PTEN Loss and BRCA1 Promoter Hypermethylation Negatively Predict for Immunogenicity in BRCA-Deficient Ovarian Cancer

Abstract

PURPOSE

Ovarian cancers can exhibit a prominent immune infiltrate, but clinical trials have not demonstrated substantive response rates to immune checkpoint blockade monotherapy. We aimed to understand genomic features associated with immunogenicity in BRCA1/2 mutation–associated cancers.

MATERIALS AND METHODS

Using the Cancer Genome Atlas whole-exome sequencing, methylation, and expression data, we analyzed 66 ovarian cancers with either germline or somatic loss of BRCA1/2 and whole-exome sequencing, immunohistochemistry, and CyTOF in 20 ovarian cancers with germline BRCA1/2 pathogenic variants from Penn.

RESULTS

We found two groups of BRCA1/2 ovarian cancers differing in their immunogenicity: (1) 37 tumors significantly enriched for PTEN loss (11, 30%) and BRCA1 promoter–hypermethylated (10, 27%; P = .0016) and (2) PTEN wild-type (28 of 29 tumors) cancers, with the latter group having longer overall survival (OS; P = .0186, median OS not reached v median OS = 66.1 months). BRCA1/2-mutant PTEN loss and BRCA1 promoter–hypermethylated cancers were characterized by the decreased composition of lymphocytes estimated by gene expression (P = .0030), cytolytic index (P = .034), and cytokine expression but higher homologous recombination deficiency scores (P = .00013). Large-scale state transitions were the primary discriminating feature (P = .001); neither mutational burden nor neoantigen burden could explain differences in immunogenicity. In Penn tumors, PTEN loss and high homologous recombination deficiency cancers exhibited fewer CD3+ (P = .05), CD8+ (P = .012), and FOXP3+ (P = .0087) T cells; decreased PRF1 expression (P = .041); and lower immune costimulatory and inhibitory molecule expression.

CONCLUSION

Our study suggests that within ovarian cancers with genetic loss of BRCA1/2 are two subsets exhibiting differential immunogenicity, with lower levels associated with PTEN loss and BRCA hypermethylation. These genomic features of BRCA1/2-associated ovarian cancers may inform considerations around how to optimally deploy immune checkpoint inhibitors in the clinic.

INTRODUCTION

BRCA1/2 are essential proteins involved in homologous recombination (HR)–based DNA repair.¹ Ovarian cancers with germline and somatic alterations in BRCA1/2 share many phenotypic characteristics including defects in double-strand break repair, replication fork stalling, and mutational signatures reflective of underlying HR deficiency (HRD).^2,3 Consequently, they exhibit similar therapeutic vulnerabilities, namely, sensitivity to inhibitors of poly(ADP-ribose) polymerase (PARP) and platinum-based chemotherapy.^4-12 However, resistance to DNA-damaging therapy, whether intrinsic in the form of the absence of BRCA1/2 allele–specific loss of heterozygosity¹³ or acquired via secondary genetic events,^9,14-18 has necessitated the investigation of orthogonal treatment strategies.

CONTEXT

• Key Objective

• To understand genomic features associated with immunogenicity in BRCA1/2 mutation–associated ovarian cancers.

• Knowledge Generated

• We found that BRCA1/2 mutation–associated ovarian groups clustered into two groups: Immune-High, associated with PTEN-loss and BRCA1 promoter–methylated tumors, and Immune-Low, with a significantly lower overall survival. BRCA1/2 mutation–associated ovarian cancers with PTEN loss had significantly higher homologous recombination deficiency scores, but exhibited significantly fewer CD3+, CD8+, and FOXP3+ T cells.

• Relevance

• Guided by molecular features, BRCA1/2 mutation–associated ovarian cancers can be divided into two groups with differing levels of immunogenicity, which may inform the use of immune checkpoint inhibitors in this patient group.

