Genomic Signatures Predict the Immunogenicity of BRCA-Deficient Breast Cancer

This article is featured in Highlights of This Issue, p. 4197

Mutations in BRCA1 and BRCA2 are the most common causes of hereditary breast cancer. BRCA1 and BRCA2 play essential functions in maintaining genome integrity, primarily through their roles in homologous recombination (HR) and contribution to double-strand DNA break repair (1). Breast cancers associated with germline BRCA1 and BRCA2 mutations have higher sensitivity to DNA-damaging agents including platinum-based chemotherapy and inhibitors of poly(ADP-ribose) polymerase (PARP1), as HR is necessary for repair of DNA crosslinks (1). However, outcomes can vary widely across patients with germline BRCA1 and BRCA2 mutations receiving DNA-damaging agents, which in part may be due to the varying degree of HR deficiency (HRD) in these tumors. Secondary genetic alterations, particularly BRCA1/2 reversion mutations, lead to acquired resistance to these therapies, necessitating the investigation of alternative treatment strategies (2, 3).

Tumors with somatic or germline defects in BRCA1/2 are hypothesized to be more immunogenic than tumors without genetic defects in the HR pathway, making them potential candidates for immune checkpoint blockade (4–7). BRCA1/2 mutation-associated breast cancers have been found to be more genomically unstable than tumors without such genetic alterations (2), with increased numbers of nonsynonymous single-nucleotide variants (SNV) likely driving the heightened immunogenicity observed in these tumors (6, 7). BRCA1/2 deficiency also has been associated with an immune activation signature (6), and histopathologic analyses have found that the level of lymphocytic infiltrates is the strongest positive predictor for recurrence-free survival in BRCA1/2 mutation-associated breast cancer cases (6). Breast cancers with a DNA damage response-deficient phenotype, characterized by a 44-gene expression signature reflecting defects in the Fanconi anemia (FA)/BRCA DNA repair pathway, have been shown to have higher CD4⁺ and CD8⁺ T-cell infiltrates, likely driven by the release of chemoattractive cytokines through STING/TBK1/IRF3 signaling (5).

However, there is a wide range in immunogenicity observed in BRCA1/2 mutation-associated breast cancers, and the molecular underpinnings of that variability have not been delineated (4–8). Although BRCA1/2-deficient breast cancers are generally more immunogenic than sporadic cancers (8), a comprehensive histopathologic analysis of BRCA1/2 mutation-associated breast cancers found that greater than 50% were characterized by “absent/mild” levels of tumor-associated lymphocytes (4). Immune metagene signatures that provide an RNA-based measure of immune cytotoxicity demonstrate that immune gene expression ranges widely among BRCA1/2-deficient breast cancers (6). Further, triple-negative breast cancers (TNBC), the most common tumor type associated with germline BRCA1 mutations, have shown a clinical response rate to PD1 checkpoint blockade monotherapy of only 18.5% (9, 10), underscoring the heterogeneity of BRCA-deficient breast cancers with respect to immunogenicity. Mouse models of Brca1 deficiency have also demonstrated that efficacy of CTLA4 and PD1 therapy in breast cancer may be dependent on concomitant treatment with DNA-damaging agents (7). A pan-cancer analysis of The Cancer Genome Atlas (TCGA) data found that the levels of genomic copy number variations (CNV) inversely correlate with a cytotoxic immune signature and clinical benefit from immune checkpoint blockade (11). Given that BRCA1/2 cancers are characterized by unique genomic scars driven by CNVs (2, 3), this finding raises the question of whether mutational burden or CNV level might predominate in driving immunogenicity in BRCA1/2 mutation-associated cancers.

Herein, we aimed to determine the underlying molecular features of immunogenicity in 115 BRCA1- and BRCA2-deficient breast cancers using genomic data from TCGA, and genomic and histopathologic analyses in a validation cohort (n = 35) from the Hospital of the University of Pennsylvania. We further aimed to stratify BRCA1/2 breast cancer immunogenicity based on these features and understand the molecular processes underlying the immunogenic and nonimmunogenic phenotypes.

Level 1 DNA sequencing data for breast tumors from TCGA were obtained by submitting a project request for approval by the National Center for Biotechnology Information Genotypes and Phenotypes Database (NCBI dbGaP) Data Access Request System Protocol #5309 “BRCA1 and BRCA2 mutations in breast and ovarian cancer.” Breast cancers with controlled-access Level 1 whole exome sequencing (WES) binary alignment map (*.bam) files were identified in the Genomic Data Commons (GDC; https://gdc.cancer.gov; n = 1,098 breast tumors). After filtering samples without RNAseq data or that failed to pass our quality control pipeline (2), 859 breast tumor/normal *.bam files in total were subjected to downstream analysis. RNAseq v2 expression data and patient clinical data for breast invasive carcinoma (TCGA, provisional) were downloaded from cBioportal (http://www.cbioportal.org/). Tumors were categorized as receptor positive or as TNBC using primary data from the Santa Cruz Cancer Genetics Hub (Supplementary Table S1).

