Metastatic Prostate Cancers with BRCA2 versus ATM Mutations Exhibit Divergent Molecular Features and Clinical Outcomes

Clinical grade genomic tests are available and widely used for patients with metastatic prostate cancer (mPC). Consortium-sequencing studies have reached a consensus that recurrent genomic alterations of homologous recombination repair (HRR) genes are a major feature in approximately 20% of mPCs (1–7). For such reasons, mutations in HRR genes are reported in genomic tests to risk-stratify patients and inform clinical decisions for patients with mPC (8–10). Although this is an exciting advancement, the field has begun to recognize that HRR genes, such as ATM or BRCA2, may govern distinct clinical responses in patients with mPC (10–13) and exhibit differential functions in preclinical models (14). To this end, elucidating their underlying mechanisms is a major objective in the field. This understanding is projected to impact treatment plans and inform avenues to improve precision therapies for patients with mPC.

HRR genes regulate critical DNA repair processes essential for tumor cell survival. For these reasons, cancer cells deficient in HRR become dependent on alternative repair mechanisms to survive. The reliance on other repair pathways is the foundation for using small-molecule PARPi for mPCs with HRR gene mutations (9, 10, 15). In 2020, after successful registrational clinical trials, the FDA approved the use of two PARPi, olaparib and rucaparib, for patients with mPC. Although this provides treatment strategies for patients who otherwise lack options, subsequent studies have determined that PARPi sensitivity is heterogeneous and specific to the underlying HRR gene mutations. Multiple studies noted that the clinical benefits were robustly driven by mutations in BRCA2, whereas limited responses were observed in ATM-mutated or even BRCA1-mutated mPC (11, 12, 16, 17). Separately, studies have also demonstrated that patients with the ATM mutation, as compared with BRCA2, have worse response to platinum therapies, but improved response to taxanes (18, 19). Particularly, in the TRITON2 study, of patients with DDR genes that received rucaparib, the PSA response in patients with the BRCA1/2 mutation was 53.6% (37/69), whereas in patients with the ATM mutation, it was 2.4% (1/41). Outside of PARPi, patients with the BRCA2 mutation have worse time-to-the development of castration resistance and have worse outcomes as compared with patients with the ATM mutation (18–21). Furthermore, our prior study in preclinical models also supported the clinical findings, in which gene depletion methods that induced the loss of BRCA2 or ATM in prostate cancer cell lines yielded distinct responses toward PARPi (14). Overall, our study demonstrated that reductions in BRCA2 expression did in fact increase PARPi response, whereas loss of ATM mediated by CRISPR did not. ATM deletions instead increased sensitivity to inhibition of ATR, a protein that senses DNA damage. Together, the clinical and laboratory findings provide evidence that BRCA2 and ATM mutations, whereas both being HRR genes, are distinct subtypes of mPCs. To this end, in both the clinical and preclinical settings, thorough interrogations are required to better define the mPC subtypes driven by ATM and BRCA2 mutations.

In this study, we aimed to address the underlying molecular differences of ATM and BRCA2-mutated mPC. Through the Caris Precision Oncology Alliance (POA) consortium, we aggregated a cohort of 1,187 microsatellite-stable mPCs collected from metastatic tissue sites, including bone, lymph nodes, and liver. Within these patients, we identified patients with mPC with mutations of ATM, BRCA2, or other homologous repair-deficient (HRDother) genes. To resolve the molecular features, we have also acquired whole-exome sequencing (WES) and whole-transcriptome sequencing (WTS) data uniformly processed through the Caris Life Sciences sequencing platform. The large number of patients permitted robust statistical comparisons of clinical and genomic features between subgroups of patients with mPC with different HRR mutations.