One treatment strategy that has been hypothesized to impart clinical benefits in BRCA1/2-deficient ovarian cancer is immune checkpoint blockade (ICB).^19-25 BRCA1/2-deficient ovarian cancers tend to have a relatively increased neoantigen load, because of a reliance on error-prone double-strand break repair,²² which can be predictive in some tumor types of ICB response.^20,22,23 BRCA1/2-deficient ovarian tumors are also characterized by a high presence of tumor-infiltrating lymphocytes (TILs),^20,22,26,27 which is a positive prognostic factor for survival.^20,22,24 PARP1 inhibition (PARPi), to which BRCA1/2-deficient ovarian cancers can respond,^5,16,28 has also been shown to increase TILs and synergistically combine with inhibitors of the immune checkpoint protein CTLA4 in mouse models.^29,30

Despite the immunogenic properties of BRCA1/2-deficient ovarian cancers, the results of clinical trials evaluating ICB have been variable.³¹ A single-agent phase Ib clinical trial of the programmed death-1 (PD1) inhibitor avelumab (n = 125) with epithelial ovarian cancer found an objective response rate (ORR) of 9.6% and a 1-year progression-free survival (PFS) of 10.2%.³² Patients with BRCA1/2 mutations did not selectively benefit over BRCA1/2 wild-type (WT) patients in this trial.^32,33 A phase II ovarian cancer study of pembrolizumab (KEYNOTE-100, n = 367) also showed modest activity with an ORR of 8.0%; BRCA mutation status was not evaluated as a biomarker.³⁴ In a single-institution study, BRCA mutation status, tumor mutational burden (TMB), and HRD were not associated with response to ICB in patients with ovarian cancer although a high fraction of genome altered was associated with improvement in PFS and overall survival.³⁵ A trial combining PARP1 inhibitor niraparib with the PD1 inhibitor pembrolizumab (n = 62) showed an ORR of 18% (5% complete response and 13% partial response³⁶), which did not differ by tumor BRCA deficiency (11 carriers, 18% response). The accompanying correlative study found that having a marker of HRD, mutational signature 3 (n = 11, 33%), a positive immune score (n = 3, 9%), or both (n = 6, 18%) was predictive of response.³⁷ Thus, it remains unclear whether responses were driven primarily by PARPi or checkpoint blockade or the treatment of two subgroups in the population.^36,38,39 Although larger phase II and III trials evaluating anti–CTLA-4 and anti-PD1/programmed cell death ligand-1 in combination with targeted therapy are underway,^21,40 these results suggest that a subset of BRCA1/2 ovarian cancers may harbor immunosuppressive mechanisms that impede clinical benefits.

Herein, we investigated 86 BRCA1/2-deficient ovarian cancers using genomic data from the Cancer Genome Atlas (TCGA) and performed genomic and histopathologic analyses in ovarian cancers from our institute to determine the genetic and genomic properties associated with intrinsic immunogenicity. Our findings illustrate that transcriptional pathways, loss of PTEN, and genomic signatures of HRD collectively inform intrinsic immunogenicity in BRCA1/2-deficient ovarian cancers and may aid in the evaluation of ICB in clinical trials.

MATERIALS AND METHODS

Materials and methods describing the acquisition and sequencing of Penn tumors, the acquisition of TCGA data, immunohistochemistry, CyTOF analysis, and bioinformatics are described in the Data Supplement.

RESULTS

Genomic and Transcriptomic Properties of BRCA1/2-Deficient Ovarian Cancers

We investigated immunogenicity in ovarian cancers from TCGA, comparing BRCA1/2-deficient with HR-proficient cancers (HR-WT cancers, Data Supplement). Ovarian cancers with mutations in non-BRCA1/2 genes involved in HR⁶ were excluded from our analysis because of the insufficient sample size for immunogenetic analysis (n = 10 RNAseq). TMB (Methods) was significantly different across BRCA1 versus HR-WT (P = .00032, Kruskal-Wallis) and across BRCA2 versus HR-WT cancers (P = .00043, Kruskal-Wallis, Data Supplement). The number of single-nucleotide nonsynonymous variants and neoantigen load were similar across BRCA1 versus BRCA2 cancers, and BRCA1/2-deficient ovarian cancers had a significantly higher mean number of single-nucleotide nonsynonymous variants (Data Supplement; P = .0036 and P = .0025) and neoantigen loads (Data Supplement; P = .018 and P = .0066, respectively) than HR-WT cancers. BRCA1/2-deficient cancers collectively exhibited a higher HRD score, a metric of genomic instability associated with BRCA1/2 dysfunction,^8,13 than HR-WT cancers (Data Supplement; P = 6.74e–05). Notably, neither HRD nor TMB were significantly correlated with sequencing coverage (Data Supplement).