DNA sequencing data from a discovery study population ascertained from academic and community hospital sites within Penn Medicine were used for this analysis. Briefly, informed consent was obtained from each participant for use of their samples in genetic studies in an IRB-approved study. Patients with diagnosis of breast cancer and mutations identified though germline BRCA1/2 sequencing in a Clinical Laboratory Improvement Amendments (CLIA)-approved laboratory were eligible. As previously described (2), formalin-fixed paraffin-embedded (FFPE) tumor blocks were macrodissected, sectioned, and stained with hematoxylin and eosin to ensure sections of over 70% invasive tumor. Tumors passing pathology quality control were dissolved in Deparaffinization Solution (Qiagen) and purified using Gentra PureGene Tissue Kit (Qiagen) following manufacturer's protocols. DNA libraries were subjected to WES using the Agilent All-Exon Kit v5, and matched tumor and normal sequencing data for 43 breast cancer patients (BRCA1, n = 25; BRCA2, n = 18) were used for HLA typing.

All WES data from TCGA tumors and matched germline were aligned to the hg38 assembly of the human genome. All WES data from Penn tumors and matched germline were aligned to the hg19 assembly of the human genome. Alignment data underwent initial quality control filtering according to GATK best practices. All exonic germline and somatic SNVs, and germline BRCA1 and BRCA2 mutations were identified and annotated as previously described (2). Exonic variants were annotated as somatic, if they had an alternate allele depth in germline of less than five reads and an alternate allele supported by greater than 10 reads in the tumor. Tumors were determined to be associated with germline BRCA1 and BRCA2 mutations if the germline allelic fraction (AF) >0.30 for the mutation, and the total depth was >30 in germline and tumor at the mutation locus. Only tumors with known pathogenic BRCA1 or BRCA2 mutations as per ENIGMA classification (43) were retained (Supplementary Tables S1 and S2). In total, we identified 35 breast tumors from the TCGA associated with germline BRCA1 (n = 18) or BRCA2 (n = 17) mutations with RNAseq data. We also studied 43 breast tumors with germline BRCA1 (n = 25) or BRCA2 (n = 18) mutations identified from Penn, 35 of which had available tumor for IHC staining (BRCA1 = 23, BRCA2 = 12).

Allele-specific LOH of the germline BRCA1/2 mutation in the tumor was determined through a combination of Varscan2 (44), purity-adjusted allele frequency comparisons, and allele specific copy number calls as previously described (2). Briefly, estimates of tumor purity and ploidy were determined using Sequenza (45) for use in Varscan2 variant calling. Samples with a significant Varscan2 somatic P-value were assigned an allele-specific LOH-positive status and those with a significant Varscan2 germline P-value were assigned an allele-specific LOH-negative status. Sequenza was also used to call allele-specific copy number states in the genomic region containing the BRCA1 or BRCA2 mutation. The copy number of the genomic region surrounding the germline BRCA1 or BRCA2 mutation (CN) and the number of mutant alleles (m) were used to assign LOH states, including allele-specific LOH positive (CN = 1, m = 1; CN = 2, m = 2; or CN > 2, m = 2) and allele-specific LOH negative (CN = 2, m = 1; or CN > 2, m = 1; Supplementary Table S1).

Using our mutation calling pipeline (2), level 3 RNAseq z-scores, microarray Z-scores, and HK27/HK450 methylation β values from The Cancer Genomics Hub (https://gdc.cancer.gov/) for the tumor/normal pairs, we identified 54 additional breast tumors with somatic alterations in BRCA1 or BRCA2 that met the following criteria: (i) presence of a deleterious or likely deleterious single nucleotide or insertion/deletion variant using our previously described variant annotation pipeline (2); (ii) RNAseq z-score < −1.5 for BRCA1 expression (with HK450m >0.5 or data not available); and (iii) BRCA1 or BRCA2 homozygous copy loss by Sequenza analysis (Supplementary Table S1). HR gene mutated breast cancers included those harboring: (i) deleterious or likely deleterious single nucleotide or insertion/deletion variants at other genes involved in HR (12) as defined in Supplementary Table S3; (ii) promoter methylation HK450m >0.25 with RNAseq←1.5 for RAD51C; (iii) PTEN homozygous copy loss or clonal mutation (12); and (iv) EMSY high copy gain (ref. 12; Supplementary Tables S1 and S3). Finally, tumors with no identifiable alteration in BRCA1/2 or other HR-related genes were classified as HR wild-type tumors. HRD scores were determined using custom R-scripts (2) and Sequenza (45), which assigned counts of non-telomeric allelic imbalance (NtAI), large state transitions (LST), and genomic LOH (2, 3).

All SNVs described in ref. 2 were used to calculate mutational burden. We defined mutational burden as the number of nonsynonymous somatic mutations per megabase of genomic DNA. We determined HLA genotypes from germline TCGA and Penn WES files using OptiType (46) requiring a minimal read depth of 40 at the HLA loci; clinical HLA typing in a subset of non-TCGA samples (n = 7) showed complete concordance with OptiType calls (Supplementary Table S4). Neoantigen prediction was accomplished through NetMHCcons (47) called from the Immune Epitope Database Application Programming Interface (IEDB-API). All samples met the minimum read depth requirements and were retained for downstream analysis. Epitope lengths of 8, 9, and 10 were queried from FASTA files for each patient containing peptide sequences of length 15, 17, and 19, respectively. All neoantigens with a predicted IC₅₀ <500 nmol/L and a rank score of less than 2% were retained for comparisons across tumor groups.

