Results
Pathological TIL Scores
Among WCHS patients, TIL scores were higher in hormone receptor–negative than hormone receptor–positive tumors; no associations with HER2 status were found after adjusting for hormone receptor status (Supplementary Figure 3, available online). In comparisons by race overall and by hormone receptor subtype, tumors from Black women had higher pathological TIL scores than those from White women (all P ≤ .001) (Figure 1, A), with similar associations noted in molecular estimates of TILs in a subset of the patients (all P ≤ .03) (Figure 1, B).
Figure 1.
Pathological and molecular estimates of tumor-infiltrating lymphocytes in Black and White breast cancer patients in the Women's Circle of Health Study. A) Boxplots of pathological tumor-infiltrating lymphocytes (TIL) scores by race, with and without stratification by tumor hormone receptor (HR) status. The bar in the middle of a box indicates the subgroup median, and the lower and upper edges indicate the first and third quartiles, respectively. The extended lines indicate the range in each subgroup. P values were derived from 2-sided Wilcoxon test between Black and White patients. B) Boxplots of molecular TIL scores estimated based on immune panel gene expression data, with and without stratification by tumor hormone receptor status.
Higher pathological TIL scores in Black patients in WCHS were associated with all-cause and disease-specific mortality (Supplementary Table 3, available online). Patients with lymphocyte-predominant breast cancer (TIL score ≥50%) had a statistically significantly lower hazard of all-cause mortality (HR = 0.38, 95% CI = 0.19 to 0.76), in comparison to those with lymphocyte-depleted breast cancer (TIL score <10%).
Molecular Estimates of Immune Infiltrates in Breast TME
Consistent with the results of pathological TIL assessment, no associations between HER2 status and the absolute or relative fractions of any immune cell subsets were found after adjusting for hormone receptor status (Supplementary Figure 4, available online). Thus, we relied on hormone receptor status to assess the tumor influences on immune estimates in TME, which is also supported by previous studies demonstrating 2 major etiological subtypes of breast cancer proximated by hormone receptor status.[22] Hormone receptor–negative cancers had statistically significantly higher absolute fractions of all immune cell subsets than hormone receptor–positive cancers, except for mast cells, which differed in an opposite direction (Supplementary Figure 4, available online). In contrast, similar comparisons of the relative fractions of immune cells revealed no differences in any of the immune cells except mast cells. Because the relative fractions were not strongly confounded by tumor hormone receptor status, they were used in subsequent analyses unless otherwise specified.
As shown in Figure 2 and Supplementary Figure 5 (available online), in contrast to lower fractions of neutrophils and dendritic cells in Blacks, the most statistically significant and consistent racial differences in immune infiltrates were higher fractions of CD4+ T cells in Black patients compared with White patients in both WCHS and TCGA. The racial differences in immune cell subsets remained after adjusting for or stratifying by tumor hormone receptor status (Supplementary Figures 6 and 7, available online). In a validation study of CD4+ T cells using immunohistochemical (IHC) staining and WCHS tissue microarrays, there was a moderate-to-strong correlation between IHC and molecular scores of CD4+ T cells (r = 0.59; P < 2.2 x 10−16); higher IHC scores of CD4 were also observed in Black than in White patients, although the differences were restricted to hormone receptor–negative cancer subtype (Supplementary Figure 8, available online).
Figure 2.
Racial differences in the relative fractions of immune cells between Black and White breast cancer patients. Relative fractions of the 10 major immune cell subsets plus CD4+ to CD8+ T-cell ratios were centered to an overall mean of 0 before subgroup medians were calculated for Whites and Blacks. The differences in subgroup medians are plotted in bar graphs, with red indicating higher immune cell estimates in Blacks, and blue indicating higher estimates in Whites. P values from 2-sided Wilcoxon test were log10-transformed, negated, and plotted as bar graphs along the median differences. The red reference line indicates a cutoff of statistical significance level after adjusting for multiple testing [-log10(0.05/11) = 2.3]. Patient populations from Women's Circle of Health Study (WCHS) and The Cancer Genome Atlas (TCGA) were included for cross-validation.