We investigated biologic pathways that may differentiate BRCA1/2-deficient ovarian cancers (Methods). Clustering of gene set variation analysis scores across all cancers found that three recurring biologic functions inclusive of immune, hematologic, and tumor-suppressive (p53 and ATM) pathways were significantly upregulated in cluster c2 (Fig 1A and Data Supplement). We evaluated the relationship between PTEN status and cluster assignment, as PTEN loss is a frequent driving event in BRCA1/2 high grade serous ovarian cancer⁴¹ and has been shown to impede immune cell infiltration in other solid tumors.^42,43 We confirmed in our cohort that BRCA1/2 mutations were associated with a higher rate of PTEN loss; 17.3% of BRCA1/2 cancers had homozygous PTEN loss versus 7.7% of HR WT tumors (P = .019, Fisher's exact test). PIK3CA amplification was found in 20% of all samples with higher rates in the BRCA1 hypermethylation group at 30% than those with BRCA1/2 somatic and germline mutations (14%) and non-BRCA1/2 (17%; Data Supplement). Using the chi-square test of independence, we found that BRCA1/2-deficient PTEN-WT cancers were significantly over-represented in one cluster (28 of 29 tumors), that is, c2, whereas BRCA1 promoter–hypermethylated and BRCA1/2-deficient cancers with homozygous loss of PTEN (Methods) were significantly over-represented in cluster c1 (14 of 37 tumors, Fig 1B, P = .0016), but not PIK3CA amplification. When comparing overall survival across cluster c1 versus cluster c2 by multivariate Cox analysis, we found significantly worse prognosis in cluster c1 relative to cluster c2 when adjusting for grade, HRD level, ploidy, BRCA mutation, and stage (Fig 1C, P = .0186, hazard ratio = 4.19 [CI, 1.27 to 13.83]). When comparing PFS across clusters, we found a borderline significantly worse prognosis in cluster c1 relative to cluster c2 when adjusting for the same set of covariates (Fig 1C, P = .062, hazard ratio = 3.26 [CI, 0.94 to 11.33]).

FIG 1.

Mutational and transcriptomic features of the Cancer Genome Atlas ovarian cancers with somatic or germline BRCA1/2 alterations: (A) GSVA illustrating significantly different canonical pathways (MSigDb, adjusted P value < .05) across c1 versus c2 GSVA signaling clusters in BRCA1/2 ovarian cancers; (B) chi-square test of independence illustrating frequency of mutational subtypes within hierarchical clusters in BRCA1/2 cancers (P = .0016); and (C) covariate-adjusted (grade, homologous recombination deficiency level, ploidy, stage, and BRCA1/2 mutation) OS (hazard ratio = 4.19 [CI, 1.27 to 13.83]) and PFS (hazard ratio = 3.26 [CI, 0.94 to 11.33]) curves across c1 versus c2 GSVA signaling clusters in BRCA1/2 ovarian cancers. *P < .05, **P < .01, Student's t-test, Cox Proportional Hazards. BRCA1-HM, BRCA1-hypermethylated; BRCA1/2-P, BRCA1/2 PTEN wild-type; BRCA1/2-PL, BRCA1/2 PTEN loss; FS, Frameshift indel; GSVA, gene set variation analysis; OS, overall survival; PFS, progression-free survival; SNV, single-nucleotide variant; WT, wild-type.