Cytolytic index for each tumor was computed as the geometric mean of PRF1 and GZMA expression from RNAseq v2 expression data downloaded from cBioportal (http://www.cbioportal.org/), as noted above. Inference of immune cell infiltration from TCGA RNA-Seq data was accomplished via the “estimate” R package (14). ESTIMATE gives a measure of immune cell infiltration by performing single-sample gene set enrichment analysis (ssGSEA) based on an inferred immune signature. The immune signature represents the intersection between two gene sets: a leukocyte methylation signature gene set and genes enriched for high expression in normal hematopoietic cells.

Gene Set Variation Analysis (GSVA) was implemented via the “gsva” Bioconductor package (48). All curated canonical pathways (C2) were downloaded from the Molecular Signatures Database (49) (http://software.broadinstitute.org/gsea/msigdb/genesets.jsp?collection=CP), and GSVA scores were determined for all breast tumors. Immune metagene gene sets for GSVA were obtained from Rooney and colleagues (13).

Tissue microarrays for Penn breast BRCA1/2 tumors were prepared as previously described (2). IHC was performed using mouse monoclonal antibodies for CD3 (Leica Biosystems, Catalog No. PA0553), CD8 (Dako, Catalog No. M7103), CD20 (Dako, Catalog No. M0755), CD45 (Dako, Catalog No. M0701), CD68 (Dako, Catalog No. M0814), FoxP3 (Biolegend, Catalog No. 320102), and PRF1 (Leica Biosystems, NCL-L-Perforin), and a rabbit monoclonal antibody for PDL1 (Cell Signaling Technology, 15161 BF). Scoring for IHC stains was performed in triplicate for each sample, and normalized counts of CD3, CD8, and PRF1-positive cells were determined by board-certified pathologists (A. Barrett, M. Feldman, and A. Nayak). PDL1 stain intensity was graded on a four-point scale (0–3) by board-certified pathologists (A. Barrett and M. Feldman), with 0 = no staining, 1 = weak or partial staining, 2 = moderate staining, 3 = strong staining. PDL1 H-scores were computed for statistical comparisons across tumor subsets.

Masson's trichrome staining was accomplished with a Staining Kit from Polysciences, Inc. (Catalog No. 25088). Tissue microarray sections were fixed in Bouin's solution at 60°C and washed. Fixed sections were stained with the following reagents: Weigert's Iron Hematoxylin, Beibrich Scarlet-Acid Fuchsin, phosphotungstic acid, and Aniline Blue. At the conclusion of the staining protocol, all sections were rinsed with 1% acetic acid, dehydrated, and mounted for microscopy. Staining intensity was quantified by board-certified pathologists (A. Nayak and M. Feldman) by grading on a four-point scale (0–3) of density and thickness for each trichrome stain (blue: collagen; red: myofibroblasts; black: nuclei). Density and thickness metrics were multiplied for each sample for statistical analysis.

Median HRD-total scores were determined for BRCA1/2 (media, n = 50.65), HR mutant (media, n = 12.08), and HR wild-type (median = −4.58) tumors and used as the basis for dichotomization for each group of tumors. P-values for correlation between gene expression and sample traits (HRD and allele-specific LOH status) were calculated via the two-sided Student t test on log-normalized expression data. Normalized enrichment scores from GSVA were computed by setting “max.diff” to “TRUE,” which provides a unimodal and approximately Gaussian distribution of enrichment scores (48). All linear statistical tests with GSVA scores were implemented via the “limma” package in R (50). T-statistics were adjusted using an empirical Bayesian model, and P-values were adjusted using the Benjamini–Hochberg method.

Cumulative hazard was compared using a Cox proportional hazards model to determine the effects of HRD level in breast patients, adjusting for stage, mutational burden, node positivity, and estrogen receptor expression status. Hazard curves were constructed in R using the “survival” (https://CRAN.R-project.org/package=survival) package

The WES data that supports this study have been deposited in the National Center for Biotechnology Information (NCBI)'s Sequence Read Archive (SRA, https://www.ncbi.nlm.nih.gov/sra) with BioProject ID PRJNA38804 and can be accessed at: http://www.ncbi.nlm.nih.gov/bioproject/388048. The TCGA data are available from the National Cancer Institute's Genome Data Commons (https://gdc.cancer.gov/). The remaining data are available within the article and its Supplementary Information files or available from the authors upon request.

We investigated immunogenicity in breast cancers from TCGA (Supplementary Fig. S1A), including those with germline or somatic alterations in BRCA1 or BRCA2, in non-BRCA1/2 genes involved in HR (HR mutant tumors; ref. 12), and in those without alterations in genes involved in HR (HR wild-type tumors; Supplementary Tables S1 and S3). Mutational burden was higher in breast cancers associated with mutations in BRCA1/2 (P = 0.006) and HR genes (P = 8E−06; Fig. 1A) than HR wild type. BRCA1 expression was significantly lower in cancers with germline or somatic BRCA1 alterations (P = 0.026), and expression of BRCA2 was lower in cancers with BRCA2 germline or somatic alterations with borderline significance (P = 0.087; Supplementary Fig. S1B). We determined neoantigen load (Supplementary Fig. S1C), which was also higher among BRCA1/2 mutation-associated (P = 0.014) and HR gene mutation-associated (P = 0.015) breast cancers than among WT tumors (Fig. 1B). Notably, neoantigen load was not significantly different when comparing BRCA1 versus BRCA2 mutant tumors, tumors with germline versus somatic mutations in BRCA1 or BRCA2, or nascent versus whole genome amplified (WGA) DNA (Supplementary Fig. S1D–S1F); thus, all were combined into a single BRCA1/2 tumor group for further analyses.