We queried the Caris Life Sciences database to assess the molecular alterations, gene expression, and related survival outcomes of ATM (n = 88) or BRCA2 (n = 98)-mutated microsatellite stable metastatic prostate cancer. No blood-based ctDNA biopsies were included in this study. The Caris Life Sciences database includes factors that are available in insurance claims. This includes gender (all male for this study), age at diagnosis, the therapeutic agents given and the dates of administration, and time of last contact. With regard to the tumor biopsy sample, the tissue site from which the sample was obtained, and the primary cancer site of the patient, all information pertains to prostate for this study. HRR-proficient (HRP, n = 808) tumors were defined as any tumor classified as not having a pathogenic or likely pathogenic mutation in any of the following HRR genes: ATM, BRCA2, RAD51B, RAD51C, RAD51D, RAD54L, FANCL, PALB2, CHEK1, CHEK2, CDK12, BRCA1, BARD1, and BRIP1. Homologous recombination–deficient (HRDother, n = 193) tumors harbored one or more “pathogenic,” or “likely pathogenic” HRR gene mutations, excluding ATM and BRCA2. Comprehensive molecular profiling, including WES, targeted next-generation sequencing (NGS; ref. 22), and WTS, was performed in a CLIA/CAP/ISO15189–certified clinical laboratory (Caris Life Sciences). The aggregate genomic and molecular features of each experimental group are summarized in Supplementary Table S1.

An automated staining platform (Ventana Medical Systems, Inc.) was used to detect PD-L1 protein expression (SP124 antibody). PD-L1 positivity was recognized if ≥5% of cancer cells exhibited moderate (2+) membranous protein expression. Benign tonsil samples served as a positive control. MMR protein expression was tested by IHC using antibody clones [MLH1, M1 antibody; MSH2, G2191129 antibody; MSH6, 44 antibody; and PMS2 (Abcam cat. No. ab203457, RRID:AB 2889230), EPR3947 antibody (Ventana Medical Systems, Inc.)]. The complete absence of protein expression of any of the four proteins tested (0+ in 100% of cells) was considered deficient MMR.

Microsatellite-stable metastatic prostate tumors were identified using NGS by Caris Life Sciences. NGS was performed on genomic DNA isolated from formalin-fixed paraffin-embedded (FFPE) tumor samples after microdissection using the nextSeq platform (Illumina, Inc.) at the Caris Life Sciences laboratory. No normal tissue pairs were sequenced. A custom-designed SureSelect XT assay was used to enrich 592 whole-gene targets (Agilent Technologies). All variants were detected with >99% confidence based on allele frequency and amplicon coverage, with an analytic sensitivity of 5% and an average sequencing depth of coverage >500. Tumor enrichment was achieved by collecting targeted tissue using manual microdissection techniques to molecular testing. Genetic variants were interpreted by board-certified molecular geneticists and classified as pathogenic, likely pathogenic, variant of unknown significance, likely benign, or benign according to the ACMG standards. When assessing mutation frequencies of individual genes, pathogenic and likely pathogenic were counted as mutations, whereas benign, likely benign, and variants of unknown significance were excluded.

On the basis of the mapped WES data, tumor mutational burden (TMB) was derived by counting all nonsynonymous missense, nonsense, in-frame insertion/deletion and frameshift mutations found per tumor that had not been previously described as germline alterations in dbSNP151, Genome Aggregation Database (gnomAD) databases, or benign variants identified by Caris geneticists. A cutoff point of ≥10 mutations per MB (muts/Mb) was used on the basis of the KEYNOTE-158 pembrolizumab trial (23), which showed that patients with a TMB of ≥10 muts/Mb across several tumor types had higher response rates than patients with a TMB of <10 mt/MB. Caris Life Sciences is a participant in the Friends of Cancer Research TMB Harmonization Project (24).

Direct Sequence analysis was executed on genomic DNA isolated from micro dissected, FFPE tumor samples using the Illumina Novaseq 6000 sequencers. A hybrid pull down of baits designed to enrich for 720 clinically relevant genes at high coverage and high read-depth was used, along with another panel designed to enrich for an additional >20,000 genes at a lower depth. A 500 Mb SNP backbone panel (Agilent Technologies) was added to assist with gene amplification/deletion detection.

