The Landscape of Immune Microenvironments in Racially Diverse Breast Cancer Patients

The tumor microenvironment plays a major role in the clinical course of breast cancer. Clinical trials have shown that high levels of tumor-infiltrating lymphocytes (TIL), consisting primarily of cytotoxic [CD8-positive (CD8⁺)] T cells, CD19-positive (CD19⁺) B cells, and a small population of natural killer (NK) cells (1, 2) positively predict therapeutic response in triple-negative (TNBC) and HER2-positive (HER2⁺) breast cancer (3–5). Gene expression surrogates of TILs and immune biomarkers have corroborated these findings (6, 7). However, few studies have evaluated immune response in diverse patient populations (8, 9). Black women and young patients may have unique immune responses (10–13), but are often underrepresented in clinical studies. Furthermore, both groups experience higher mortality rates than older and non-Hispanic White women (14–16) and are more likely to be diagnosed with basal-like and TNBC subtypes (9, 16–18), which tend to be more immune-infiltrated (6).

Several studies have shown increased immune infiltrates in tumors from Black patients with breast cancer (19–24), but studies have conflicted. Resolution of this literature has been challenging due to focus on small numbers of immune-cell–specific markers, and smaller sample sizes of Black and young women, which has limited ability to simultaneously consider the role of tumor subtype, grade, and age. Prior studies have also emphasized tumor banks and clinical trials, which tend not to include earlier stage, smaller tumors that are an important part of the clinical population of breast cancers. In light of intensive ongoing research on immune-targeting therapies, studies clearly defining the tumor immune landscape among clinically and racially diverse patient populations and with a broad panel of immune markers are needed to develop a clearer picture of the immune landscapes of breast cancers.

Here, we used gene expression profiling and histologic approaches to characterize the breast cancer immune microenvironment, leveraging data from the Carolina Breast Cancer Study (CBCS; N = 1,952 cases), a population-based study that oversampled Black (n = 1,030) and younger (n = 1,039) women. We selected 48 RNA-based markers indicative of 10 major cell types [B cell, T cell, CD8 T cell, Th cell, regulatory T cell (Treg), T follicular helper (Tfh), eosinophil, neutrophil, NK cell, and macrophage] to evaluate overall global patterns of immune response and to assess the role of immune gene expression in recurrence within a diverse, population-based sample.

The CBCS (25) is a three-phase population-based study that utilized rapid case ascertainment with the North Carolina Central Cancer Registry to identify women, ages 20 to 74 years across 44 counties diagnosed with first primary breast cancer from 1993 to 1996 (phase I), 1996 to 2001 (phase II), and 2008 to 2013 (phase III). Black and younger women (<50 years) were oversampled using randomized recruitment (25). Of 4,806 breast cancer cases enrolled, 1,952 bulk tumor samples were profiled by Nanostring (phase 1: N = 252; phase 2: N = 454; phase 3: N = 1,246) after exclusions for depleted tissue (n = 1,188) or low-quality RNA (n = 241). Samples with depleted tissue and degraded RNA were predominantly from the older, phase I study where fewer sections were collected and stored in suboptimal conditions for RNA isolation. This study was approved by the University of North Carolina at Chapel Hill (UNC-CH; Chapel Hill, NC) School of Medicine Institutional Review Board in accordance with the revised U.S. Common rule, and participants provided written informed consent.

Health history, demographic variables, and measurements for body mass index (BMI) were collected by a nurse during in-home interviews. Race was self-reported and categorized as White/non-Black or African-American/Black; less than 5% of non-Black participants self-identified as multiracial, Hispanic, or other race/ethnicity and were grouped with non-Black for statistical analyses. While genetic ancestry and self-reported race are strongly concordant in CBCS (26), we interpret race herein as a social construct, representing the culmination of biological, social, and environmental exposures. Tumor size, American Joint Committee on Cancer (AJCC) stage, estrogen receptor (ER), progesterone receptor (PR), HER2 receptor, node status, and tumor grade were obtained from medical records, pathology reports, and IHC staining performed at UNC-CH. Tumor grade was assigned by a pathologist in phases I and III. For grade adjustment analyses, missing grade (474/1,952) was imputed with the Multivariate Imputation by Chained Equations package (27), incorporating ER/PR/HER2 status, node status, race, age, tumor stage, size, p53 mutation status, survival, grade, and study phase as predictor variables, using the method described by Ali and colleagues (28). In a sensitivity analysis including clinically assigned grade and a missing value indicator, relative frequency differences (RFD) and 95% confidence intervals (CI) remained stable relative to imputed grade (Supplementary Table S1). Patient characteristics are described in Table 1. Sample percentages are displayed both unweighted and weighted to original North Carolina demographics to account for the sampling design of CBCS, which oversampled Black and younger women using randomized recruitment. Sampling weights were set based on incidence to ensure equal proportions of younger Black, older Black, younger non-Black, and older non-Black participants (18).