We determined the HRD score for each tumor (2, 3), as described previously (2). We found that HRD score was higher among BRCA1/2 mutation-associated versus HR gene mutation-associated (P = 1.28E−09) or HR gene wild-type breast cancers (P = 3.03E−22; Supplementary Fig. S1G). Dichotomizing tumors by median HRD score, we found that high HRD scores significantly correlated with increased neoantigen load amongst BRCA1/2 (P = 0.02) and HR gene mutation-associated breast cancers (P = 0.016; Fig. 1C). Notably, HRD scores and estimated tumor ploidy also correlated with focal CNV levels and chromosome/arm CNV levels, respectively, as computed in a pan-TCGA cancer analysis (11) (Supplementary Fig. S2A). BRCA1/2 cancers with higher HRD tended to exhibit a greater degree of aneuploidy (HRD high mean ploidy = 3.6, HRD low mean ploidy = 3.1, P = 0.047) than HRD-low cancers. These data show that deficiency in the HR pathway is correlated with genomic instability and an increase in predicted neoantigens.

To infer immunogenicity in BRCA1/2 mutation-associated breast cancers, we computed cytolytic index (13) as a measure of CD8⁺ T-cell cytotoxicity and immune ESTIMATE scores (14) as a measure of lymphocyte infiltration. Neither immune measure differed between breast cancers with mutations in BRCA1 and BRCA2 nor germline versus somatic alterations (Supplementary Fig. S2B and S2C); hence, all tumors were combined for further analysis. We then investigated the association of these immune metrics with HRD level. Cytolytic index was higher in BRCA1/2 mutation-associated (P = 0.0016) and HR gene mutation-associated (P = 0.0019) than HR wild type breast cancers (Fig. 1D), consistent with prior reports that HR pathway deficiency leads to increased immunogenicity (4–7). Unexpectedly, when BRCA1/2 mutant breast cancers were dichotomized by median HRD score, higher HRD scores were associated with lower cytolytic index (P = 0.043) and immune ESTIMATE score (P = 0.002), despite correlating with a higher predicted neoantigen load (Fig. 1E and F). The magnitude of both immune metrics was independent of tumor stage (Supplementary Fig. S2D). Importantly, HRD-high BRCA1/2 cancers more closely resembled HR wild-type cancers (cytolytic index: BRCA1/2 HRD high = 132.7, HR wild type = 118.5, P = 0.56; immune ESTIMATE score: BRCA1/2 HRD high = −2.7, HR wild type = 191.1, P = 0.12). We also interrogated immune metagenes (13) by GSVA and found lower enrichment of the type II IFN metagene (P = 0.002; Fig. 1G) among HRD-high BRCA1/2 mutation-associated breast cancers, suggesting lower immune effector activity in this tumor subset. HRD-low breast cancers were also associated with a significantly lower death hazard ratio from Cox proportional hazards analysis (P = 0.03; Supplementary Fig. S2E).

To gain further insight into factors that may contribute to lower immunogenicity in HRD-high BRCA1/2 mutated breast cancers, we compared all curated, canonical pathways (MSigDb) by GSVA (Supplementary Fig. S2F; Supplementary Table S5). Multiple pathways including Notch, tumor necrosis factor (TNF), and MAPK signaling linked to NF-κB, a key transcription factor involved in CD28-mediated T-cell activation and the expression of chemoattractive T-cell cytokines (15, 16), and were elevated across HRD-low relative to HRD-high tumors (Supplementary Table S5). Type I IFN signaling, known to induce NF-κB activation and elicit antitumor immune responses in lymphocytes (17), was upregulated in HRD-low tumors (P = 1.44E−14). Pathways related to MAP-kinase-induced NF-κB activation from toll-like receptors (P = 8.89E−04, P = 1.22E−03, P = 1.25E−03), the pattern recognition receptors of innate immunity that play an important role in facilitating antigen-driven immune responses (18), were elevated in HRD-low tumors, likely mediated by altered IKK signaling observed in this subset (P = 0.013; ref. 18; Supplementary Table S5). Notch1 and Notch4 signaling, known to enhance CD4⁺ and CD8⁺ effector activity (19), also was elevated in HRD-low tumors (P = 7.98E−3, P = 0.05, respectively), both of which are capable of activating NF-κB canonical signaling (19). Elevated TGFβ signaling, observed in HRD-low tumors (P = 0.045), can induce NF-κB via TAK1 (20). Further, the TNF pathway can activate NF-κB specifically through the actions of TNFR1 and TNFR2 (21), both higher in HRD-low tumors (P = 0.011, P = 0.016), and promote antitumor immune responses by amplifying T-cell receptor (TCR)-induced signals. NF-κB activity can regulate the transcription of defensins and β-defensins, small cationic peptides involved in tumor suppression as well as in T-cell chemotaxis (22), and defensin-related pathways were found at far higher levels in HRD-low tumors (P = 2.6E−36, P = 7.5E−34). As STING (TMEM173) protein has been implicated in driving immunogenicity in DNA damage response-deficient (DDRD) breast cancers (5, 23) and acts through NF-κB (24), we also compared the expression of STING by HRD level and found that HRD-low tumors expressed higher levels of STING than HRD-high tumors (P = 0.047; Supplementary Fig. S2G). These data collectively suggest that low HRD BRCA1/2 breast cancers are associated with increased immunogenicity likely driven by NF-κB signaling.