We queried the deidentified real-world evidence (RWE) outcomes dataset from the Caris Life Sciences POA registry and insurance claims data. RWE overall survival (OS) from the insurance claims repository was used to determine outcomes of patients, including OS and treatment with ADT or taxanes. To estimate OS, we determined the times of sample collection and times of either death or last contact. For outcomes of patients treated with ADT or taxanes, we determined periods in which ADT or taxanes were initiated and the time of either death or last contact. As previously reported, patient death was assumed for any patient without a claim for more than 100 days, which holds true for more than 95% of patients with a recorded death in the NDI (National Death Index; refs. 25–28). Cox proportional hazard ratios (HR) were calculated for each comparison group, and significance was determined as P values of <0.05 using the log-rank statistic.

The Qiagen RNA FFPE tissue extraction kit was used for extraction. RNA quality and quantity were determined using the Agilent TapeStation (Agilent TapeStation Laptop, RRID:SCR_019547). Biotinylated RNA baits were hybridized to the synthesized and purified cDNA targets, and the bait–target complexes were amplified in a postcapture PCR reaction. The Illumina NovaSeq 6500 was used to sequence the whole transcriptome from patients to an average of 60M reads. Raw data were demultiplexed by the Illumina Dragen BioIT accelerator, trimmed, counted, removed of PCR duplicates, and aligned to human reference genome hg19 by the STAR aligner. For transcription counting, transcripts per million molecules were generated using the Salmon expression pipeline. Immune cell fraction was inferred by Quantiseq using transcriptome data (29). Differential gene expression was analyzed using the Limma R package (q < 0.001, logFC > 1.5, −log₁₀ FDR > 20). The comparisons of WTS profiles between ATMm and BRCA2m with the HRP group are summarized in Supplementary Table S3.

We generated gene profiles for ATM or BRCA2-mutated tumors by considering the differential expression values based on Limma analyses. For subsequent analyses, we included differential expressed genes as significant based on the adjusted P values of <0.01. We then used Gene Set Enrichment Analysis (GSEA; ref. 30) preranked tools on these profiles to identify enrichment of gene signatures from the C5 Go Biological Process gene signatures in MSigDB (31). On the basis of the output files, we tabulated the normalized enrichment scores that were significant on the basis of FDR. We considered NES >1.5 and FDR <0.05 as significantly associated profiles.

Statistical significance was determined using Fisher's-exact/Mann–Whitney/χ² test with Benjamini–Hochberg correction. Differential gene expression was analyzed using the Limma R package (q < 0.001, logFC > 1.5, −log₁₀ FDR > 20). GSEA was used with a P value of <0.01 cutoff. Significance was determined as P values <0.05, unless otherwise defined.

The deidentified sequencing data are owned by Caris Life Sciences.

The datasets generated and analyzed during the current study are available from the authors upon reasonable request and with permission of Caris Life Sciences. Qualified researchers may contact the corresponding author with their request.

This study was conducted in accordance with the guidelines of the Declaration of Helsinki, Belmont report, and U.S. Common Rule. In keeping with 45 CFR 46.101(b)32, this study was performed using retrospective, deidentified clinical data. Therefore, this study is considered institutional review board exempt, and no patient consent was necessary from the subjects.

We assembled a cohort of 1,187 patients with mPC from the POA consortium. We used the DNA/RNA-sequencing platform at Caris Life Sciences and executed according to the ACMG standards (33). We excluded microsatellite-instable (MSI) tumors to prevent elevation of bystander mutation events in BRCA2, ATM, or other HRR genes (34). This yielded 88 (7.41%) and 98 (8.26%) mPCs with pathogenic mutations in ATM (ATMm) and BRCA2 (BRCA2m), respectively (Fig. 1A). We identified a control group that included 808 mPCs (68.07%) that were HRP based on wild-type status of 12 HRR genes exclusive of ATM or BRCA2, which were identified and reported as genes that regulate HRR (9, 15, 35, 36). These genes included RAD51B, RAD51C, RAD51D, RAD54L, FANCL, PALB2, CHEK1, CHEK2, CDK12, BRCA1, BARD1, and BRIP1. When mutations in any of these genes were observed, we considered these mPCs to have non-BRCA2/ATM HRR-deficient cancers (HRDother, 16.26%; Fig. 1A). We removed three ATM and four BRCA2-mutated samples with concurrent HRDother mutations. Therefore, ATMm, BRCA2m, and HRDother subgroups are exclusive for the named mutations. These biopsy samples were obtained from metastatic sites, including bone, lymph, liver, lung, and bladder (Fig. 1B). We did not observe significant differences between the frequencies of ATM and BRCA2 mutations across all specimen sites (Fig. 1C). With respect to mutation distribution, few of the mutations in ATM and BRCA2 were recurrent (Fig. 1D and E), indicating the lack of somatic hotspot mutations.