Among CD4+ T-cell subsets, the fraction of T follicular helper (TFH) cells, as estimated by CIBERSORT algorithm,[23] showed the most statistically significant difference between Black and White patients in WCHS and TCGA (Supplementary Figure 9, available online). In a recent pan-cancer TCGA analysis, it was reported that rs3366 in the 3' untranslated region (UTR) of SIK1 was associated with the fraction of TFH cells present in bulk tumor tissues.[24] We thus assessed the racial differences in the distribution of this variant. According to gnomAD data,[25] the allele associated with a higher TFH fraction had a markedly higher frequency in Black than in White populations (0.45 vs 0.09). This corroborated our findings of higher fractions of TFH cells in Black than in White breast cancer patients.
Consistent with a stronger CD4+ T-cell response, B-cell fractions and B-cell receptor (BCR) clonality were also higher in Black than White patients in TCGA, but not T-cell receptor clonality (Supplementary Figure 10, available online). Moderate-to-strong correlations were noted between the fractions of CD4+ T cells and B cells and BCR diversity and richness (Supplementary Figure 11, available online).
T-cell Exhaustion Markers and "ExCD8-r" Signature
Because no consistent racial differences were seen in the CD8+ T-cell fractions (Figure 2), the major immune cell subtype with antineoplastic activity, we examined the expression of 4 major inhibitory receptors (IRs: PD-1, LAG-3, CTLA-4 and TIGIT) and the PD-1 ligand. The co-expression of these markers represents cardinal features of T-cell exhaustion and serves as targets of immune checkpoint inhibitors.[26] As expected, these IRs and the PD-1 ligand displayed a strong co-expression pattern (Supplementary Figure 12, available online). When the expression levels were analyzed relative to the absolute fraction of CD8+ T cells, PD-1, LAG-3, and CTLA-4 were statistically significantly higher in breast tumors from Black patients in comparison with White patients in both WCHS and TCGA (all P < .001; Figure 3).
Figure 3.
Racial differences in T-cell exhaustion markers and ExCD8-r signature between Black and White breast cancer patients. The differences between Blacks and Whites in the expression levels of T-cell exhaustion markers relative to the absolute fractions of CD8+ T cells, as well as the ExCD8-r signatures, are plotted in bar graphs, with red indicating higher immune cell estimates in Blacks, and blue indicating higher estimates in Whites. P values from Wilcoxon test are log10-transformed, negated, and plotted as bar graphs along the median differences. The red reference line indicates log10(0.05/6) = 2.1. Patient populations from Women's Circle of Health Study (WCHS) and The Cancer Genome Atlas (TCGA) were used for cross-validation. PD-1 = programmed cell death protein 1; PD-L1 = PD-1 ligand; LAG-3: lymphocyte-activation gene 3; CTLA-4: cytotoxic T-lymphocyte-associated protein 4; TIGIT = T Cell Immunoreceptor With Ig And ITIM Domains.
To develop an expression signature reflecting the balance between the cytotoxic and exhausted states of CD8+ T cells, we leveraged the co-expression pattern of PD-1 and LAG-3 as hallmarks of T-cell exhaustion, demonstrated in previous studies from our group and others.[27–29] Because Eomesodermin (Eomes) is a key transcription factor upregulated in terminally differentiated exhausted T cells,[30] we defined the resultant signature "ExCD8-r" as the ratio of the aggregated expression of PD-1, LAG-3, and Eomes to the absolute CD8+ T-cell fraction. The signature scores were statistically significantly higher in tumors from Black patients (all P ≤ .002; Figure 3), regardless of hormone receptor status (Supplementary Figure 13, available online), suggesting that the CD8+ T-cell responses in breast tumors from Black women were more likely to assume an exhausted state than in Whites. In analysis of a recently developed T-cell exhaustion signature by Cai et al.,[31] a strong correlation was noted with the signature we developed prior to adjusting for the absolute CD8+ T-cell fraction (r = 0.89; P < 2.2 x 10−16); the signature scores by Cai et al. were also statistically significantly higher in tumors from Black vs White patients in WCHS (Supplementary Figure 14, available online).