Immunogenicity of BRCA1/2-Deficient Ovarian Cancers by PTEN Status and Mechanism of BRCA1/2 Loss of Function

Given the transcriptional clusters observed in Figure 1A, we categorized BRCA1/2-deficient cancers on the basis of their PTEN status and mechanism of BRCA1/2 loss.⁴⁴ We found no significant differences in TMB or neoantigen load when comparing on the basis of PTEN loss status within either germline or somatic BRCA1/2-deficient ovarian cancers or HR-WT cancers (Data Supplement). However, HRD scores were significantly higher in BRCA1/2-deficient PTEN-loss and BRCA1 promoter–hypermethylated cancers as compared with BRCA1/2-deficient PTEN-WT and HR-WT cancers (Fig 2A; P = .024 and P = .024, respectively). Among the three components of the HRD score (Methods), large scale state transitions (LST) were significantly lower in BRCA1/2-deficient PTEN-WT ovarian cancers when compared with BRCA1/2-deficient PTEN-loss (P = .012) and BRCA1 promoter–hypermethylated (P = .010 for LST) cancers (Fig 2B). The LST scores of BRCA1/2-deficient PTEN-WT cancers were similar to those of HR-WT cancers. Notably, non-telomeric allelic imbalance (NtAI) scores were significantly higher in BRCA1/2-deficient PTEN WT and HR-WT cancers versus BRCA1/2-deficient PTEN-loss cancers (Fig 2B).

FIG 2.

BRCA1/2 ovarian cancer immunogenicity in relation to the mechanism of BRCA1/2 loss of function and PTEN status: (A) HRD score in Penn + TCGA BRCA1/2 and TCGA HR WT (N = 108) ovarian cancers by PTEN status (PTEN loss, n = 17; PTEN WT, n = 69) versus TCGA BRCA1-hm ovarian cancers (n = 11); (B) LST, NtAI, and LOH scores in Penn + TCGA BRCA1/2 and TCGA HR WT (N = 108) ovarian cancers by PTEN status (PTEN loss, n = 17; PTEN WT, n = 69) versus TCGA BRCA1-hm ovarian cancers (n = 11); (C) in TCGA, CD8 T-cell CIind in BRCA1/2 ovarian cancers by PTEN status (PTEN loss, n = 12; PTEN WT, n = 43) versus TCGA BRCA1-hm ovarian cancers (n = 11) and TCGA HR WT (N = 108); (D) immune infiltration inferred by ESTIMATE in BRCA1/2 ovarian cancers by PTEN status (PTEN loss, n = 12; PTEN WT, n = 43) versus TCGA BRCA1-hm ovarian cancers (n = 11) and TCGA HR WT (N = 108); (E) expression of immune-regulatory molecules in BRCA1/2 ovarian cancers by PTEN status (PTEN loss, n = 12; PTEN WT, n = 43) versus TCGA BRCA1-HM ovarian cancers (n = 11); and (F) coefficient weights from SVM analysis of copy number variations illustrating gene importance across TCGA BRCA1/2-P (n = 43) versus TCGA BRCA1/2-PL (n = 12) and BRCA1-HM (n = 11) ovarian cancers. *P < .05, **P < .01, Student's t-test, analysis of variance. BRCA1/2-P, BRCA1/2 PTEN wild-type; BRCA1/2-PL, BRCA1/2 PTEN loss; CIind, cytolytic index; HR, homologous recombination; HRD, homologous recombination deficiency; iEst, immune ESTIMATE; LOH, loss of heterozygosity; LST, large scale state transitions; NtAI, non-telomeric allelic imbalance; SVM, support vector machine; TCGA, the Cancer Genome Atlas; WT, wild-type.