To evaluate survival outcomes associated with each molecular subset of mPCs, we first stratified the samples based on their hormone therapy treatment status. We defined noncastrate and castrate mPCs based on the relationship to the onset of androgen-deprivation therapy (ADT): degarelix, goserelin, leuprolide, or triptorelin (with or without abiraterone, apalutamide, bicalutamide, or darolutamide). Tissues obtained within <3 months of ADT initiation were considered noncastrate (n = 572), whereas tissues collected >3 months after ADT initiation were categorized as castrate (n = 328). We estimated OS based on our prior study (25) that considered the time of sample collection to last contact with the patient with mPC. In non-castrate and castrate samples, we compared the outcomes of each group with the HRP group based on Kaplan–Meier curves, HRs with confidence intervals (CI; Supplementary Fig. S1A–S1D). To determine whether mPCs with ATM, BRCA2, and HRDother mutations have different outcomes as compared with one another, we compared the outcomes based on the same analyses (Supplementary Fig. S1E–S1F). Altogether, although some trends were noted, ATM, BRCA2, and HRDother groups did not exhibit significant differences in OS (from sample collection to inferred death) in any comparison.

We next sought to determine whether mutations in BRCA2 or ATM predict clinical response to specific classes of standard therapies. On the basis of our definitions of castration status, we again compared all experimental groups and evaluated these outcomes based on dates of the beginning of a given treatment with the last contact for each patient (25). We evaluated therapies that inhibit AR or AR signaling (ART), including abiraterone, enzalutamide, bicalutamide, darolutamide, and ADT, including degarelix, triptorelin, goserelin, and leuprolide. We compared clinical outcomes of mPCs with ATM, BRCA2, and HRDother mutations against HRP, which was calculated from the start of ADT and established Kaplan–Meier curves in the noncastrate (Fig. 2A) and castrate settings (Fig. 2B). In parallel, we calculated HRs in the noncastrate and castrate groups, respectively (Fig. 2C and D). To determine whether there were differences in treatment outcomes between mPCs with ATM, BRCA2, and HRDother, we then compared treatment outcomes between each group in noncastrate and castrate samples (Fig. 2E-H). In the noncastrate samples, we found that patients with BRCA2 mutation had significantly worse outcomes compared with the HRP group (Fig. 2A and C; HR, 1.529; 95% CI, 1.009–2.316; P = 0.044). No significant differences were found in ATMm and HRDother groups compared with the HRP group (HR, 0.875 and 1.386; 95% CI, 0.515–1.487 and 0.962–1.99; P = 0.622, 0.079, respectively). In the castrate samples, we observed significantly worse outcomes in mPCs with either ATM, BRCA2, or HRDother mutations after ADT/ART therapies (Fig. 2B and D; HR, 1.6, 1.7, and 1.8; 95% CI, 0.998–2.423, 1.094–2.583, and 1.245–2.515; P = 0.049, 0.017, 0.001, respectively) compared with HRP tumors. Reflecting our other observations, BRCA2-mutated mPCs demonstrated a trend toward worse outcomes when directly compared with ATM-mutated mPCs, but only in the noncastrate setting (Fig. 2E–H). Altogether, although BRCA2 mutations were associated with worse clinical outcomes with ADT/ART treatment in both noncastrate and castrate settings, ATM and HRDother mutations only exhibited worse outcomes in the castrate setting.