Recurrence data were available for CBCS phase III (2008–2013; n  =  1,246). Recurrence-free survival (RFS) was defined as the time between date of diagnosis to first local, regional, or distant recurrent breast cancer and verified through medical record review. Recurrence data are complete through October 2019 with 5-year follow-up for all women. Among 1,246 eligible women, 47 participants were stage IV at diagnosis and excluded from recurrence analysis. Among 1,199 patients (stage I–III), 143 recurrences were identified.

RNA was isolated from bulk tumor tissue using the Qiagen formalin-fixed, paraffin-embedded (FFPE) RNeasy isolation kit and assayed using Nanostring nCounter technology as previously described (18). Multiple code-sets, including the PAM50 molecular subtype predictor (29) and an immune expression panel were used; therefore, we utilized Remove Unwanted Variation (RUV) to harmonize across batches as previously described (Supplementary Table S2; ref. 30). PAM50 molecular subtyping was performed using a research version of the predictor to classify tumors as Luminal A, Luminal B, HER2-enriched, basal-like, or normal-like, and to generate risk of recurrence scores (ROR-PT) incorporating tumor size, proliferation, and subtype (18, 29). We also curated a 48-gene panel of immune markers based on previous work (31, 32), representing 10 major cell types from both adaptive and innate arms of the immune system (B cell, T cell, CD8 T cell, T helper cell, Treg, Tfh, eosinophil, neutrophil, NK, and macrophages), cytotoxic cells and PD-L1 (CD274; Supplementary Table S3).

Three tiers of immune variables were considered in this study: (i) global immune clusters based on clustering across all immune genes; (ii) adaptive-cell versus innate-cell scores calculated across multiple cell types based on median expression; and (iii) individual cell-type scores calculated based on median expression across cell-type–specific genes.

For each participant, 10 cell-type–specific scores were calculated for all n genes related to a given cell type (e.g., B-cell genes, n = 7), in addition to scores for cytotoxic cells, adaptive cells, innate cells, and PD-L1, as listed in Supplementary Table S3 (31, 32). The median and average log₂ expression (computed for each participant across the n genes) were similar and the median was ultimately selected to minimize skew due to extreme values. Adaptive-cell scores were calculated by computing the median log₂ expression among all genes related to B cells, T cells, CD8 T-cells, Th cells, Tregs, and Tfh cells, and an innate-cell score was calculated by computing the median log₂ expression among all genes related to eosinophil, neutrophil, NK, and macrophages. Cytotoxic-cell genes and PD-L1 (CD274) were not included in adaptive-cell and innate-cell scores due to expression of these markers on cells from both arms of the immune system (33). In a validation experiment, we used immunofluorescence and protein-based digital spatial profiling (DSP; ref. 34) to assess concordance between RNA-based and protein-based measurements (Supplementary Fig. S1). Immunofluorescence-based CD19 was positively correlated with RNA-based CD19 quantification and RNA-based B-cell scores (Supplementary Fig. S1A and S1B). Similarly, RNA-based ICOS and CD8A expression was positively correlated with DSP-based expression (Supplementary Fig. S1C and S1D).

For each tumor, we also assigned a single global immune class. Global immune classes (clusters) were based on clustering analysis in CBCS and The Cancer Genome Atlas (TCGA), and used to group tumors based on similarity in their immune-related gene expression patterns across all 48 immune genes in our panel. Due to differences in RNA expression platforms [i.e., NanoString versus RNA sequencing (RNA-seq)], the scope of immune genes present, and sample population, we began with independent immune class discovery in CBCS and TCGA to validate use of our immune panel. To ensure stability of these global immune clusters in each dataset, the ConsensusClusterPlus Bioconductor package (35) was used to run 1,000 clustering iterations with 90% subsampling, the Pearson distance metric and average linkage method. Gene expression was median-centered and visualized using the ComplexHeatmap R package (36). To explore relationships with tumor and patient characteristics, we also developed a classifier of CBCS immune clusters using Classification to the Nearest Centroid (ClaNC; ref. 37) and applied to TCGA.