We determined tumor immunogenicity by comparing expression-based immune indices, including cytolytic index (CIind)⁴⁵ and immune ESTIMATE (iEst) score.⁴⁶ Neither index differed significantly when comparing across cancers with BRCA1 versus BRCA2 alterations or germline versus somatic BRCA1/2 alterations, or PTEN-loss versus PTEN-WT among HR-WT cancers (Data Supplement). Both indices correlated negatively with tumor purity estimated by copy number variation (CNV; see the Methods; Data Supplement; CIind: R = –0.25, P = .038; iEst: R = –0.38, P = .0013). The LST score correlated negatively with iEst scores and not with CIind (Data Supplement; CIind: R = –0.079, P = .52; iEst: R = –0.29, P = .019), whereas loss of heterozygosity score and NtAi scores did not exhibit any significant associations (data not shown). The HRD score showed a borderline significantly negative association with the iEst score, but not CIind (Data Supplement; CIind: R = –0.077, P = .53; iEst: R = –0.23, P = .067). When stratifying further, we found that BRCA1/2-deficient PTEN-loss and BRCA1 promoter–hypermethylated cancers had lower CIind (P = .034 and P = .042, respectively) and, for the latter group, decreased iEst versus BRCA1/2-deficient PTEN-WT cancers (P = .003; Figs 2C and 2D). Tumor purity was also lower in BRCA1/2-deficient PTEN-WT versus BRCA1/2-deficient PTEN-loss tumors (Data Supplement, P = 9.34e–05). We analyzed the expression of multiple immune-regulatory genes that we curated from the literature^{26,29,45,47,48} as having important roles with respect to response to ICB. We found significantly higher expression of ADORA2A (P = .033), DOK3 (P = .0026), HAVCR2 (P = .00055), CD28 (P = .0037), CD86 (P = .00091), ICOS (P = .05), and TNFRSF17 (P = .033) in BRCA1/2-deficient PTEN-WT tumors versus BRCA1 promoter–hypermethylated and BRCA1/2 PTEN-loss tumors (Fig 2E), consistent with the hypothesis that elevated expression of immune inhibitors (ADORA2A, DOK3, and HAVCR2) serves to counter-regulate heightened immune activity.^25,26 T-cell chemoattractants (CCL5, CXCL9, CXCL10, and CXCL11) coordinately trended toward higher expression in BRCA1/2-deficient PTEN-WT tumors, but not in BRCA1 promoter–hypermethylated or BRCA1/2-deficient PTEN-loss tumors. HLA expression did not have any discernable pattern across the three groups.

We aimed to identify CNVs affecting gene regions involved in immune system function (Reactome, n = 946 immune-related genes) that may distinguish immunologically cold (BRCA1/2-deficient PTEN-loss + BRCA1-hypermethylated) cancers versus hot cancers (BRCA1/2-deficient, PTEN-WT), as focal CNVs affecting genes involved in immune function influence tumor immunophenotypes.⁴⁵ By an L1-regularized support vector machine,⁴⁹ we identified 42 genes distinguishing the two groups with an area under the receiver operating characteristic curve of 0.96 (Data Supplement, hg38 genomic locations). We observed deletions in EIF2AK2, NCF2, and KPNA3, among BRCA1/2-deficient PTEN-WT ovarian cancers. Gains in DEFB4A and NLRP1 were more frequently observed in BRCA1/2-deficient PTEN-loss or BRCA1 promoter–hypermethylated ovarian cancers.

Association of PTEN Loss and HRD Score With Immune Infiltration and Cytotoxicity in BRCA1/2-Deficient Penn Ovarian Cancers

To validate the immunologic effects of PTEN loss, we evaluated Penn ovarian cancers associated with germline BRCA1/2 mutations (n = 18, Data Supplement). Multiple markers corresponding to adaptive and innate immune cell function were compared by PTEN status (Methods). BRCA1/2-deficient PTEN-loss cancers (n = 5) were characterized by significantly lower intratumoral CD3+ cells (P = .05), intratumoral CD8+ (P = .012), and intratumoral and stromal FOXP3+ (P = .0087 and P = .037, respectively) immune cells (Figs 3A and 3B), revealing that PTEN loss is associated with T-cell exclusion in BRCA1/2-deficient ovarian cancer. Other immune cell types (CD4, CD20, and CD68) did not demonstrate a statistically significant difference by PTEN status (Data Supplement). Furthermore, BRCA1/2-deficient PTEN-loss cancers were characterized by lower numbers of PRF1 (perforin 1)-positive cells (P = .041, Figs 3C and 3D), suggesting lower antitumor cytolytic activity in this subset. To determine whether the difference in cytolytic activity was due to fewer CD8+ cells or due to lower immune activity on a per-cell basis, we calculated the ratio of PRF1+ to CD8+ cells and found no significant difference (P = .71) across PTEN-WT versus PTEN-loss BRCA1/2-deficient cancers. MATH score, a measure of intratumoral heterogeneity,⁵⁰ was significantly higher in BRCA1/2-deficient PTEN-loss cancers than BRCA1/2-deficient PTEN-WT cancers (Fig 3E, P = .0042), in agreement with previous work demonstrating that CD8+ T-cell infiltration is negatively associated with ovarian cancers exhibiting greater clonal diversity.⁵¹

