The Breast Tumor Immune Microenvironment of DNA Double-Strand Break Repair Pathogenic Variant Carriers Is Enriched with Tumor-Associated Macrophages Available to Purchase

An estimated 5% of patients with breast cancer carry a rare germline pathogenic variant (PV) that predisposes them to breast cancer (1). PVs in specific genes are associated with a higher risk of breast cancer (BRCA1 or BRCA2, 50% lifetime risk; PALB2, 32% lifetime risk) and specific tumor molecular subtypes [estrogen receptor (ER) positive: CHEK2 and ATM; ER negative and triple negative: BARD1, RAD51C, and RAD51; ref. 1]. These genes encode proteins with functional roles in DNA double-strand break (DSB) repair pathways, affirming the role of aberrant or deficient DNA repair in carcinogenesis. Defects in DSB repair mechanisms have been shown to activate the cGAS–STING signaling pathway to generate an immune response that removes the damaged cell (2). Ideally, the adaptive immune response removes DSB repair–deficient cells; however, cancer cells with persistent DSBs can upregulate immune checkpoint proteins to evade the antitumor immune response and continue to proliferate in the inflammatory microenvironment (3–5). For instance, persistent DSBs lead to ATM/ATR/Chk1-dependent upregulation of PDL1 and resistance to cell death despite infiltrating immune cells (5). Further characterizing the tumor immune microenvironment (TIME) of individuals with DSB repair–related PVs can further delineate the mechanisms promoting carcinogenesis in this population.

Differences in TIME phenotypes have been observed across PV carriers of specific DSB-related genes. Tumor infiltrating lymphocytes (TIL), a good prognostic factor in breast cancer (6–8), are enriched in individuals with germline PVs in BRCA1. The relationship between germline BRCA2 PVs and TILs is less clear because of limited studies that differentiate BRCA1 and BRCA2 PV carriers, although tumors from germline BRCA2 PV carriers may have higher sensitivity to immune checkpoint inhibition (9), suggesting that additional factors besides TIL abundance may influence TIME immune checkpoint signaling. ATM and PALB2 PV carriers show no evidence for increased TIL infiltration compared with noncarriers (10, 11). Solinas and colleagues (12) also reported null associations between PV carrier status and tumor TIL density, although more tumors in the combined BRCA1/2 mutant group were found to be TIL positive.

Beyond TIL quantification methods, deconvolution of gene expression data has been used to characterize immune cell abundance and immune signaling pathways in the TIME (13). A recent pancancer study of germline genetic variation and TIME features suggests that both common and rare germline genetic variations shape the composition of the TIME (14). In breast cancer specifically, polygenic risk of autoimmune conditions is inversely associated with IFN signaling in the tumor microenvironment, suggesting that genetic susceptibility can influence inflammation in the TIME (15).

Characterizing the immune cell composition and signaling pathways active in tumors of breast cancer PV carriers may reveal common features underlying their relationship with the TIME. Here, we investigated the breast tumor immune profiles of 33 DSB repair–related PV carriers and 526 noncarriers in the Nurses’ Health Study (NHS; RRID: SCR_019144) and NHSII.

NHS and NHSII are longitudinal cohort studies of US female registered nurses described in detail previously (16). NHS enrolled 121,700 participants, ages 30 to 55 years, in 1976, and NHSII enrolled 116,429 participants, ages 25 to 42 years, in 1989. Participants reporting an incident breast cancer diagnosis were asked to review their medical records, and cases with pathology information were confirmed by medical review (>99%; refs. 17–20). Formalin-fixed, paraffin-embedded (FFPE) tumor blocks were collected prior to treatment and were later used for microarray analysis of tumor RNA expression and IHC analyses (21). Blood samples from 32,826 participants of NHS and 29,611 participants of NHSII were collected from 1989 to 1990 and 1996 to 1999, respectively.

Age at diagnosis was retrieved from the NHS and NHSII questionnaires. ER and HER2 statuses were determined from pathologist assessment of tissue microarrays (TMA); refs. 20, 22). Tumor grade, stage at diagnosis, and tumor morphology were obtained from pathology reports. Genetic principal components for population substructure were based on genotyping data collected previously (23).