Based on our prior work (2), we investigated whether allele-specific loss of heterozygosity (LOH) of the germline BRCA1 or BRCA2 mutation was predictive of immunogenicity in TCGA data. Allele-specific LOH was determined, as described previously (2), by comparing purity-adjusted mutation allele frequencies and copy number states across matched tumor and normal samples (see “Materials and Methods”). Breast cancers with allele-specific LOH of the BRCA1/2 germline mutation had significantly higher HRD score than tumors without allele-specific LOH (2). In the TCGA data set, we examined breast cancers associated with germline mutations in BRCA2 only, including those with (n = 10, 59%) and without (n = 7, 41%) allele-specific LOH. We found that allele-specific LOH associated with germline mutations in BRCA2 was associated with lower type II IFN signaling, cytolytic index, and immune ESTIMATE scores (Fig. 2A; P = 0.01, P = 0.011, P = 0.039, respectively). We observed a similar trend in breast cancers associated with germline BRCA1 mutations, although only one breast cancer in the TCGA cohort lacked allele-specific LOH of the BRCA1 mutation (Supplementary Fig. S3A). Given that the difference in immunogenicity dependent on heterozygous versus homozygous germline BRCA1/2 mutations, we wanted to determine whether somatic BRCA1/2 mutation clonality also was associated with differences in immunogenicity. We categorized all breast cancers with somatic mutations in BRCA1/2 at a purity-adjusted allele frequency of less than 0.50 as subclonal (n = 14) to compare to cancers with either clonal mutations [corrected tumor allele frequency (CTAF) >0.50], promoter hypermethylation of BRCA1, and/or low expression (RNAseq z-score ←1.5 of BRCA1 or BRCA2 [referred to collectively as “clonal” BRCA1/2 somatic mutation-associated cancers] (n = 33); Supplementary Table S1). We found that cancers with clonal BRCA1/2 mutations had higher mutational burden (P = 0.05) and HRD scores (P = 2.37E−07; Fig. 2B), but lower cytolytic index (P = 0.0033) and immune ESTIMATE scores (P = 4.98E−05) than cancers with subclonal BRCA1/2 mutations (Fig. 2C). When evaluating BRCA1 cancers alone, we found that subclonal BRCA1 alterations were associated with borderline significantly higher cytolytic index (subclonal: n = 4, mean cyt ind = 285.18; clonal: n = 4, mean cyt ind = 109.65; P = 0.054) and significantly higher immune ESTIMATE scores (subclonal, mean score = 1410.28; clonal, mean score = 71.81; P = 0.022). BRCA2 cancers with subclonal alterations exhibited borderline significantly higher cytolytic index (subclonal: n = 10, mean cyt ind = 329.60; clonal: n = 12, mean cyt ind = 92.41; P = 0.067), and immune ESTIMATE scores (subclonal: mean score = 484.42; clonal: mean score = −227.48; P = 0.079). As STING can mediate inflammation through NF-κB and IRF7 (24), both of which were elevated in cancers with lower HRD, we then evaluated STING expression as a function of partial or complete loss of BRCA1/2 function. We found that tumors with either BRCA1/2 LOH or clonal somatic BRCA1/2 alterations had lower STING expression than tumors negative for BRCA1/2 LOH or those with subclonal BRCA1/2 somatic mutations (P = 0.04 and P = 2.1E−04; Supplementary Fig. S3B). Collectively, these data suggest that the disparity in immunogenicity as a function of HRD in breast cancers is driven by full loss of BRCA1 or BRCA2 function, either by allele-specific LOH of a germline mutation or clonality of a somatic mutation.