FIG 3.

Evaluation of tumor infiltrates, antitumor immune activity, and somatic mutation clonality in Penn BRCA1/2 ovarian cancers by PTEN status. Immune cell populations and immune effector molecules as a function of PTEN status in Penn ovarian cancers associated with germline BRCA1/2 mutations. (A) Levels of intratumoral and stromal CD3+, CD8+, and FoxP3+ T cells in BRCA1/2-PL (n = 5) or BRCA1/2-P (n = 13) ovarian cancers. (B) Representative 10× immunohistochemical images of CD3+, CD8+, and FoxP3+ T cells in BRCA1/2 ovarian cancers (the scale bar represents 0.1 mm at 0.01 mm increments). (C) Number of PRF1-positive cells in BRCA1/2-PL (n = 5) or BRCA1/2-P (n = 13) ovarian cancers. (D) Representative images of intratumoral PRF1 expression in BRCA1/2-PL or BRCA1/2-P ovarian cancers (the scale bar represents 0.1 mm at 0.01 mm increments). (E) MATH scores by PTEN status. Error bars, standard error, *P < .05, **P < .01, Student's t-test. BRCA1/2-P, BRCA1/2 PTEN wild-type; BRCA1/2-PL, BRCA1/2 PTEN loss; MATH, Mutant Allele Tumor Heterogeneity.

CyTOF analysis was performed on 11 available Penn tumor digests (n = 8) or ascites samples (n = 3) with BRCA1/2 mutations (2 of 11 with homozygous PTEN loss) to measure the expression of immune-regulatory molecules in CD3+ T-cell subsets in the tumor microenvironment (Methods, Data Supplement). viSNE plots mapped the location of immune subsets such as CD8+ and CD8– T cells (Data Supplement). viSNE analysis^52,53 identified dense clusters and higher frequencies of T cells in BRCA1/2-deficient PTEN-WT tumors expressing relatively high amounts of inhibitory immune checkpoints CTLA4, LAG3, and TIGIT, as well as FOXP3+ cells in a representative BRCA1 ovarian cancer, in contrast to comparatively lower overall expression in a BRCA1 PTEN-loss cancer (Fig 4A). The proinflammatory and costimulatory molecules interleukin (IL)-2, interferon gamma, IL-6, and CD28 were also expressed by more T cells at high levels in the PTEN-WT cancer relative to the PTEN-loss cancer (Fig 4B), underscoring the association of PTEN loss with reduced T-cell activation in the tumor microenvironment.

FIG 4.

CyTOF analysis of immunoinhibitory and cytolytic activity in Penn BRCA1/2 ovarian cancers by PTEN status and HRD score: (A) viSNE maps of CD4+ and CD8+ T cells illustrating differences in expression of inhibitory immune checkpoint molecules CTLA4, LAG3, TIGIT, and FOXP3 from a patient with a germline BRCA1/2-PL tumor versus a patient with a BRCA1/2-P; (B) viSNE maps of CD4+ and CD8+ T cells illustrating differences in expression of proinflammatory and costimulatory immune molecules IL2, IFNG, IL6, and CD28 from a patient with a germline BRCA1/2-PL versus a patient with a BRCA1/2-P; and (C) Scatterplots illustrating negative correlations between the HRD score and the expression of inhibitory immune molecules and immune cell subsets (top) and between the HRD score and proinflammatory and immune-activating molecules (bottom) by CyTOF in BRCA1/2 ovarian cancers (n = 11). BRCA1/2-P, BRCA1/2 PTEN wild-type; BRCA1/2-PL, BRCA1/2 PTEN loss; HRD, homologous recombination deficiency; IFNG, interferon gamma; IL, interleukin; r², Pearson Correlation Coefficient.