The Genie algorithm from Aperio's digital pathology software (Leica Biosystems) was trained to digitally quantify TILs from hematoxylin and eosin (H&E)-stained tissue microarrays, excluding cores with degraded tissue, more than 50% red blood cells, or cysts (n = 996 with RNA immunoprofiling). The tissue classifier was trained using a representative feature library of manually annotated epithelium, stroma, adipose and immune (TILs) tissue compartments, and optimized through iterative rounds of adjustment parameter modification and visual assessment. Each quantified compartment area was then divided by the total tissue area per case and multiplied by 100. Reproducibility of digital lymphocyte quantification was evaluated by a study pathologist, where high agreement was found between digital and pathologic review (38). Percentage of TILs in tissue was considered as a continuous variable and log₂-transformed for analysis.

Comparison of expression levels and TILs across global immune clusters was performed using ANOVA with Tukey multiple comparisons test and Welch two sample t tests. Generalized linear models (glm) were used to calculate RFDs as the measure of association between immune clusters and covariates of interest. RFDs are estimated based on a general linear model, and are interpretable as the percentage difference between index and referent groups. Multivariable models were adjusted for age and race in reduced models, and additionally adjusted for tumor grade in full models. Note that in reduced models comparing age or race, age comparisons were only adjusted for race, and race comparisons were only adjusted for age. Kaplan–Meier curves and log-rank tests were used to compare mean time to recurrence across global immune clusters in stage I to III cases (n = 1,199). HRs and 95% CI were calculated using Cox proportional hazard models, and adjusted for patient age, race, and tumor stage. The assumption of proportionality was assessed via the Wald p-value. There was evidence of nonproportional hazards, however point estimates from models that included covariate-time interaction terms did not differ substantially from the model without the time interaction term. All statistical analyses were performed in R version 4.0.3.

TCGA breast cancer dataset, including 1,095 primary tumors, is publicly available under dbGaP accession phs000178.v1.p1. TCGA breast cancer dataset was used to validate the use of our immune panel in breast cancer samples, and leveraged for the availability of multiple data platforms for each case, including RNA-seq, leukocyte-specific DNA methylation markers (39), and histologic TIL quantification by study pathologists. These data and description of related methods are available at https://gdc.cancer.gov/about-data/publications/PanCan-CellOfOrigin (39), with patient characteristics described in Table 1. CBCS data are available upon request (https://unclineberger.org/cbcs).

We evaluated immune gene expression in two datasets [CBCS (n = 1,952) and TCGA breast cancer (n = 1,095)], that differed according to clinical and demographic variables. The population-based CBCS sample was comprised of 53.2% young women (<50 years) and 52.8% Black participants, while 38.9% of tumors were classified as low-grade, 33.6% low-stage, 56.8% node-negative 62.9% ER-positive (ER⁺), and 27.5% basal-like (Table 1). Compared with TCGA, CBCS had higher proportions of young (<50 years) and Black participants, and higher proportions of low-stage, node-negative, ER-negative (ER⁻), and basal-like tumors (Table 1). After accounting for randomized recruitment, the distribution of molecular tumor subtypes was similar between both studies, but younger age, low stage, node-negative, and ER⁻ remained more prevalent in CBCS.

We identified three stable global immune clusters in CBCS using consensus clustering with our 48-gene panel: (i) adaptive-enriched, (ii) innate-enriched, and (iii) immune-quiet (Fig. 1A). Tumors in the adaptive-enriched cluster displayed the highest median immune expression (Fig. 1B) and was characterized by the highest expression of the overall adaptive-cell score, and highest levels of B-cell, Tfh, Treg, Th cell, T-cell, CD8 T-cell, and PD-L1 (CD274) scores (Supplementary Fig. S2). The innate-enriched cluster had the highest eosinophil and neutrophil scores (Supplementary Fig. S1) and the highest overall innate-cell score expression. The immune-quiet cluster had the lowest overall immune expression (Fig. 1B), including the lowest adaptive-cell and innate-cell score expression, but displayed significantly elevated macrophage scores (Supplementary Fig. S2). Corresponding with pathologic evaluation, TILs were significantly higher in adaptive-enriched tumors compared with immune-quiet (P = 0.00001) and innate-enriched (P < 0.000001; Fig. 1C and D). Thus, these clusters represent both overall immune expression patterns and cell-type–specific differences in immune response.

