Integrative Analysis of Gene Expression and DNA Methylation Depicting the Impact of Obesity on Breast Cancer

Abstract

Obesity has been reported to be a risk factor for breast cancer, but how obesity affects breast cancer (BC) remains unclear. Although body mass index (BMI) is the most commonly used reference for obesity, it is insufficient to evaluate the obesity-related pathophysiological changes in breast tissue. The purpose of this study is to establish a DNA-methylation-based biomarker for BMI (DM-BMI) and explore the connection between obesity and BC. Using DNA methylation data from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO), we developed DM-BMI to evaluate the degree of obesity in breast tissues. In tissues from non-BC and BC population, the DM-BMI model exhibited high accuracy in BMI prediction. In BC tissues, DM-BMI correlated with increased adipose tissue content and BC tissues with increased DM-BMI exhibited higher expression of pro-inflammatory adipokines. Next, we identified the gene expression profile relating to DM-BMI. Using Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, we observed that the DM-BMI-related genes were mostly involved in the process of cancer immunity. DM-BMI is positively correlated with T cell infiltration in BC tissues. Furthermore, we observed that DM-BMI was positively correlated with immune checkpoint inhibitors (ICI) response markers in BC. Collectively, we developed a new biomarker for obesity and discovered that BC tissues from obese individuals exhibit an increased degree of immune cell infiltration, indicating that obese BC patients might be the potential beneficiaries for ICI treatment.

Keywords: DNA methylation; biomarker; breast cancer; immune checkpoint inhibitor; obesity.

FIGURE 1

Flow chart of the study design. We enrolled 221 normal breast tissues as training set to develop a lasso regression to predict DM-BMI and validated the accuracy of the model with data from normal breast tissues and tumor-adjacent breast tissues. Then, we predicted DM-BMI in 775 BC tissues and 76 matched tumor-adjacent breast tissues. The correlation between DM-BMI and clinical characteristics was assessed in BC tissues. Further, we identified the DM-BMI related gene profile and evaluated the relationships between DN-BMI and tumor immune response in BC tissues.

FIGURE 2

Development and validation of the DA-BMI predicting model. (A,B) Correlation of DM-BMI with (A) BMI and (B) proportion of adipose tissue in normal breast tissues based on the training set (GSE88883 and GSE101961). (C,D) Correlation of DM-BMI with (C) BMI and (D) proportion of adipose tissue in normal breast tissues based on Validation Set 1 (GSE67919 and GSE74214). (E,F) Correlation of DM-BMI with (E) BMI and (F) proportion of adipose tissue in tumor-adjacent breast tissues based on Validation Set 2 (GSE67919). (G) Analyzing the differences of DM-BMI between tumor tissues (n = 76) and matched tumor-adjacent breast tissues (n = 76) based on the TCGA-BRCA dataset. (H) Correlation of DM-BMI with proportion of adipose tissue in BC tissues based on the TCGA-BRCA dataset (n = 775). (A-F and H) r, Spearman correlation coefficient. (G) p-values were determined by paired t-test.

FIGURE 3

Characteristic analyses of BMI predictors. (A) Distribution of 42 BMI predictors referring to (Left) chromosome and (Middle) transcription Start Sites; CpG islands were listed as number (proportion). For chromosome, no BMI predictors were located in Chr14, 18, 19, 20, and 21. (B) Identification of differential methylation BMI predictors between tumor tissues and matched tumor-adjacent breast tissues in TCGA dataset (n = 76). *p < 0.05, **p < 0.01, ***p < 0.001. (C) Forest plot of the prognostic related BMI predictors referring to DNA methylation level from TCGA BC tissues (n = 774): three BMI predictors (mapping to FARP1, PLEC, and cg05731218 located in the intergenic region/IGR) negatively correlated with overall survival. (D) Volcano plot of the correlation analysis between DM-BMI and methylation level of BMI predictors. r, Spearman correlation coefficient; 39 BMI predictors positively correlated with DM-BMI (BMI predictors with correlation coefficient >0.3 were marked as red dot; n = 10), and five BMI predictors negatively correlated with DM-BMI (BMI predictors with Spearman correlation coefficient < −0.3 were marked as green dot; n = 1). (E) Venn diagram of DMP; DM-BMI-correlated and expression-correlated BMI predictors. BMI predictors differentially methylated between tumor and tumor-adjacent tissues were labeled as blue; methylation levels of the BMI predictors correlated with DM-BMI were labeled as red; methylation levels of BMI predictors negatively correlated with gene expression were labeled as green.

FIGURE 4

Functional and clinical characteristics analysis of DM-BMI-related gene profile in BC. (A,B) Analyzing the differences of DM-BMI based on (A) menopause status of patients (n = 749) or TP53 mutation status (n = 768) based on TCGA-BRCA dataset. (C,D) Correlation of DM-BMI with copy number of (C) ERBB2 and (D) MYC. r, Spearman correlation coefficient. (E) Correlation of DM-BMI and expression of pro/anti-inflammatory adipokines in BC tissues (n = 771). Expressions of adipokines significantly correlated with DM-BMI were labeled as red. (F) GO function analysis of DM-BMI related gene. (Left) Analysis of gene whose mRNA expression positively correlated with DM-BMI (Ayers et al., 2017); analysis of gene whose mRNA expression negatively correlated with DM-BMI. (G) Analysis of DM-BMI-related gene enrichment in immunologic pathway based on KEGG database. Gene ratio was defined as: number of genes enriched to target pathway/number of DM-BMI related genes included in the KEGG dataset.

FIGURE 5

DM-BMI correlated with T cell infiltration and ICI response in BC. (A) Correlation of DM-BMI with tumor mutation burden (TMB) in BC tissues (TCGA-BRCA, n = 775). Numerical distribution of DM-BMI and TMB is shown on the above x- and the right y-axis, respectively. (B) Correlation of DM-BMI with the level of infiltrating immune cells (Left, estimate-immune score) and the level of stromal cells (Right, estimate-stromal score) in BC tissues (TCGA-BRCA, n = 772). Numerical distribution of DM-BMI and estimate-immune/stromal score is shown on the above x- and the right y-axis, respectively. (C) Correlation of DM-BMI with 22 types of immune cell components is shown by dotplot. Five of 7 T-cell contents correlated with DM-BMI, which were labeled in red. (D) GSVA analysis identified an immunologic pathway which positively correlated with DM-BMI. Enrichment scores of pathways from the GSEA-Molecular Signatures Database were calculated using GSVA in BC tissues (TCGA-BRCA-mRNA data, n = 771). Immunologic gene sets which significantly correlated with DM-BMI are displayed (r > 0.3). (E) Correlation of DM-BMI with markers for ICI response/resistance is shown by dotplot. For (A,B), r, Spearman correlation coefficient.