Genetic Ancestry-specific Molecular and Survival Differences in Admixed Patients With Breast Cancer

Abstract

Objective: We aim to determine whether incremental changes in genetic ancestry percentages influence molecular and clinical outcome characteristics of breast cancer in an admixed population.

Background: Patients with breast cancer are predominantly characterized as "Black" or "White" based on self-identified race/ethnicity or arbitrary genetic ancestry cutoffs. This limits scientific discovery in populations that are admixed or of mixed race/ethnicity as they cannot be classified based on historical race/ethnicity boxes or genetic ancestry cutoffs.

Methods: We used The Cancer Genome Atlas cohort and focused on genetically admixed patients that had less than 90% European, African, Asian, or Native American ancestry.

Results: Genetically admixed patients with breast cancer exhibited improved 10-year overall survival relative to those with >90% European ancestry. Within the luminal A subtype, patients with lower African ancestry had longer 10-year overall survival compared to those with higher African ancestry. The correlation of genetic ancestry with gene expression and DNA methylation in the admixed cohort revealed novel ancestry-specific intrinsic PAM50 subtype patterns. In luminal A tumors, genetic ancestry was correlated with both the expression and methylation of signaling genes, while in basal-like tumors, genetic ancestry was correlated with stemness genes. In addition, we took a machine-learning approach to estimate genetic ancestry from gene expression or DNA methylation and were able to accurately calculate ancestry values from a reduced set of 10 genes or 50 methylation sites that were specific for each molecular subtype.

Conclusions: Our results suggest that incremental changes in genetic ancestry percentages result in ancestry-specific molecular differences even between well-established PAM50 subtypes which may influence disparities in breast cancer survival outcomes. Accounting for incremental changes in ancestry will be important in future research, prognostication, and risk stratification, particularly in ancestrally diverse populations.

Figure 1.. Overview of the admixed population.

Flowchart showing the filtering steps and the number of samples included in our cohort as well as their distribution in molecular subtypes. Stacked bar plots represent the ancestry fractions per sample in the population.

Figure 2:. Overall survival in luminal A breast cancer patients.

Overall survival in luminal A breast cancer patients stratified by (A) admixed, >90% European, >90% West African, and >90 Asian and (B) 25.5% West African ancestry cutoff among the admixed cohort.

Figure 3:. Correlation of gene expression with West African and European ancestries.

(A) Heatmap and hierarchical clustering (metric: Euclidean distance) showing the Spearman correlation coefficient of the top 100 genes per subtype per with European (EU) or West African (WA) ancestry. (B-C) Heatmaps showing the enrichment of Hallmark sets (B), KEGG pathways (C) in the correlations of gene expression with West African or European ancestry. (D) GSEA leading edge plots showing the enrichment of a pan-cancer stemness signature in the positive or negative correlations with ancestry. The vertical black bars indicate the positions of the stemness genes in the ranked dataset. NES: normalized enrichment score.

Figure 4:. Correlation of DNA methylation with West African and European ancestries.

(A) Heatmap and hierarchical clustering (metric: Kendall correlation) showing the Spearman correlation coefficient of the top 100 methylation probes per subtype with European (EU) or West African (WA) ancestry. (B) Heatmap showing the enrichment of KEGG pathways in the genes closest to the methylation probes correlated with ancestry per subtype.

Figure 5:. Toward a gene expression or DNA methylation signature predictive of genetic ancestry.

(A) Bar plot showing the Spearman correlation coefficients between the superPC-estimated ancestry with the original genetic ancestry when using all genes. (B) Bar plots showing the same coefficient as in (A) when using subset of the most significant genes. The vertical dashed red line marks the respective correlation coefficient in (A). Asterisks indicate statistically significant difference among the correlation coefficient in (A) and the respective coefficient (P-value<0.05; Pearson and Filon test). (C) Heatmap showing the top 10 genes in terms of significant for estimating ancestry in each group. (D) Bar plot showing the Spearman correlation coefficients between superPC-estimated ancestry with the original genetic ancestry using all DNA methylation probes. (E) Bar plots showing the same information as in (D) but using a subset of the most significant methylation probes. Vertical lines and asterisks mark the same respective information as in (B). (F) Heatmap showing the top 50 methylation probes in terms of significant for estimating genetic ancestry in each group.