Given the availability of the full CIBERSORT 547-gene immune deconvolution panel in TCGA RNA-seq data (40), we compared our classification with CIBERSORT-based estimation, filtering to the cell types represented in our targeted immune panel. CIBERSORT expression patterns mirrored those in our targeted panel (Fig. 2A, lower panel). However, in independent analysis, only two stable immune clusters were identified in TCGA: overall Immune-High and Immune-Low (Fig. 2A and B), which could be reflective of differing tumor and demographic characteristics in this dataset. The Immune-High group shared features of the CBCS adaptive-enriched cluster, with higher DNA methylation-based estimates of leukocytes (Fig. 2C; ref. 39), and higher TIL counts (Fig. 2D and E).

Because the TCGA seemed not to include the immune-quiet cluster based on unsupervised clustering in independent discovery, we used CBCS centroids to identify all three immune classes in TCGA. Distance to centroid showed that 85.5% of Immune-High tumors were classified as adaptive-enriched, while 87.5% of Immune-Low tumors were innate-enriched or immune-quiet. The adaptive-enriched cluster was found in nearly half (n = 489, 44.7%) of TCGA tumors, while innate-enriched was the other dominant class (n = 532, 48.6%). The immune-quiet cluster was rare in TCGA (n = 74, 6.8%) and similar to CBCS, consisted predominantly of low stage (I/II) and Luminal A tumors (Supplementary Fig. S3).

We compared our three clusters to six published immune-related subtypes identified using 160 validated immune signatures in TCGA PanCancer (41). The adaptive-enriched cluster had the highest frequency of “C2-IFNγ-dominant” and “C3-Inflammatory” subtypes. Innate-enriched and immune-quiet were associated with “C1-Wound-healing” and “C4-Lymphocyte-depleted”. In addition, the immune-quiet had the highest frequency of the rare (3% prevalence in TCGA PanCancer) “C6-TGFβ-dominant” subtype, which is characterized by an immunosuppressive phenotype (Supplementary Fig. S3).

We evaluated associations between immune clusters and patient age at diagnosis, race, tumor grade, stage, node status, and BMI in CBCS. Relative to immune-quiet, the adaptive-enriched cluster was associated with young age, high grade, and low BMI, while both adaptive-enriched and innate-enriched were associated with Black race (Fig. 3). Adaptive-enriched and innate-enriched clusters remained significantly associated with Black race when also adjusting for tumor grade, but associations between adaptive-enriched, young age, and BMI were attenuated. There were no significant associations with node status or tumor stage.

Global immune clusters were strongly associated with both clinical and molecular breast cancer subtypes. Adaptive-enriched was associated with IHC-based HER2⁺/HR-negative (HR⁻) (HER2⁺) breast cancer, and both adaptive-enriched and innate-enriched were strongly associated with TNBC, the RNA-based basal-like subtype, and high ROR-PT scores (Fig. 4). Combined race and age adjustments were not possible in PAM50 and ROR-PT models due to high collinearity. However, we performed a sensitivity analysis restricting to ER⁺ tumors, since this subtype is known to be less immunogenic (6). Among ER⁺ tumors only, adaptive-enriched remained strongly associated with Black race and high grade, but not age (Supplementary Fig. S4). Conversely, young age was associated with both adaptive- and innate-enriched clusters among ER⁻ tumors, despite race and grade adjustments (Adaptive RFD [95%CI]:14.2 [2.4, 25.9]; Innate: 14.2 [2.3, 26.0]).

The CBCS identified 143 recurrences during the first 5 years of follow up and we assessed associations between the three immune clusters and recurrence using both Kaplan–Meier analyses and multivariate Cox proportional hazards models. Results underscore the importance of ER as a modifier. Considering all tumor subtypes, immune-quiet and adaptive-enriched tumors were associated with improved RFS, while innate-enriched had the poorest RFS (Fig. 5A). However, after stratification by ER status, significant associations were limited to ER⁻ tumors (Fig. 5B and C), where adaptive-enriched tumors had the best RFS and innate-enriched tumors had the poorest RFS.

