Breast Tumor Microenvironment in Black Women

A Distinct Signature of CD8+ T-Cell Exhaustion

Song Yao, PhD; Ting-Yuan David Cheng, PhD; Ahmed Elkhanany, MD; Li Yan, PhD; Angela Omilian, PhD; Scott I. Abrams, PhD; Sharon Evans, PhD; Chi-Chen Hong, PhD; Qianya Qi, MS; Warren Davis, PhD; Song Liu, PhD; Elisa V. Bandera, MD, PhD; Kunle Odunsi, MD, PhD; Kazuaki Takabe, MD; Thaer Khoury, MD; Christine B. Ambrosone, PhD


J Natl Cancer Inst. 2021;113(8):1036-1043. 

In This Article


Patient Population

Tumor tissues and data for this study were derived from WCHS as previously described[16] and in the Supplementary Methods (available online). Clinicopathological data were obtained from pathology reports, supplemented with data obtained from the New Jersey State Cancer Registry, which actively maintains updated information on deaths and causes of deaths through various follow-up sources. All study participants provided consent for the use of their data and specimens for research. The studies were approved by the institutional review boards at participating institutions.

Pathological Tumor-infiltrating Lymphocytes (TIL) Assessment

As part of the protocol of receiving unstained slides and tumor blocks in WCHS, hematoxylin and eosin sections were prepared and reviewed for histopathologic scoring of stromal TILs by a board-certified breast pathologist (TK) blinded to patient characteristics, following the recommendations by the International TILs Working Group.[17] TIL data were available from 1315 WCHS patients with invasive cancer (920 Blacks and 395 Whites). Descriptive characteristics of this patient population are summarized in Table 1.

Gene Expression Profiling of Breast TME

The NanoString PanCancer Immune Panel was used to quantify immune cell subsets in the breast TME from WCHS patients, complemented with samples from the Roswell Park Pathology Network Shared Resource. Tissue samples from 190 Black breast cancer patients were frequency matched to 177 White patients by age at diagnosis and tumor subtypes. Women with HER2-positive and triple-negative breast cancer were oversampled to allow comparisons between White and Black women by subtype. Patient characteristics are summarized in Table 1. Two 10-micron curls were cut from a formalin-fixed parafin-embedded block for RNA extraction.

Normalized and transformed gene expression data were used to estimate the absolute fractions (ie, relative to all types of cells in the bulk tissue) of 10 major immune cell subsets derived from the literature where consensus could be reached across multiple algorithms[18] (total T cells, CD4+ T cells, CD8+ T cells, regulatory T cells, B cells, macrophages, neutrophils, mast cells, dendritic cells, and natural killer cells). The relative fractions of each immune cell type (ie, relative to all infiltrating immune cells) were calculated as the ratio of the absolute fraction to CD3+ cells. In a subset of patients with both pathological and molecular TIL scores, the correlations between the 2 were moderate to strong (Supplementary Figure 1, available online; r = 0.65; P < 2.2 x 10−16).

Publicly Accessible Datasets

We used 2 publicly accessible datasets: TCGA breast cancer subset (n = 1080, including 180 Blacks and 714 Whites)[19] and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC; n = 1904 Whites only).[20] TCGA was used as a validation cohort for racial differences in immune infiltrates identified in WCHS, and TCGA and METABRIC were used for survival analysis of tumor immune phenotypes. In addition, genomic estimates of immune receptor repertoire were obtained from TCGA.[21] Patient selection and inclusion of the studies are depicted in a CONSORT flow diagram (Supplementary Figure 2, available online).

Statistical Analysis

Comparisons of tumor-immune phenotypes between tumors from Black and White patients were conducted using Wilcoxon tests. Multivariable linear regression models were used to derive standardized residuals for race after controlling for hormone receptor status (estrogen receptor and progesterone receptor). Analyses of all-cause mortality with tumor immune phenotypes were conducted using Cox proportional hazards regression, with hazard ratios (HR) and 95% confidence intervals (CI) adjusted for clinical prognostic variables and the assumption of proportionality verified by time-independent Schoenfeld residuals. Analyses of disease-specific mortality were conducted by treating other causes of death as competing risk. The mean follow-up time (range) of TCGA, WCHS, and METABRIC was 27 (0–287) months, 72 (9–153) months, and 115 (0–355) months, respectively. All analyses were 2-sided and performed using R 3.6.1. Multiple comparison error was controlled for using the Bonferroni method, with a family-wise error rate of .05.