We performed IHC in breast cancers ascertained from Penn Medicine patients with confirmed germline BRCA1 (n = 23) or BRCA2 mutations (n = 12; Supplementary Fig. S3C; Supplementary Tables S1 and S2). We stained for multiple immune cell markers spanning adaptive and innate immune cell function including CD3, CD8, FOXP3, CD20, CD45, and CD68 in tissue microarray sections of BRCA1/2 breast cancers. T cells (CD3, CD8, FOXP3), B cells (CD20), and cells of the myeloid lineage (CD68) were all detectable in our breast cancer set. Intratumoral, stromal, and total CD8⁺ (P = 0.0018, 0.041, and 0.037, respectively) T cells were lower in BRCA1/2 breast cancers with elevated levels of HRD (Fig. 3A and B), consistent with our results from our analysis of TCGA dataset, whereas no differences were observed in other immune markers (Supplementary Fig. S3D). Differences in CD45⁺ T cells in association with HRD level were more strongly associated with BRCA1 mutation-associated cancers (intratumoral, stromal, and total; P = 0.032, 0.043, and 0.039, respectively; Supplementary Fig. S3E and S3F). We investigated the effects of allele-specific LOH associated with germline mutations in BRCA1 (LOH positive, n = 18, 78%; LOH negative, n = 5) and BRCA2 (LOH positive, n = 5, 42%; LOH negative, n = 7) cancers. No significant differences were observed across CD3⁺, FOXP3⁺, and CD68⁺ cells when analyzing germline BRCA1/2 cancers collectively by allele-specific LOH status (Supplementary Fig. S3G). However, we found those that were BRCA1/2 LOH positive had a lower number of CD8⁺ T cells (intratumoral P = 0.13, stromal P = 0.035, and total P = 0.015), CD20⁺ B cells (intratumoral P = 0.043, stromal P = 0.223, and total P = 0.189), and CD45⁺ T cells (intratumoral P = 0.002, stromal P = 0.041, and total P = 0.037; Fig. 3C and D). Again, these changes were more strongly associated with BRCA1 mutation-associated tumors; CD8⁺ (intratumoral P = 7.03e−5, stromal P = 0.0014, and total P = 0.00016), CD45⁺ (intratumoral P = 0.00043, stromal P = 0.00014, total P = 8.0e−4), and CD20⁺ (intratumoral P = 0.0345, stromal P = 0.0038, and total P = 0.0016) cells were all found at lower levels in BRCA1 allele-specific LOH-positive cancers (Supplementary Fig. S3H). Further, BRCA1 allele-specific LOH-positive cancers specifically exhibited lower levels of CD3⁺ (intratumoral P = 0.016, stromal P = 0.0092, total P = 0.0067) and CD68⁺ (intratumoral P = 0.019, stromal P = 0.106, total P = 0.140) cells (Supplementary Fig. S3H). Differences for all cell types in BRCA2 cancers were nonsignificant (Supplementary Fig. S3I). We stained for the programmed cell death protein ligand 1 (PDL1), a component of the dominant PD1–PDL1 immune checkpoint pathway induced by tumors to counter-regulate immune attack (25). We found lower levels of macrophage membrane (P = 0.048) and macrophage + tumor PDL1 (P = 0.012) in cancers with allele-specific LOH relative to cancers without allele-specific LOH (Fig. 3E and F), indicating lower tumor inflammation (25). These changes were driven by BRCA1 cancers, whereby allele-specific LOH-positive cancers had lower macrophage membrane (P = 0.019) and total (tumor + macrophage, P = 7.4e−4) PDL1 relative to LOH-negative cancers (Supplementary Fig. S3J). BRCA2 cancers trended in the same direction but failed to reach significance (Supplementary Fig. S3K).

Given that both TGFβ and NF-κB signaling were differentially regulated across HRD-high versus HRD-low tumors in TCGA cohort and given their known roles in remodeling the tumor microenvironment (Supplementary Table S5; refs. 26, 27), we investigated extracellular matrix composition by Masson's Trichrome staining in the Penn cohort (28). We found lower red myofibroblast staining in HRD-high versus HRD-low tumors (P = 0.0071; Fig. 3G and H).

We analyzed the association between hormone receptor expression and immunogenicity in the background of BRCA1/2 alterations, as TNBCs have been found to be generally more immunogenic than hormone receptor positive (Rec+) cancers (tumor types in Supplementary Table S1; ref. 6). Although there were no significant differences in cytolytic index and immune ESTIMATE within each hormone receptor subgroup based on BRCA1 or BRCA2 mutation status (Supplementary Fig. S4A), TNBCs had higher cytolytic index overall (P = 0.025; Fig. 4A) despite a similar mutational burden (Supplementary Fig. S4B). An expression level comparison of immunomodulatory genes across TNBCs and Rec+ BRCA1/2 mutant breast cancers found that TNBCs had higher expression of most immune markers than Rec+ tumors, suggesting a more inflamed microenvironment (Fig. 4B; ref. 25). Inflamed tumors often express counter-regulatory checkpoint proteins such as PDL1 to evade immune attack (25). In Penn BRCA1/2 germline mutation-associated breast cancers, membrane PDL1 was higher in TNBCs when summing with tumor and macrophage membrane PDL1 (P = 0.035; Fig. 4C).

We sought to investigate a potential interaction effect between HRD and hormone receptor statuses, performing a stratified comparison in BRCA1/2 TCGA tumors. Comparing cytolytic index and immune ESTIMATE in TCGA, we found the greatest difference between TNBC HRD-low (n = 10) and Rec+ HRD-high BRCA1/2 breast cancers (n = 22; P = 0.0013; Fig. 4D and E). Similarly, when partitioning by BRCA1/2 allele-specific LOH status and hormone receptor expression, we found that TNBC, LOH-negative cancers had the highest cytolytic index (n = 2, mean cytolytic index = 303.1) at a borderline significantly higher level than Rec+, LOH-positive cancers (n = 8, mean cytolytic index = 97.7, P = 0.052), and Rec+, LOH-negative and TNBC, LOH positive had middling levels of cytolytic index (mean cytolytic index = 143.6 and 156.4, respectively). We confirmed these observations in Penn BRCA1/2 germline mutation-associated breast cancers; LOH-positive cancers had significantly higher HRD score than LOH-negative cancers (Supplementary Fig. S4C; P = 6.65e−6). TNBC, LOH-negative (n = 4) cancers had the highest CD8 infiltrate when compared to TNBC, LOH-positive (n = 14; intratumoral P = 9.22e−5, stromal P = 0.0063, total P = 0.0023) and Rec+, LOH-positive (n = 8; intratumoral P = 0.00051, stromal P = 0.0016, total P = 0.0011) cancers, but nonsignificantly higher than Rec+, LOH-negative (n = 8; intratumoral P = 0.13, stromal P = 0.088, total P = 0.13) cancers (Supplementary Fig. S4D). The same observations were found when comparing TNBC, LOH-negative to LOH-positive CD3⁺ and CD45⁺ infiltrates (CD3: intratumoral P = 0.00024, stromal P = 0.038, total P = 0.0012; CD45: intratumoral P = 0.022, stromal P = 0.0022, total P = 0.0020), Rec+, LOH-positive (CD3: intratumoral P = 0.00073, stromal P = 0.014, total P = 0.0068; CD45: intratumoral P = 0.025, stromal P = 0.011, total P = 0.010), and Rec+, LOH-negative (CD3: intratumoral P = 0.068, stromal P = 0.069, total P = 0.066; CD45: intratumoral P = 0.021, stromal P = 0.041, total P = 0.044) cancers (Supplementary Fig. S4E and S4F). Further, the number of PRF1-positive cells in the tumor microenvironment was highest in TNBC, LOH-negative cancers relative to TNBC, LOH-positive (P = 0.024), Rec+, LOH-positive (P = 0.014), and Rec+, LOH-negative (P = 0.021) cancers (Supplementary Fig. S4G). TNBC, LOH-negative membrane also had the highest levels of membrane and total PDL1 when compared with TNBC, LOH-positive (macrophage membrane P = 0.015, tumor+macrophage membrane P = 0.039, total P = 0.034) and Rec+, LOH-positive cancers (macrophage membrane P = 0.012, tumor+macrophage membrane P = 0.028, total P = 0.013), but not Rec+, LOH-negative cancers (macrophage membrane P = 0.28, tumor+macrophage membrane P = 0.055, total P = 0.814; Supplementary Fig. S4H).

