Identification of Distinct Heterogenic Subtypes and Molecular Signatures Associated with African Ancestry in Triple Negative Breast Cancer Using Quantified Genetic Ancestry Models in Admixed Race Populations

Abstract

Triple negative breast cancers (TNBCs) are molecularly heterogeneous, and the link between their aggressiveness with African ancestry is not established. We investigated primary TNBCs for gene expression among self-reported race (SRR) groups of African American (AA, n = 42) and European American (EA, n = 33) women. RNA sequencing data were analyzed to measure changes in genome-wide expression, and we utilized logistic regressions to identify ancestry-associated gene expression signatures. Using SNVs identified from our RNA sequencing data, global ancestry was estimated. We identified 156 African ancestry-associated genes and found that, compared to SRR, quantitative genetic analysis was a more robust method to identify racial/ethnic-specific genes that were differentially expressed. A subset of African ancestry-specific genes that were upregulated in TNBCs of our AA patients were validated in TCGA data. In AA patients, there was a higher incidence of basal-like two tumors and altered TP53, NFB1, and AKT pathways. The distinct distribution of TNBC subtypes and altered oncologic pathways show that the ethnic variations in TNBCs are driven by shared genetic ancestry. Thus, to appreciate the molecular diversity of TNBCs, tumors from patients of various ancestral origins should be evaluated.

Keywords: African ancestry; RNAseq analysis; disparities; oncologic pathways; triple negative breast cancer.

Figure 1

Differentially expressed genes (DEGs) associated with quantified genetic ancestry (QGA) in treatment-naïve TNBC RNA-seq. (A) QGA estimates for each cancer case, derived from RNAseq variants. Geographic ancestry super-group categories are indicated as European (EUR, light blue), East Asian (EAS, dark blue), American Native (AMR, light green), South Asian (SAS, dark green), and African (AFR, pink). Samples are grouped by treatment status (treatment naïve or residual tumor) and self-reported race (SRR). (B) Clustergram heatmap of the 156 (p < 0.05) genes that show differential expression levels by QGA, where rows represent genes and columns represent individuals. SRR is shown in the top row of the color map (red indicating EA, and blue indicates AA); the remaining color map rows indicate ancestry estimations for each individual. The red box indicates genes that are associated with non-European admixture (EAS, SAS, and AMR). Constellation plot, right, representing the hierarchical structure of the individuals shown at the bottom of the heatmap. Red dots indicate SRR EAs; blue dots are SRR AAs. The red arrow points to the substrata of EA individuals with increased admixture; this corresponds to the non-European admixture genes in the red box of the heatmap. (C) Multidimensional analysis using 156 ancestry-associated genes indicates that the expression patterns separate individuals into SRR groups. Red indicates EA, and blue indicates AA. The blue arrow indicates an individual that self-reported as EA but has mostly AFR ancestry, clustering with the AA group. (D) De novo network analysis using QGA DEGs using Ingenuity Pathway Analysis (IPA) software. Molecules in green are upregulated in individuals with increased AFR ancestry; those in blue are downregulated in individuals with AFR ancestry. Molecules in orange are drawn into the network and predicted to be activated based on the state of DEGs in the network, using published interactions from the curated Ingenuity Knowledge Base. Orange lines between molecules indicate relationships leading to activation; blue lines indicate relationships leading to inhibition. Yellow lines indicate that the relationship between two molecules is not in the expected direction. For example, in this network TP53 is known to inhibit AKT1. TP53 is activated, and so it is expected that AKT1 would be downregulated. However, it is not. Because of this, the line showing the interaction between these two molecules is shown as yellow.

Figure 2

TNBC subtyping and distribution among SRR and treatment groups. (A) Distribution of re-assigned TNBC subtype calls across SRR race groups using the Vanderbilt calling method. (B) Clustergram heatmap of TNBCType call correlations for BL1, BL2, LAR, and M subtypes from use of the Vanderbilt tool with TPM (transcripts per million) and FPKM (fragments per kilobase of exon model per million reads mapped) values as input and our TNBC subtyping method (TNHF mean and median). Rows represent the different subtype correlations for the tools, and columns represent individuals. Using these correlations, our samples separate into six clusters, numbered at the bottom. Color map columns of the call reassignments removing IM and MSL are shown at the top of the heatmap (key to the upper right). Sample names are colored based on their cluster assignment. Reassignment of TNBC subtypes based on a dual-tool reduction method. Cluster Nodes: 1 = LAR+/BL1−, 2 = M−, 3 = M+, 4 = BL2+/BL1−, 5 = BL1+/BL2−, 6 = Indistinct (IND). (C) Parallel plots for each of the six clusters showing the correlation for the samples within the cluster to a given TNBC subtyping call/method (bottom). Cluster coloring matches that in panel 2B. (D) Sankey plot showing how samples reassign from the original TPM call, to the second most correlated call (for re-assignment of IM, MSL, and UNS samples) and their cluster assignment from panel B. (E) TNBC clusters (from panel B) and their distribution among SRR. (F) Total abundance of tumor-associated leukocytes estimated using CIBERSORT deconvolution methods is shown in comparison to SRR and treatment groups.

Figure 3

African ancestry-associated genes that are current drug targets in cancer treatments show different survival outcomes between SRR groups and breast cancer subtypes. (A) Gene expression levels between QGA differentially expressed genes identified as drug targets between SRR CAs and AAs in our TNBC dataset. (B) Gene expression levels between QGA differentially expressed genes in Figure 3A, but from the TCGA cohort. (C) Relapse-free survival curve of PIM3 for SRR AAs shows that higher expression of PIM3 is associated with higher probability of relapse-free survival (p = 0.051). (D) Relapse-free survival curve for PIM3 for SRR CAs shows that higher expression of PIM3 is associated with lower probability of relapse-free survival (p = 0.11). (E) Relapse-free survival curve of PIM3 for TNBC basal-like 1 tumors shows that higher expression of PIM3 is associated with higher probability of relapse-free survival (p = 0.0051). (F) Relapse-free survival curve of PIM3 for TNBC mesenchymal tumors shows that higher expression of PIM3 is associated with lower probability of relapse-free survival (p = 0.24).