Adipogenesis in triple-negative breast cancer is associated with unfavorable tumor immune microenvironment and with worse survival

Abstract

Cancer-associated adipocytes are known to cause inflammation; however, the role of adipogenesis, the formation of adipocytes, in breast cancer is unclear. We hypothesized that intra-tumoral adipogenesis reflects a different cancer biology than abundance of intra-tumoral adipocytes. The Molecular Signatures Database Hallmark adipogenesis gene set of gene set variant analysis was used to quantify adipogenesis. Total of 5,098 breast cancer patients in multiple cohorts (training; GSE96058 (n = 3273), validation; TCGA (n = 1069), treatment response; GSE25066 (n = 508) and GSE20194 (n = 248)) were analyzed. Adipogenesis did not correlate with abundance of adipocytes. Adipogenesis was significantly lower in triple negative breast cancer (TNBC). Elevated adipogenesis was significantly associated with worse survival in TNBC, but not in the other subtypes. High adipogenesis TNBC was significantly associated with low homologous recombination deficiency, but not with mutation load. High adipogenesis TNBC enriched metabolism-related gene sets, but neither of cell proliferation- nor inflammation-related gene sets, which were enriched to adipocytes. High adipogenesis TNBC was infiltrated with low CD8⁺ T cells and high M2 macrophages. Although adipogenesis was not associated with neoadjuvant chemotherapy response, high adipogenesis TNBC was significantly associated with low expression of PD-L1 and PD-L2 genes, and immune checkpoint molecules index. In conclusion, adipogenesis in TNBC was associated with cancer metabolism and unfavorable tumor immune microenvironment, which is different from abundance of adipocytes.

Figure 1

Association of the adipogenesis score with expression of adipogenesis- and adipocyte-related genes, and correlation with adipocyte score in the GSE96058 and TCGA cohorts. Boxplots comparing low and high adipogenesis score tumors of gene expression levels of (A) adipogenesis-related genes; Acetyl-CoA carboxylase/ACLY, ATP citrate lyase/ACACA, NADP-dependent malic enzyme/SLC25A10, and mitochondrial dicarboxylate carrier/ME1, and (B) adipocyte-related genes; adiponectin/ADIPOQ, leptin/LEP, lipoprotein/LPL, and perilipin 1/PLIN1 in breast cancer. P-value was calculated using Mann–Whitney U test. (C) Correlation pots between adipogenesis and adipocytes scores. Spearman correlation coefficient (r) was used to the analysis.

Figure 2

Association of the adipogenesis pathway score with clinical features in the GSE96058 and TCGA cohorts. Boxplots of the adipogenesis scores of tumors by (A) American Joint Committee on Cancer stage, tumor size (T-category) and lymph node positivity (N-category), Nottingham pathological grade, and (B) breast cancer subtype, and Pam50 classifications. P-value was calculated using Kruskal–Wallis test. Group sizes are shown underneath the plots.

Figure 3

Association of adipogenesis pathway score with patient survival by whole and each subtype in the GSE96058 and TCGA cohorts. Kaplan–Meier plots of Disease-specific survival (DSS) in the TCGA cohort, and overall survival (OS) in both cohorts by adipogenesis low (blue) and high (red) score with (A) whole, (B) Estrogen receptor (ER)+/HER2−, (C) human epidermal growth factor receptor 2 (HER2)-positive, or (D) Triple negative breast cancer (TNBC) patients. Median value was used as the cut-off to divide into low and high adipogenesis score groups within each cohort. P-value was calculated using log-rank test.

Figure 4

Gene Set Enrichment Assay (GSEA) on adipogenesis score in TNBC of both GSE96058 and TCGA cohorts. Gene set enrichment plots of (A) Hallmark metabolism-related gene sets (oxidative phosphorylation, fatty acid metabolism, peroxisome, reactive oxygen species pathway) (B) inflammation-related gene sets (inflammatory response, interferon (IFN)-α response and IFN-γ response) and (C) cell proliferation-related gene sets (E2F target, G2M checkpoint, mitotic spindle, and MYC target v1 and v2) in both cohorts with normalized enrichment score (NES) and false discovery rate (FDR).

Figure 5

Homologous recombination deficiency (HRD), mutation load, fractions of tumor infiltrating immune cells and cytolytic activity between low and high adipogenesis triple negative breast cancer (TNBC) in the GSE96058 and TCGA cohorts. (A) Boxplots of HRD, altered fraction, silent and non-silent mutation, single nucleotide variation (SNV) and indel neoantigens by the adipogenesis in TNBC. (B) The comparison of immune cells; CD8⁺ T cells, CD4⁺ memory T cells, T helper type 1 cells (Th1) and Th2, regulatory T cells (Tregs), M1 and M2 macrophage, and (C) cytolytic activity score (CYT). Median value was used as the cut-off to divide into low and high adipogenesis score groups within each cohort. P-value was calculated using Mann–Whitney U test.

Figure 6

Association of adipogenesis score with neoadjuvant chemotherapy (NAC) response and expression of immune checkpoint molecules in TNBC. (A) Bar plots of pathological compete response (pCR) rate after NAC between low (blue) and high (red) adipogenesis score in ER+/HER2− and TNBC in the GSE25066 and GSE20194 cohorts. The numbers under the chart are pCR case number/number of patients in that group that makes the pCR rate. Error bar showed Standard Error. P-value was calculated by Fisher’s test. (B) Boxplots of comparison between low- and high-adipogenesis score with mRNA expression of PD-1/PDCD1, PD-L1/CD274, PD-L2/PDCD1LG2, and CTLA4 genes, and (C) immune checkpoint molecules index in TNBC in the GSE96058 and TCGA cohorts. P-value was calculated using Mann–Whitney U test. Median value was used as the cut-off to divide into low- and high- adipogenesis score groups within each cohort. PD-1 programmed cell death 1; PD-L1 programmed cell death 1 ligand 1, PD-L2 programmed cell death 1 ligand 2, CTLA4 cytotoxic T-lymphocyte-associated protein 4.