All participants provided written informed consent, and the study was approved by the institutional review boards of Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health. The study was conducted in accordance with recognized ethical guidelines (Declaration of Helsinki, Council for International Organizations of Medical Sciences, Belmont Report, and US Common Rule). The study protocol was approved by the institutional review boards of Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health and those of participating registries as required.

RNA extraction from FFPE tissue blocks and gene expression profiling have been described previously (17, 20, 21, 24). In brief, RNA was extracted from 1.0- or 1.5-mm cores from FFPE tissue blocks using the Qiagen AllPrep RNA isolation kit. Gene expression profiling was performed in two batches using Affymetrix Glue Grant Human Transcriptome Arrays (HTA 3.0 pre-release version 2012 to 2014 and HTA 2.0 Affymetrix array 2015 to 2018; Affymetrix) by the Molecular Biology Core Facilities, Dana-Farber Cancer Institute (Boston, MA). Gene expression data were normalized and summarized using Robust Multiarray Average (Affymetrix Power Tools v1.18.0). Data quality was evaluated using Affymetrix Power Tools probeset summarization–based metrics with AUC <0.55 excluded from further analysis. Samples that failed nonoutlier analysis by arrayQualityMetrics v3.24.0 were also excluded (25). Batch effects were corrected using the ComBat package (26). Unannotated probes, probes from the Y chromosome, and low-expression probes were removed from analysis. A total of 869 tumor tissue samples and 17,787 gene-level annotated probesets of coding and noncoding RNAs remained for CIBERSORTx (27) deconvolution analysis and expression signature scores. CIBERSORTx uses a support vector regression approach based on prior training for leukocyte subsets to estimate both relative and absolute quantities of infiltrating cell types. The LM22 deconvolution matrix was used as a reference to estimate the absolute abundance for 22 tumor-infiltrating immune cells. Three cell types returned scores of zero (>99%) and were omitted from further study (naïve CD4⁺ T cells, activated CD4⁺ memory T cells, and gamma-delta T cells). Gene expression signature scores for 68 immune features were computed as shown in Wolf and colleagues (28) and Amara and colleagues (29). In brief, signature scores were calculated as mean, median, principal components, and weighted mean values based on their original methods and were mean-centered with deviations scaled to 1. Gene expression data are available from the Gene Expression Omnibus (GEO: GSE115577) database. Cell enrichment values and gene signature scores are referred to subsequently as immune features.

Germline DNA was extracted from blood samples, and a multigene amplicon-based panel was sequenced as described in Hu and colleagues (1). In brief, dual bar–coded QIAseq (Qiagen) multiplex amplicon-based analysis of 746 targeted regions in 37 cancer predisposition genes was performed to prepare libraries. Samples were pooled and sequenced using the HiSeq 4000 system (Illumina). Variants were evaluated using the Genome Analysis Toolkit HaplotypeCaller tool and the VarDict variant caller tool. Variants met the following criteria: allele frequency <0.01 in cases or <0.003 in gnomAD (or not present in gnomAD); altered allele frequency >0.05 or <0.95, or equal to 1; and read depth of both altered reads and reference reads >5. Variants considered “loss of function” (nonsense, frameshift, consensus splice sites (±1 or 2) in panel genes without an entry in ClinVar were classified as PVs. Missense variants and in-frame deletions were called as variants of unknown significance and were excluded, except those occurring in the first codon. Intronic and synonymous variants were considered benign. Copy-number variants were evaluated individually based on log₂ ratios. Variants reported with reduced penetrance were excluded. Variants considered “likely pathogenic” in the ClinVar database were called “pathogenic.” PVs in 16 DSB-related cancer predisposition genes were considered (ATM, BARD1, BLM, BRCA1, BRCA2, BRIP1, CHEK2, FANCC, FANCM, MRE11A, NBN, PALB2, RAD50, RAD51C, RECQL, and XRCC2) based on their functional roles in the Fanconi anemia–related signaling and homologous recombination pathways (30, 31). No participants had PVs in MRE11A or XRCC2. The 33 individuals with PVs in 14 genes comprised the “PV carrier” group in subsequent analyses.