Comparing all differentially regulated canonical pathways between these two groups in the TCGA (Supplementary Fig. S4I; Supplementary Table S6), we found that pathways related to TCR, chemokine, and PD1 signaling were the most positively upregulated in TNBC HRD-low tumors (P < 0.005). Collectively, these data suggest that HRD and hormone receptor expression independently stratify immunogenicity and, when considered jointly, can more precisely categorize the immune properties of BRCA1/2 breast cancers.

We have delineated the molecular properties of immunogenicity in breast cancers associated with BRCA1/2 alterations (Fig. 4F); our observations have potential implications for treatment of these cancers with checkpoint blockade. BRCA1/2-mutation associated breast cancers exhibited slightly higher mutational load and predicted neo-epitopes compared with wild type tumors—features that were most prominent among the HRD-high subset; however, our findings now reveal that it is the HRD-low subset of breast cancers in this cohort that are the most highly immunogenic with the strongest evidence of CD8-driven T-cell responses. Thus, BRCA1/2-related breast cancers represent another example in which mutational burden and T-cell responses are not linked (29, 30); rather, tumor intrinsic features (31, 32) that regulate immune response and suppression are increasingly appreciated as driving forces. Further, our findings are consistent with a pan-cancer analysis which found that aneuploidy and a higher prevalence of large somatic copy number alterations, a feature of HRD, were associated with lower immune cytotoxicity, possibly due to these genomic events imparting dysfunction in cellular machinery involved in the recruitment of immune cells (11).

Immunogenicity in BRCA1/2 breast cancers as a function of HRD may be related to multiple pathways that contribute to dendritic and T-cell function (Notch, TNF, and TLR signaling) that converge on NF-κB (21), a key proinflammatory transcription factor known to influence the transcription of genes involved in survival, inflammation, and lymphocyte activation (15) and shown to promote immune surveillance by augmenting the antitumor responses of T cells (16). As NF-κB signaling is influenced by DNA damage and BRCA1/2 mutation status (33–36), our data collectively indicate that both tumor cell intrinsic (BRCA1/2 and HRD) and extrinsic signals (TNF and TLR signaling) may determine NF-κB activity and in turn immunophentoypes in BRCA1/2 mutation-associated breast cancers. HRD-low cancers also exhibited heightened TGFβ signaling, which is known to promote trans-differentiation of fibroblasts to myofibroblasts, a prominent feature of the HRD-low tumor microenvironment in our investigation (27). Myofibroblasts have previously been associated with sites of inflammation and their depletion has been associated with loss of T-cell effector cells, as we observed (37, 38). Further, an NF-κB positive feedback loop has been shown to maintain activated myofibroblasts in breast tumors through IL6 (39), a pathway which was elevated in HRD low tumors (Supplementary Table S5). Collectively, these data underscore that the inflammation and heightened immune infiltrates observed in HRD-low tumors may be driven by NF-κB, which may promote the presence of stromal myofibroblasts potentially via TGFβ and IL6 signaling.

Our study also highlights the relationship between partial or complete loss of BRCA1/2 function and immunogenicity in patients with germline BRCA1/2 mutations. Loss of the nonmutated wild-type allele in breast tumors associated with BRCA1 or BRCA2 germline mutations has been commonly reported (40). We have found that absence of BRCA locus-specific LOH is more frequent than previously thought, particularly in association with BRCA2 germline mutations, and that patients with absence of allele-specific LOH in their ovarian cancers have significantly poorer survival when treated with adjuvant platinum therapy and thus may drive intrinsic resistance to this therapy (2). In the current study, we found that breast cancers associated with BRCA1/2 germline mutations without allele-specific LOH were characterized by a higher cytotoxic CD8⁺ T-cell infiltrate, elevated markers of cytolytic activity, and increased macrophage membrane PDL1. Moreover, tumors with subclonal somatic mutations in BRCA1/2 also were more immunogenic than those with clonal somatic mutations. These associations were particularly pronounced in BRCA1-deficient cancers when analyzed separately, with generally similar trends observed in BRCA2-deficient cancers, suggesting that breast cancers without allele-specific LOH of the germline BRCA1/2 mutation or demonstrating subclonality of somatic BRCA1/2 mutations may be more intrinsically responsive to immune checkpoint blockade. Although our findings have important implications for the treatment of breast cancers with BRCA1 and BRCA2 alterations, whether patients with low HRD may preferentially respond to immune checkpoint blockade will require direct evaluation in the clinic.