This study investigated the breast cancer immune microenvironment in a large and diverse population-based study and identified a novel class of immune response that is immune-quiet. This subtype was present at very low prevalence in TCGA, emphasizing that diverse cohorts representing the full range of tumor phenotypes are valuable for understanding the diversity of immune response. In this racially diverse cohort, we also showed that Black women had higher frequencies of adaptive-enriched and innate-enriched tumors. These racial differences persisted in ER-stratified analyses, suggesting that they are not driven exclusively by subtype and may reflect other race-associated exposures or stressors. Young age and high grade were also associated with adaptive response. Immune response differences showed the strongest relationships with recurrence among ER⁻ cancers.

Our results showing associations between immune response and tumor molecular subtypes are in line with previous literature, where the highest immune expression levels were observed in aggressive tumors [TNBC, basal-like, HER2-enriched subtypes (6), high ROR-PT scores, and high grade (42)]. Several smaller studies have reported immunologic differences between Black and non-Black patients with breast cancer suggesting elevated immune infiltrates in tumors from Black women (19–24). Here, we observed strong and independent associations between race and the immune microenvironment; but race differences were consistently smaller in magnitude than those for grade and subtype. Our finding of increased adaptive-enriched expression was consistent with a smaller study by Yao and colleagues (23) that identified higher TILs in tumors from Black women while matching on age and subtype. However, adaptive immune responses in cancer are complex, with conflicting associations between the presence of certain lymphocyte populations and patient outcomes (43–46). Thus, further delineation and spatial evaluation of the distribution of specific immune cell populations may be needed to resolve some of the conflicting studies. We also observed a high frequency of innate-enriched tumors in Black women. This cluster may be particularly important, as it was associated with aggressive subtypes/high ROR-PT scores and had the highest recurrence hazards in our study.

Studies of tumor immune microenvironment have emphasized clinical features, but herein we also assessed immune differences by age and BMI, as both can systemically impact immune function (11). Building upon previous work in rodent models (12), young age was associated with the strongest immune response among ER⁻ tumors. Conversely, tumors from patients with BMI ≥ 25 were more frequently immune-quiet. Previous studies have suggested that high BMI is associated with increased macrophage infiltration, and we found evidence that immune-quiet tumors had high macrophage infiltration (while lacking other innate immune cell signals). Thus, the association of high BMI with this cluster appears consistent with multiple studies linking obesity with macrophage-mediated breast cancer pathogenesis (47–49). Identification of other social or institutional variables that impact immune phenotypes is important, particularly in understanding race as a social construct. As such, disentangling race, age, and a larger range of individual and community-level variables is an important future direction.

While our study recapitulated previous findings emphasizing abundance of immune-cell infiltrates (i.e., TILs) showing that robust adaptive response predicts lower recurrence among ER⁻ tumors (3–5), we also present novel data suggesting that the character of immune response, not just abundance, is important in breast cancer outcomes. Specifically, we show that innate-enriched tumors had the poorest RFS. While the innate-enriched group had the lowest lymphocyte-related expression, this finding also suggests that the patterns of specific innate cell types may be important, particularly given that the immune-quiet cluster also had lower lymphocyte expression but did not convey the poorest survival. Previous studies have found some associations between innate immune cell expression, poor survivorship (50), and resistance to neoadjuvant chemotherapy in breast cancer (51). Our data extends those previous findings in largely Caucasian tumor bank studies to a population-based cohort enriched for younger and Black women. Given that Black race was strongly associated with both adaptive-enriched and innate-enriched clusters, these data suggest that some Black women may be candidates for immune-checkpoint blockade. However, in-depth investigation of the role of innate immune cells in breast cancer is needed to address the high prevalence of poor-prognosis innate-enriched tumors among this patient population.

The CBCS and TCGA differ in that TCGA is skewed toward more late stage, large breast tumors due to the minimum tissue requirements, and has larger proportions of older, white women (PMID: 30131556; ref. 52). In contrast, CBCS represents a broader and more natural distribution of stage in the population, including an increased frequency of small, early-stage tumors. In line with our finding that immune-quiet tumors tended to be smaller and ER⁺, our results suggest higher proportions of immune-quiet tumors in CBCS, making this cluster more readily discernable. These data suggest this immune subtype would be higher in clinical populations than predicted based on TCGA breast cancer data.