Of the 559 participants with both gene expression data and PV sequencing in the analysis, 87 participants from the NHSII cohort underwent IHC evaluation of four immune cell markers from a previous nested case–control study (32). In brief, TMAs were constructed at the Dana-Farber Harvard Cancer Center Tissue Microarray core facility (Boston, MA). Three 0.6-mm tissue cores were taken from tumor regions on each tissue block, and IHC staining was performed on TMA sections using antibodies for CD4 (Dako 7310, clone 4B12), CD8 (Dako 7103, clone C8/144B), CD20 (Dako 0755, clone L26), and CD163 (Vector Labs VP-C374, clone 10D6). Staining positivity was assessed using an automated computational image analysis system (Definiens Tissue Studio software). The mean percent positivity was calculated as the average number of positive cells over the total number of cells for three patient cores (32, 33).

Prior to statistical analyses, CIBERSORTx-based immune cell infiltration, immune-related gene expression signature scores, and IHC mean percent positivity were rank-based normal-transformed to account for skewed distributions.

For clustering analysis, Pearson correlation coefficients were calculated for the 87 immune features across the 559 participants. Correlations were converted to distances using a distance metric of 1 – correlation and clustered with the complete agglomerative approach (13, 29).

Linear regression models were fit to test the associations between carrier status of DSB repair–related genes with immune cell infiltration, gene expression signature scores, and mean percent positivity for IHC markers, adjusting for age at diagnosis, NHS study cohort, tumor grade (well-differentiated, moderately differentiated, and poorly differentiated), tumor stage at diagnosis (stage 1, stage 2, and stages 3 and 4), dichotomous ER status (negative/low ER and high ER:ER status were quantified as a percentage of positive cells, and three categories were determined: ER negative: <1%, low ER: 1%–10%, and high ER >10%), the top four genetic principal components, and genotyping array. The ER variable was dichotomized based on the observation that ER-low tumors may behave more similarly to ER-negative tumors than ER-positive tumors in the context of immune response in early breast cancer (34). To account for multiple testing for the 87 immune feature regression models, the threshold for significance of P values was adjusted to P ≤ 5.75 × 10⁻⁴ based on the Bonferroni correction for 87 tests. We define nominal association as a P value between 5.75 × 10⁻⁴ and 0.05.

The NHS/NHSII microarray dataset is publicly accessible in the GEO (GEO; RRID: SCR_005012; GEO #GSE11577). Codes for computing the gene signatures and CIBERSORTx variables from the NHS/NHSII microarray dataset are available at https://github.com/yuxi-liu42/PRS_immune. The targeted sequencing data are controlled access and available through a signed access request to the Database of Genotypes and Phenotypes (dbGaP RRID: SCR_002709; accession: phs002820.v1.p1). Other individual-level data are available via collaborative request and subjected to institutional approval at https://nurseshealthstudy.org/researchers.

To investigate the effect of germline DSB repair–related PV carrier status on tumor immune features, we identified individuals in NHS and NHSII who had data from both gene expression profiling of their breast tumor and targeted germline sequencing of their cancer predisposition genes (N = 559; Fig. 1). Of these 559 individuals, 33 (5.9%) had germline PVs in genes with functional roles in DSB repair (Fig. 2). Of the 33 carriers, 10 (30.3%) had PVs in BRCA2 and 2 (6.1%) had PVs in BRCA1. Two (6.1%) individuals were observed to have PVs in multiple genes, and most carriers (29, 87.9%) had ER-positive disease. Carriers were younger at diagnosis (59 vs. 63 years; P = 0.038), as would be expected, were more likely to be from the NHSII cohort (P = 0.002), and were more likely to have poorly differentiated tumors (P = 0.009; Table 1).

Immune signaling cascades are comprised of many cell types and signaling molecules, and many features within the TIME are correlated. To inspect the correlation among the 87 immune features across the entire sample, we performed unsupervised hierarchical clustering of the immune features to visualize their general relationships (Supplementary Fig. S1). Upon inspection of the dendrogram, we observed two main clusters: one branch which contained multiple clusters of highly correlated features previously described in Amara and colleagues (29) such as IFN-related immune features, TGF-β features, and T cell/B cell–related features, and the second branch which had fewer clusters, and several small clusters of features that had inverse correlations with the TGF-β features. The second branch also included CD8 T cell–related features such as CIBERSORT-inferred CD8⁺ T cells (T.cells.CD8) and expression of CD8A mRNA (CD8A). The branch also contains several of the single-gene expression features such as the immune checkpoint proteins CTLA4, PDL1, and PD1 (CTLA4_data, PDL1_data, and PD1_data, respectively).