We conducted similar analyses examining clinical outcomes after taxanes (cabazitaxel and docetaxel) treatment. We first compared the OS of the mPCs with ATM, BRCA2, and HRDother mutations against HRP, which was calculated from the start of taxanes and established Kaplan–Meier curves in the noncastrate (Fig. 3A) and castrate samples (Fig. 3B). We also determined HRs in the noncastrate and castrate groups, respectively (Fig. 3C and D) and then examined potential differences in treatment outcomes between mPCs with ATM, BRCA2, and HRDother in noncastrate and castrate samples (Fig. 3E and F). In the noncastrate samples, we did not observe significant differences in clinical outcomes in the mPCs with ATM, BRCA2 and HRDother mutations as compared with the HRP group (Fig. 3A and C). In the castrate samples, ATM, BRCA2, and HRDother mutations exhibited worse clinical outcomes based on notable trends (HR, 1.89, 1.692, and 1.483; 95% CI, 0.896–3.986, 0.849–3.375, and 0.819–2.685; P = 0.09, 0.131, 0.191, respectively; Fig. 3B and D). With respect to clinical outcomes after taxane treatment, BRCA2-mutated mPCs exhibited significantly improved outcomes as compared with HRDother (Fig. 3E; HR, 0.36; 95% CI, 0.131–0.987; P = 0.038). Altogether, with regard to taxanes, ATM, BRCA2, or HRDother gene mutations exhibited trends toward worse outcomes in both noncastrate and castrate samples. BRCA2-mutated tumors exhibited improved clinical outcomes as compared with HRDother-mutated mPCs in noncastrate samples.

When we considered other forms of systemic therapies, there were limited numbers of patients who received PARPi, platinums, or combined platinum/taxane chemotherapies (n = 38, 14, 27, respectively), precluding further analyses of these subsets.

To improve our understanding of why ATM and BRCA2-mutated mPCs exhibit distinct clinical outcomes, we sought to interrogate the landscape of co-occurring molecular features. Genome-wide loss of heterozygosity (gLOH) has been associated with HRR gene mutations in mPC (37). On the basis of a computed gLOH score, mPCs with mutations in BRCA2 and HRDother genes exhibited increased gLOH scores compared with patients in the HRP group, whereas ATM-mutated mPCs did not exhibit a significant difference (Fig. 4A). On the basis of the findings from consortium studies, genomic deregulatory events in AR or events that promote AR signaling are key drivers in mPCs (1–7). We examined whether mutations in ATM, BRCA2, or HRDother exhibited differences associated with AR mutations, AR-V7 transcripts, FOXA1 mutations, TMPRSS2-ERG fusions, or SPOP mutations (Fig. 4B). There was no significant difference in association with AR-activating mutations in BRCA2 or ATM-mutant mPCs. However, HRDother tumors were depleted of TMPRSS2 fusions relative to HRP tumors. TP53, RB1, and PTEN are recurrent genomic aberrations associated with advanced mPC (1–7) and are observed in aggressive variants and lineage-plastic forms of mPC, including neuroendocrine prostate cancer (NEPC; ref. 38). We observed that ATM-mutated tumors exhibited significantly lower rates of TP53 alterations and had lower rates of RB1 or PTEN mutations as compared with BRCA2-mutated or HRP mPCs (Fig. 4C). When examining coalterations in an unbiased fashion, we found that ATM-mutated tumors exhibited significant coamplifications with genes that reside on chromosome 11q13 that includes CCND1, FGF4, FGF19, and FGF3 (Fig. 4D). BRCA2-mutated mPCs were not enriched for these amplification events (Fig. 4D). BRCA2-mutated mPCs exhibited significantly greater TMB-High status compared with HRP in frequency (7.36% vs. 0.49% in HRP, P < 0.001; Fig. 4E) and higher median (five muts/Mb) compared with ATMm (three muts/Mb), HRDother (four muts/Mb), and HRP (three muts/Mb) mPCs (q <0.001 for all comparisons; Supplementary Fig. S2). IHC staining revealed significantly increased PD-L1 protein expression in HRDother tumors compared with HRP tumors (Fig. 4E). No significant difference was observed in BRCA2 and ATM mutant mPCs in PD-L1 staining compared with HRD other tumors. Overall, these observations indicate that a subset of BRCA2-mutated tumors, independent of MSI status, may have improved susceptibility to checkpoint inhibitors as suggested by recent clinical studies (39). Finally, we considered that treatment status may impact the rates of genomic features we observed. We thus compared the representation of these molecular markers between the noncastrate and castrate samples with ATM, BRCA2, HRDother, and HRP mutations (Supplementary Table S2). We noted enrichment of AR mutations in the castrate groups with BRCA2 or HRP (P = 0.0267, 0.0036, respectively). Otherwise, AR V7 and gLOH scores were enriched in the castrate HRP group compared with the noncastrate group (P = 0.002, 0.036, respectively). No other genomic features were enriched in the castrate compared with the noncastrate samples.