A strength of our analysis was use of a large, population-based cohort for which we optimized a custom immune cell-focused code-set suited to FFPE specimens. This targeted approach may miss some rare cell types (i.e., mast cells) and differs from immune panels focused on activation/exhaustion states. Nevertheless, our custom panel has twice the number of innate cell-specific genes than the commonly used nCounter Breast Cancer 360 Panel (BC360). This balanced inclusion of cell markers resulted in strong representation from both adaptive and innate pathways and correlated with large and validated immune signatures (40, 41). However, future studies of cell-type distribution and spatial arrangement of cell types may be valuable. Future studies, with larger recurrence rates or longer follow-up times, could also consider how subtype, age, race, and immune response interact on a more granular level (e.g., stratifying on age, race, and subtype).

Given that younger patients and Black women are more frequently diagnosed with aggressive breast cancer subtypes and have higher burden of poor outcomes, it is important to understand immunologic differences in these patients. Our discovery of a novel immune-quiet cluster with 27% prevalence in a population-based cohort also suggests that it is important to study immune response in diverse cohorts. In addition, methods that utilize cell-type–specific markers are important because of distinct survivorship patterns among ER-negatives that depend on the dominant immune phenotype present.

Y. Abdou reports other support from Exact Sciences (consultant) outside the submitted work. J.S. Serody reports grants from Merck Inc., GlaxoSmithKline, Carisma; and other support from Tessa Therapeutics during the conduct of the study. C.M. Perou reports grants from NCI Breast SPORE program P50-CA058223 during the conduct of the study; personal fees from Bioclassifier LLC outside the submitted work; in addition, C.M. Perou has a patent for U.S. Patent No. 12,995,459 issued, licensed, and with royalties paid. B.C. Calhoun reports personal fees from Luminex Corp. outside the submitted work. No disclosures were reported by the other authors.

A.M. Hamilton: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. A.N. Hurson: Data curation, methodology, writing–review and editing. L.T. Olsson: Data curation, methodology, writing–review and editing. A. Walens: Methodology, writing–review and editing. J. Nsonwu-Farley: Data curation, writing–review and editing. E.L. Kirk: Data curation, writing–review and editing. Y. Abdou: Writing–review and editing. S.M. Downs-Canner: Writing–review and editing. J.S. Serody: Funding acquisition, writing–review and editing. C.M. Perou: Conceptualization, funding acquisition, writing–review and editing. B.C. Calhoun: Conceptualization, writing–review and editing. M.A. Troester: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing. K.A. Hoadley: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing.

We are grateful to CBCS participants for their generous participation, as well as study staff. The authors would like to acknowledge the UNC-CH BioSpecimen Processing Facility for sample processing, storage, and sample disbursements (http://bsp.web.unc.edu/). This work and the CBCS was supported by a grant from UNC Lineberger Comprehensive Cancer Center (Chapel Hill, NC), which is funded by the University Cancer Research Fund of North Carolina; the Susan G. Komen Foundation (OGUNC1202 and TREND21686258 to M.A. Troester); and the NCI of the NIH (P01CA151135, to M.A. Troester), including the NCI Specialized Program of Research Excellence (SPORE) in Breast Cancer (P50CA058223, to C.M. Perou, J.S. Serody, S.M. Downs-Canner, M.A. Troester, and K.A. Hoadley). In addition, this work was supported by R01CA253450 (to M.A. Troester and K.A. Hoadley), F31CA257388 (to A.M. Hamilton), Komen Career Catalyst grant (CCR16376756, to K.A. Hoadley), and UNC-CH Cancer Control Education Program (T32CA057726, to A Walens). This research recruited participants and/or obtained data with the assistance of Rapid Case Ascertainment, a collaboration between the North Carolina Central Cancer Registry and UNC Lineberger Comprehensive Cancer Center. Rapid Case Ascertainment is supported by a grant from the NCI of the NIH (grant no. P30CA016086). The Pathology Services Core is supported in part by NCI of the NIH Center Core Support Grant (P30CA016080) and the UNC-CH University Cancer Research Fund.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.