Importantly, hormone receptor expression, previously found to independently influence immunogenicity (6), further stratified tumor immune properties. HRD-low TNBC tumors were the most immunogenic subset characterized by high PD1–PDL1 and TCR signaling, whereas HRD-high receptor-positive tumors were the least immunogenic.

Overall, our study addresses a pressing clinical need to understand the potential causes of heterogeneity in response to immune checkpoint blockade in breast cancer patients with high-penetrance germline BRCA1/2 mutations or with somatic alterations in BRCA1/2. Our data are consistent with prior reports that BRCA1/2 mutation-associated tumors are generally more immunogenic than BRCA1/2 wild type tumors (4–7), and we build on this work by accounting for the wide range in immunogenicity observed in BRCA1/2 tumors (4) through molecular stratification, which can offer key therapeutic insights. Our study further demonstrates that clinically available molecular genetic testing, namely assessment of tumor HRD, BRCA1/2 allele-specific LOH, and somatic BRCA1/2 clonality, coupled with hormone receptor status can be used to stratify breast cancer immunogenicity. As BRCA1 and BRCA2 mutations are associated with many other cancer types including ovarian and aggressive prostate cancer, our findings may have applicability beyond breast cancer. Although our study is one of the largest BRCA1/2 breast tumor analyses to date, a larger sample size would be invaluable to further understand differences in BRCA1 versus BRCA2 tumor immunity. Although our study did not specifically address cancers emerging from germline non-BRCA1/2 HR germline mutations, the evidence to date suggest that such tumors are quite diverse in that some are similar to BRCA1/2 mutation-associated tumors and display “BRCAness” (e.g., PALB2) whereas others are not BRCA-like (e.g. CHEK2, ATM; refs. 41, 42). Breast cancers associated with each gene would have to be profiled in equivalent numbers to our BRCA1/2 mutation-associated cancers, which is outside the scope of the current study.

In sum, the results of our study give novel insights into the relationship between HRD and immunogenicity in breast cancers associated with BRCA1/2 alterations, which can potentially guide treatment strategies utilizing DNA damaging agents and checkpoint blockade alone or in combination.

J.J.D. Morrissette reports receiving speakers bureau honoraria from TriCon and CHI NHS Summit, and is a consultant/advisory board member for Novartis and Loxo. S.M. Domchek reports receiving speakers bureau honoraria from AstraZeneca, Clovis, and Bristol-Myers Squibb. R.H. Vonderheide reports receiving speakers bureau honoraria from Celgene and Janssen, is a consultant/advisory board member for Apexigen, AstraZeneca, Genentech, Lilly, MedImmune, Merck, and Verastem, and reports receiving commercial research grants from Apexigen, Fibrogen, Inovio, Janssen, and Lilly. No potential conflicts of interest were disclosed by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conception and design: A.A. Kraya, K.N. Maxwell, M. Feldman, S.M. Domchek, R.H. Vonderheide, K.L. Nathanson

Development of methodology: K.N. Maxwell, A.J. Rech, A. Nayak, R.H. Vonderheide

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.A. Kraya, K.N. Maxwell, A. Barrett, J.J.D. Morrissette, M. Feldman, A. Nayak, S.M. Domchek, K.L. Nathanson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.A. Kraya, K.N. Maxwell, B. Wubbenhorst, B.M. Wenz, J. Pluta, A.J. Rech, N. Mitra, J.J.D. Morrissette, M. Feldman, A. Nayak, S.M. Domchek, R.H. Vonderheide, K.L. Nathanson

Writing, review, and/or revision of the manuscript: A.A. Kraya, K.N. Maxwell, N. Mitra, J.J.D. Morrissette, M. Feldman, A. Nayak, S.M. Domchek, R.H. Vonderheide, K.L. Nathanson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.M. Dorfman, N. Lunceford, A. Nayak, K.L. Nathanson

Study supervision: R.H. Vonderheide, K.L. Nathanson

The authors thank the patients that participated in their studies. The authors further thank E. John Wherry, PhD (University of Pennsylvania), for helpful comments on the manuscript. Financial support for this work was provided by the National Institutes of Health (P30 CA016520), Basser Center for BRCA at the University of Pennsylvania (to K.L. Nathanson, S.M. Domchek, and R.H. Vonderheide), Breast Cancer Research Foundation (to K.L. Nathanson, S.M. Domchek, and R.H. Vonderheide), Susan G. Komen Foundation (to S.M. Domchek), Rooney Family Foundation (to K.L. Nathanson and S.M. Domchek), National Institutes of Health (K12-CA076931; to K.N. Maxwell), Konner Family Foundation (to K.N. Maxwell), Parker Institute for Cancer Immunotherapy (to R.H. Vonderheide, A.J. Rech, and K.L. Nathanson), National Center for Advancing Translational Sciences of the National Institutes of Health (TL1TR001880; to A.A. Kraya), and National Human Genome Research Institute of the National Institutes of Health (5T32HG009495-02; to A.A. Kraya).

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.