In addition to genomic features, we also examined the transcriptomes of mPCs. We first identified differential transcripts that are associated with ATM and BRCA2 mutations compared with HRP. On the basis of these analyses, ATM and BRCA2 mutations were significantly associated with distinct expression of transcripts (Fig. 5A and B, Supplementary Table S3). For specific examples, ATM-mutated tumors exhibited decreased levels of HES4 and SNHG6 (Fig. 5C and D). BRCA2-mutated tumors exhibited robust correlations with increased genes that regulate cell cycle, including E2F7 and CBX2 (Fig. 5E and F). To examine whether ATM or BRCA2 mutations may be directly associated with NEPC or lineage plasticity, we evaluated the potential differences in AR activity (40) and NEPC (38) activity based on gene signatures. We found trends in which BRCA2-mutated tumors exhibited lower AR activity (Supplementary Fig. S3A) but increased NEPC signatures (Supplementary Fig. S3B). To further characterize biology through transcriptional programs, we conducted GSEA analyses (30) after generating transcription profiles based on genes differentially enriched in the ATM and BRCA2-mutated mPCs, compared with HRP cases. We determined the relative enrichment of the transcription profiles associated with ATM or BRCA2 based on the 15,703 gene signatures from Gene Ontology (31). Here, BRCA2-mutated mPC were enriched of all 14 transcription profiles that indicate accelerated cell cycle (Fig. 5G–I). Unlike BRCA2-mutated mPC, ATM-mutated tumors were not enriched in any of the same cell-cycle profiles. Altogether, our gene expression and transcriptome profiling analyses indicated that these two groups of mPCs were associated with distinct transcriptional programs and tumor biology, with BRCA2-altered tumors demonstrating dysregulated cell-cycle control.

ATM and BRCA2 mutations are associated with divergent responses to therapies that inhibit PARPi function. In this study, we interrogated the molecular and clinical features associated with mutations in ATM and BRCA2, two HRR genes reported in clinical-grade genomic tests. We examined 1,187 patients with mPC based on samples aggregated by the Caris POA. This included tissue biopsies from metastatic sites commonly observed in patients, such as lymph, bone, and liver. Overall, 88 (7.4%) mPCs had pathogenic ATM mutations, whereas 98 (8.3%) harbored pathogenic BRCA2 mutations. These observed frequencies were consistent with prior genomic studies (1–7). Relative to PC samples that lacked mutations in HRR genes (HRP group), mutations in ATM or BRCA2 were associated with a trend toward worse clinical outcomes with taxane treatment and inferior outcomes to ADT or AR-targeted therapies. Furthermore, the genomic and transcriptional landscapes of these two subsets of mPCs exhibited divergent molecular features. Although neither ATM nor BRCA2-mutated tumors were associated with AR-activating mutations, ATM loss was associated with reduced TP53 mutations but higher amplification of 11q13 genes, including CCND1, FGF4, FGF19, and FGF3. Conversely, BRCA2-mutated mPC exhibited greater levels of TMB-high disease. When examining the transcriptome, we observed that ATM and BRCA2 mutations had distinct associations with individual genes and signaling programs. Particularly, ATM-mutated tumors had reduced expression of HES4, a key transcription factor in the Notch pathway, whereas BRCA2-mutated tumors exhibited enrichment of the cell cycle as well as known genes in this pathway. In summary, although ATM and BRCA2 mutations each account for close to 10% of mPCs, one should devise distinct strategies in treating patients with mPC that harbor these mutations.