We next examined the relationship between DSB repair-related PV carrier status and immune features. In the unadjusted analysis, we observed 3 significant associations and 23 nominal associations for carrier status and immune features (Supplementary Fig. S2; Supplementary Table S1), but after adjustment for tumor features and other covariates, there were 20 nominal associations with the largest effect estimates for STAT1_19272155 (standardized β = 0.59; 95% CI, 0.20–1.0; P = 3.5 × 10⁻³) and CIBERSORTx-inferred macrophages.M1 (standardized β = 0.56; 95% CI, 0.22–0.90; P = 1.4 × 10⁻³; Fig. 3; Supplementary Table S2). Several of these nominal associations were with gene expression signatures within the IFN feature cluster (STAT1_19272155, Module3_IFN_score, and Interferon_Cluster_21214954). Given these relationships between carrier status and IFN-related signaling, we also tested the association between expression-based indicators of tertiary lymphoid structures in breast cancer (35) and carrier status (Supplementary Fig. S3; Supplementary Table S3). CXCL13 and LTB expression was positively associated with PV carrier status, but only LTB expression was nominally associated.

We observed an association of PDL1 expression with PV carrier status (PDL1_data: standardized β = 0.39; 95% CI, −0.02 to 0.81; P = 0.06), as we would have anticipated from in vitro studies (5). We also observed a positive association between carrier status and T-cell inhibitory CTLA4 expression (CTLA4_data: standardized β = 0.54; 95% CI, 0.14–0.95; P = 9.0 × 10⁻³). To bolster and validate findings from our gene expression–based analyses of tumor immune features, we analyzed the protein expression of four cell type–specific immune markers from IHC staining of a subset of tumors from NHSII (Supplementary Table S4). Although we observed positive associations of DSB repair–related carrier status with the staining intensity of the four immune markers, these were not statistically significant (Supplementary Fig. S4; Supplementary Table S5).

The carrier group is comprised of individuals with a heterogeneous group of PVs (14 genes), and almost a third of the individuals have a PV in BRCA2 (10, 30.3%). To determine if the observations for the PV carrier group are driven by BRCA2-specific biology, we performed sensitivity analyses among PV carriers without a BRCA2 variant (non-BRCA2 carrier group, N = 17) and those with a BRCA2 variant (BRCA2-only carrier group, N = 8; Supplementary Table S6). Excluding the BRCA2 carriers attenuated the association between inferred macrophage abundance (macrophages.M1) and carrier status (standardized β = 0.48; 95% CI, 0.08–0.88; P = 0.02), but the directionality was consistent (Supplementary Fig. S5, Supplementary Tables S2 and S7). The BRCA2-only subgroup exhibited a stronger effect of carrier status (standardized β = 0.76; 95% CI, 0.18–1.4; P = 0.01), although the CIs were wider because of the reduced sample size (Supplementary Table S8).

In this study, we found that tumors from individuals with germline PVs in at least 1 of 14 DSB repair–related genes were enriched for tumor-associated macrophages (TAM) and IFN-related signaling compared with noncarriers. Carrier status was also associated with higher CTLA4 expression and PDL1 expression but not PD1 expression.

The strongest association between carrier status and an immune feature was STAT1_19272155, an expression signature associated with the response to IFN signaling (28). Both M1 macrophages and the STAT1_19272155 score clustered with IFN signaling–related signatures (Interferon_19272155, IFN_21978456, and Module3_IFN_score), suggesting that these features are occurring within an inflammatory TIME. TAMs are involved in carcinogenesis, metastasis, and response to treatment in breast cancer (36), making their enrichment in the carrier group of particular interest. Further studies establishing the etiologic relationship among inflammation, macrophages, and breast cancer in DSB-related PV carriers is warranted to determine if compromised DSB repair could be a cause or consequence of inflammation in PV carriers.

