (Log2 FC, FDR <0.05) and differential H3K27me3 promoter occupancy (FDR <0.0.5) in tazemetostat treated cells relative to DMSO treatment.

Fig. 7. EZH2 inhibition increases the efficacy of paclitaxel chemotherapy in a syngeneic murine TNBC model.

a Immunoblots show H3K27me3 and MHC-I protein expression at 1, 3, 5, 7 days after a single 10 μM treatment with tazemetostat, CPI-1205, or MAK-683. b Histograms show the distribution and c quantification of cell-surface MHC-I protein expression 5 days after a 10 μM treatment with the indicated PRC2 inhibitors. Results are representative of three experiments. Error bars represent the standard error of the mean. d Treatment schedule for mice bearing syngeneic 4T1 xenograft tumors. Mice were treated with vehicle, twice daily (BID) with 250 mg/kg tazemetostat, twice a week with 10 mg/kg paclitaxel, or the combination of tazemetostat and paclitaxel. Results are representative of ten tumors. e Graphs show 4T1 tumor volume (mm³) across time of mice treated with vehicle, tazemetostat, paclitaxel, or the combination. Error bars represent the standard error of the mean. Significance determined by two-tailed Student’s t-tests, *p = 0.0353. f Barplot shows the distribution of final tumor weight (mg) from mice treated with vehicle, tazemetostat (TAZ), paclitaxel (TAX), or the combination (TAZ + TAX). Significance determined using Dunnett’s multiple comparison test. *p = 0.0423. g Plot shows IHC quantification of intratumor CD3+ T-cells in 4T1 xenograft tumors by treatment group. h Schematic shows CpG methylation and H3K27me3 epigenetic states in non-mesenchymal TNBC or mesenchymal TNBC with or without EZH2 inhibition.