Our results indicate that ATM and BRCA2 mutations were associated with worse clinical outcomes after receiving treatments that target androgens or AR signaling. We also observed differential trends in median survival with regard to taxanes. Too few patients in our cohort were treated with PARPi. However, as these mutations indicate worse clinical outcomes in patients receiving hormonal or taxane therapies, our findings support the need to investigate targeted therapies against ATM or BRCA2 mutations, which may ultimately require different treatment strategies. As PARPi therapies are becoming more widely used in prostate cancer, we also anticipate prospective collection of more ATM- versus BRCA2-altered patients receiving PARPi treatment over time. Finally, outcomes based on insurance claims represent a significant limitation when comparing our study with other clinical studies with more detailed health records for all patients. Although we identified significant differences between ATM and BRCA2-mutated mPCs, there are merits in pursuing these observations in sizable patient cohorts that have aggregated detailed clinical records and omics data.

Our results indicate that ATM and BRCA2 both predicted inferior patient outcomes. However, prior clinical and functional studies indicated that these two classes of mutations behave differently in patients and prostate cancer cell lines (10–13). This finding would also indicate that ATM and BRCA2-mutated mPCs may have distinguishing co-occurring molecular features. This was the case when examining both the genomes and transcriptomes of these mPC. ATM-mutated tumors exhibited significantly less features of DNA damage, including lower genomic instability (gLOH scores), fewer TP53 mutations, and lower TMB. These observations potentially suggest that ATM-mutated tumors, as compared with all other groups, are unable to tolerate additional perturbations that can cause functional loss of DNA repair. In support of this notion, we previously identified that ATM suppression in cell lines led to increased sensitivity to the inhibition of ATR, another DNA damage-sensing protein (14). To expand on these ideas in the clinical setting, one might consider using novel agents (other than PARPi) that attenuate the function of DNA repair genes as treatment strategies for ATM-mutated tumors. ATM-mutated tumors also exhibited enrichment for amplifications in 11q13 genes, including CCND1 and FGF genes. The FGF signaling pathway promotes the development of double-negative prostate cancers (DNPC) that are more resistant to AR-targeted therapies (41). DNPCs demonstrate limited androgen signaling but are also different from NEPCs, which are histologically and molecularly distinct from adenocarcinomas. DNPCs are associated with worse response to ADT or ART, which coincides with our observations in which castrate samples with ATM mutations had worse clinical outcomes with abiraterone and enzalutamide treatment. At the transcriptome level, ATM-mutated mPC had significant reduction of HES4 expression and did not exhibit increases in AR or NEPC activity. Hair and enhancer of split homolog-4 (HES4) is a transcription factor that responds to the Notch pathway (22, 42) and activated by Delta-like ligands, including DLL3. DLL3 is upregulated in about 77% of NEPC and can activate HES4 (43, 44). In ATMm mPCs, we observed decreased HES4 expression and limited changes in the NEPC activity signature or expression of other NEPC-specific markers (INSM1, CHGA, and SYP). Altogether, ATM-mutated mPC harbor molecular features reminiscent of DNPCs, and they may theoretically show sensitivity to inhibitors of FGF signaling.