We also observed relationships between carrier status and immune checkpoint–related features. Both CTLA4 mRNA expression and PD1 mRNA expression have been correlated with TILs and BRCA1 mutation carrier status in individuals specifically with triple-negative subtype disease in The Cancer Genome Atlas (8), but the present population contains patients with mostly high ER tumors and has few BRCA1 carriers, so the lack of PD1 enrichment may be due to differences in specific tumor features. It was recently reported that BRCA2 carriers had improved responses to immune checkpoint inhibition compared with BRCA1 carriers (9). Mouse models of Brca1 and Brca2 deficiency displayed differential TAM infiltration, which was confirmed in human tumor samples, suggesting that BRCA2-specific biology may be involved in TAM-related phenotypes in the TIME (9). To ensure that our TAM-related observations were not driven by the high proportion of individuals with BRCA2 PVs in the carrier group compared with the few BRCA1 variants and other less frequent PVs, we performed a sensitivity analysis to measure the relationship between carrier status and immune features without BRCA2 carriers. The sensitivity analysis demonstrated that even with the removal of the BRCA2 carriers, there was still a positive effect of carrier status on M1 macrophage infiltration. When only BRCA2 carriers were considered, the effect of carrier status on M1 macrophages and CTLA4 coefficients was even stronger but did not reach statistical significance. Replication of this observation in studies with high numbers of tumors from BRCA2 PV carriers would be of clinical interest.

There are limitations to the generalizability of this study. The study population consists of largely early-stage and ER-positive tumors. The carrier group was small, limiting statistical power to detect associations and preventing stratification by specific predisposition genes and molecular subtypes. Also, the low number of BRCA1 carriers limited the ability to consider BRCA1-specific biology in the analysis. Studies with sequencing of predisposition genes in larger and more diverse cohort studies are needed to confirm these findings and further characterize the DSB repair–related carrier-specific TIME and the functional role of the enriched macrophages, especially as treatment strategies targeting immune phenotypes are developed.

Overall, this study provides evidence for an enrichment of TAMs in invasive, treatment-naïve breast cancers of individuals with germline PVs in DSB repair–related genes, which becomes increasingly relevant as the field considers precision medicine approaches around breast cancer prevention and treatment for this high-risk population.

R.L. Kelly reports funding from the Cancer Prevention Fellowship Program. R.M. Tamimi reports grants from NIH/NCI during the conduct of the study, as well as being a consultant for Sterigenics. P. Kraft reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

R.L. Kelly: Conceptualization, formal analysis, writing–original draft. Y. Liu: Conceptualization, data curation, formal analysis, supervision, writing–original draft, writing–review and editing. A.R. Harris: Writing–original draft, writing–review and editing. C. Peng: Data curation, writing–review and editing. Y.J. Heng: Data curation, writing–review and editing. G.M. Baker: Data curation, writing–review and editing. D.G. Stover: Data curation, writing–review and editing. R.M. Tamimi: Resources, data curation, funding acquisition, project administration, writing–review and editing. P. Kraft: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing.

The authors would like to acknowledge the contribution to this study from central cancer registries supported through the Centers for Disease Control and Prevention’s National Program of Cancer Registries and/or the NCI’s Surveillance, Epidemiology and End Results Program. Central registries may also be supported by state agencies, universities, and cancer centers. Participating central cancer registries include the following: Alabama, Alaska, Arizona, Arkansas, California, Colorado, Connecticut, Delaware, Florida, Georgia, Hawaii, Idaho, Indiana, Iowa, Kentucky, Louisiana, Massachusetts, Maine, Maryland, Michigan, Mississippi, Montana, Nebraska, Nevada, New Hampshire, New Jersey, New Mexico, New York, North Carolina, North Dakota, Ohio, Oklahoma, Oregon, Pennsylvania, Puerto Rico, Rhode Island, Seattle Surveillance, Epidemiology, and End Results Registry, South Carolina, Tennessee, Texas, Utah, Virginia, West Virginia, and Wyoming. The authors would like to acknowledge Audrey Goldbaum for review and critical discussion of the manuscript. This work was supported by NIH/NCI grants R01CA260352 (to R.M. Tamimi), UM1CA186107, P01CA087969, R01CA49449, U01CA176726, R01CA67262, R01CA50385, U19CA148065, and R01CA166666. R.L. Kelly and A.R. Harris are supported by the NCI Cancer Prevention Fellowship Program.