We investigated the association of the HRD score with immune cell subsets and immunomodulatory molecules in BRCA1/2-deficient ovarian cancers (Fig 4C). We found negative correlations between the HRD score and the frequency of tumor-associated T cells expressing CTLA4, LAG3, and CD160 immune inhibitory molecules and FOXP3+ T cells, which have been shown to accumulate at sites of cytotoxic T cells to mitigate antitumor immune attack.^54,55 Furthermore, proinflammatory cytokines (interferon gamma, IL-2), cytolytic pore-forming molecules (PRF and GZMB), and markers of proliferating or activated T cells (pSTAT5, Ki67, CD25, and GITR) were all negatively correlated with the HRD score (Fig 4C). Despite the negative correlations, the P values are nonsignificant likely because of the sample size.

Notably, the BRCA1/2-deficient PTEN-loss tumors had a higher preponderance of CD103+CD69+CD127– resident memory T cells (Trm) expressing PD-1 relative to the PTEN-WT tumor (Data Supplement). Furthermore, the HRD score is positively associated with CD103+CD69+CD127– Trm cells expressing PD-1 (Data Supplement).^56,57 Taken together, these results further illustrate that PTEN loss or high HRD may inform immunologic states in BRCA1/2-deficient ovarian cancers.

DISCUSSION

Our work sheds light on the tumor-immune heterogeneity in ovarian cancers with BRCA1/2 alterations and potentially gives novel insights into the treatment of these cancers with ICB. Similar to other studies of ovarian cancer, we found PIK3CA amplification in 20% of ovarian cancers but did not identify an association with immunogenicity.⁵⁸ Our results are similar to immunologic studies in other cancers^59-64 with PTEN loss and are also consistent with a study of 5,400 ovarian cancers, with 3,244 being high-grade serous, which demonstrated a correlation between cytoplasmic PTEN staining and low CD8+ T cells.⁶³ Notably, in BRCA1/2 breast cancers, we also found that markers of immunogenicity were inversely correlated with the HRD score in both TGCA and local tumor analyses, although likely through distinct underlying genetic mechanisms.⁶⁵ Furthermore, Davoli et al and Thorrson et al found negative correlations between CNV burden and measures of immunogenicity,^51,64-69 consistent with our analysis.

Our findings are also consistent with previous observations in melanoma in which T-cell exclusion is a mechanism of mitigating immune attack in PTEN-loss cancers.^69,70 Our data suggest that immune exclusion in PTEN-loss cancers may result from reduced expression of chemokines that play important roles in T-cell recruitment.⁷⁰ This finding is consistent with our recent finding that a reduction of both tumor intrinsic CCL5 expression and CCL5-driven CXCL9 expression by macrophages promotes TIL desertification and immune blunting, whereas CCL5 and CXCL9 overexpression in the ovarian cancer microenvironment is associated with CD8+ T-cell infiltration.⁴⁸

Our observations with HRD and immunogenicity highlight that although HRD contributes to an increase in the number of neoantigens, chronic exposure of TILs to their cognate antigens may contribute to a quiescent phenotype in ovarian cancer.^71-73 CyTOF analysis showed a higher preponderance of PD1-expressing CD103+CD69+CD127– Trm cells in PTEN-loss or HRD-high cancers. Trm cells have been shown in ovarian cancer to be quiescent,⁷¹ and anti-PD1 therapy can rapidly induce tumor cytotoxicity.^57,74 Notably, cancers with higher HRD are more intrinsically responsive to DNA-damaging agents,^8,13 which can work with ICB to synergistically eliminate solid tumors in mouse models.^29,30 Our data suggest that BRCA1/2-deficient PTEN-loss and BRCA1 promoter–hypermethylated cancers may respond to PARPi, whereas ICB may be optimal for patients BRCA1/2-deficient PTEN-WT cancer. Thus, combination therapy may be efficacious, as PARPi and ICB each treat a subgroup of BRCA1/2-deficient ovarian cancers.⁷⁵