Prognostic Value of ExCD8-r Signature
We next examined the relationships of the ExCD8-r signature with patient survival in WCHS, TCGA, and METABRIC with all races combined because of limited sample size for race-stratified analysis. A higher ExCD8-r signature was consistently associated with poorer all-cause mortality (Figure 4) and disease-specific mortality (Supplementary Figure 14, available online). The associations remained statistically significant after adjusting for age and clinical prognostic factors (Supplementary Table 2, available online, for all-cause mortality, and Supplementary Table 3, available online, for disease-specific mortality). Further adjusting for other immune cell subsets in the models had no substantial effects. Upon stratification by hormone receptor status, the prognostic value of ExCD8-r signature was limited to hormone receptor–positive breast cancer and not statistically significant in hormone receptor–negative cancer (Figure 4; Supplementary Tables 2 and 3, available online).
Figure 4.
Kaplan-Meir survival curves of all-cause mortality by the levels of ExCD8-r signature. Kaplan-Meier survival curves of all-cause mortality by the levels of ExCD8-r signature are plotted for Women's Circle of Health Study (WCHS), The Cancer Genome Atlas (TCGA), and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) datasets, with and without stratification by tumor hormone receptor (HR) status. The signature levels in WCHS were categorized into binary based on the median because of limited sample size and categorized into thirds in TCGA and METABRIC datasets with large sample sizes. P values from log-rank test are displayed within the plots. The medians and ranges of the follow-up (F/U) time are shown at the bottom. The analyses were conducted with all races combined because of limited sample size of Black patients in the cohorts.
Because high absolute fractions of CD8+ T cells have been consistently associated with better prognosis among hormone receptor–negative breast cancer patients,[32] we sought to determine the prognostic value of ExCD8-r signature in the context of CD8+ abundance. Notably, the absolute fractions of CD8+ T cells were statistically significant only in hormone receptor–negative patients in METABRIC but not in WCHS or TCGA, possibly because of limited sample size. In METABRIC, when CD8+ T-cell fractions were combined with the ExCD8-r signature, patients classified as CD8lowExCD8-rhigh had the highest all-cause mortality and disease-specific mortality; although unexpectedly, those classified as CD8highExCD8-rhigh, instead of CD8highExCD8-rlow, had the lowest mortality (Supplementary Figure 15 and Supplementary Table 4, available online).
Lastly, we compared the proportions of the 4 subgroups defined by the combination of the absolute fractions of CD8+ T cells and the ExCD8-r signature between Black and White patients. As shown in Figure 5, in both WCHS and TCGA, a CD8lowExCD8-rhigh profile was the most common subgroup in Black patients, in contrast to the CD8highExCD8-rlow profile being the most prevalent in White patients. Even among those with high CD8+ T-cell fractions, Black patients were still more likely than White patients to have tumors with CD8highExCD8-rhigh features. Similar findings were observed when stratified by tumor hormone receptor status (Supplementary Figure 16, available online).
Figure 5.
Breast cancer patient subgroups defined by the combination of the absolute CD8+ T-cell fractions and the ExCD8-r signature in Blacks and Whites. Stacked bar graphs of the proportions of the 4 patient subgroups were defined on the basis of the combination of the absolute fractions of CD8+ T cells and the ExCD8-r signature in Black and White breast cancer patients. The bar height of each subgroup corresponds to the percentage of that group within the racial group. P values from χ2 test between subgroup and race are displayed for Women's Circle of Health Study (WCHS) and The Cancer Genome Atlas (TCGA) datasets.
J Natl Cancer Inst. 2021;113(8):1036-1043. © 2021 Oxford University Press