From pathway analysis, we found that BRCA1/2-deficient PTEN-WT ovarian cancers had higher expression of ATM-driven DNA repair pathways, consistent with previous studies demonstrating a prevalent role of PTEN in HR-based genomic repair.^76,77 Along with lower prevalence of LSTs in this group, these results indicate that partial retention of DNA repair activity may be less immunosuppressive than full loss. These findings also indicate that BRCA1/2-deficient ovarian cancers may represent another example in which aneuploidy and large CNVs^68,78 are important driving forces that regulate immune responses and may supersede TMB as the primary immune influence.⁴⁷

Although nearly all BRCA1/2-deficient ovarian cancers in our study harbored a TP53 mutation, p53 signaling in BRCA1/2-deficient PTEN-WT cancers was elevated relative to the remaining cancers by gene set variation analysis. TP53 may be a key driver of immune responses in this subset of ovarian cancers, as a previous study found that pharmacologic activation of p53 in the tumor microenvironment enhances CD8-driven immunogenic cell death in mouse models.⁷⁹ Furthermore, our analyses of somatic CNVs identified a greater prevalence of deletions of EIF2AK2, NCF2, and KPNA3, all of which are either targets of or influence p53 signaling, in the BRCA1/2 PTEN-loss and BRCA1 promoter–hypermethylated group.^80-82 In particular, EIF2AK2 is a target of TP53 activity, serving a proapoptotic role for tumor suppression.⁸³ The more prevalent deletion of EIF2AK2 in PTEN-WT ovarian cancers in conjunction with heightened p53 signaling suggests an intricate proinflammatory mechanism of p53 in ovarian cancer with concomitant selection against the tumor-suppressive role of EIF2AK2.

Although the study benefited from inclusion of both TCGA and locally generated data, each sample set had limitations. TGCA data did not include linked immunohistochemistry data examining immune cells, and the Penn data derived from formalin-fixed paraffin-embedded tissues were limited in terms of expression analysis. Neither sample set had whole genome sequencing data, which most accurately characterize TMB and large-scale CNVs. Several findings warrant independent validation, in particular, the CyTOF analysis and association of the expression signatures with outcomes. In addition, a study of BRCA1/2 breast cancer using publicly available TCGA and Wellcome Trust Institute data suggests that PTEN status modulates immunogenicity but, by contrast, found that PTEN loss was associated with a more T-cell–inflamed signature.⁸⁴ However, PTEN loss also correlates with BRCA1 mutation status and triple-negative breast cancer status, and in their analysis, the latter may account for the enrichment of T-cell–inflamed signatures, which we and many others have found to be relatively more immunogenic than hormone-positive breast cancers.⁶⁵

Taken together, our study gives novel insights into the genetic events that may contribute to immunosuppression in BRCA1/2-deficient ovarian cancers, defining a subset of immunologically cold tumors. This understanding may help craft more efficacious use of ICB in the clinic.

DATA SHARING STATEMENT

The WES data that support 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 are available from the authors upon request.

AUTHOR CONTRIBUTIONS

Conception and design: Adam A. Kraya, Kara N. Maxwell, Michael Feldman, Susan M. Domchek, Robert H. Vonderheide, Katherine L. Nathanson

Financial support: Katherine L. Nathanson

Administrative support: Katherine L. Nathanson

Provision of study materials or patients: Katherine L. Nathanson

Collection and assembly of data: Adam A. Kraya, Monika A. Eiva, Michael Feldman, Anupma Nayak, Daniel J. Powell, Susan M. Domchek, Katherine L. Nathanson

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).