As compared with ATM mutations, we found that BRCA2 mutations generally had opposing genomic and transcriptomic features. We previously found that BRCA2 mutations were associated with higher TMB in MSI-positive prostate cancers (34), suggesting that MSI-high cases had to be excluded from this analysis. After excluding such MSI-high patients here, our results indicated that BRCA2 mutations are still associated with slightly higher TMB, even independently of MSI status. In addition, BRCA2 mutations were associated with genomic instability as evident by significantly elevated gLOH scores, another potential marker of PARPi inhibitor responsiveness (34, 45). As compared with other cancers, mPCs are overall less susceptible to immune checkpoint blockade (46). However, studies demonstrate that certain subsets of mPC may be more immunotherapy responsive (47). On the basis of our findings, a subset of BRCA2-mutated mPC with high TMB, as compared with other HRR groups, may potentially exhibit the greatest immunotherapy response. In this regard, we also found that BRCA2-mutated tumors also had greater amplification rates of PDCD1, the gene encoding PD-1. Unfortunately, in these bulk-sequenced samples, we are unable to resolve the genomics of individual cells in this heterogeneous mixture. It would be of future interest to examine, at the single-cell resolution, if these amplification events are observed in tumor, fibroblasts, or immune cells. In addition to these immunologic features, BRCA2 tumors also exhibited greater rates of concurrent RB1 loss, and robust increases in cell-cycle signaling. This would indicate that interventions that can kill dividing tumor cells could affect patients with mPC who have mutations in the BRCA2 gene. In partial support of these findings, prior studies have demonstrated that platinum agents have improved efficacy against BRCA1/2 tumor models and patients with breast cancer (48). However, studies also indicate that BRCA1 and BRCA2 are not functionally equivalent (49, 50). BRCA1 and BRCA2 mutations occur at distinct rates in breast cancers and are known to have distinct cell intrinsic and immune regulatory functions (49, 50). In addition, these studies did not specifically examine the effects of ATM mutations as compared with BRCA1 or BRCA2. Altogether, it is worth determining whether mutations in BRCA2, as compared with ATM or BRCA1, regulate specific responses to immunologic and platinum therapies in patients with mPC.

In summary, this work establishes differential genotypes and phenotypes of mPCs with tumoral mutations in either BRCA2 or ATM. Our findings indicate that these two common mutational events can be used to stratify mPCs into molecular subgroups to predict future treatment strategies or response to standard-of-care treatments. Although further mechanistic studies are required, our analysis identified specific forms of cell signaling events in each of these mPCs. These observations establish a path toward the development of precision therapies for ATM and patients with mPC who have mutations in the BRCA2 gene.

There are many clinical and pathologic factors that could have influenced clinical outcomes in these patients beyond HRD status and ATM or BRCA2 genotype. The lack of detailed clinical information in the Caris Database thus precluded multivariable analyses in the “non-castrate” and “castrate” settings to interrogate the independent contribution of HRD status or ATM/BRCA2 genotype relative to other established prognostic factors. As such, all clinical correlations should be considered exploratory and will require validation in future studies that are able to adjust for other known prognostic variables.

A. Elliott reports other support from Caris Life Sciences outside the submitted work. J. Xiu reports other support from Caris Life Sciences during the conduct of the study. E. Lou reports research grants from the American Cancer Society (RSG-22-022-01-CDP) from 2022 to 2026 and the Minnesota Ovarian Cancer Alliance in 2019, 2021, and 2022; honorarium and travel expenses for a research talk at GlaxoSmithKline in 2016; honoraria and travel expenses for laboratory-based research talks from 2018 to 2021, and equipment for laboratory-based research from 2018 to the present, and Novocure, Ltd.; honorarium for panel discussion organized by Antidote Education for a CME module on diagnostics and treatment of HER2+ gastric and colorectal cancers, funded by Daiichi-Sankyo, 2021 (honorarium donated to laboratory); compensation for scientific review of proposed printed content, Elsevier Publishing and Johns Hopkins Press; consultant, Nomocan Pharmaceuticals (no financial compensation); scientific advisory board member, Minnetronix, LLC, from 2018 to 2019 (no financial compensation); and consultant and speaker honorarium, Boston Scientific US, 2019. In addition, E. Lou is an institutional principal investigator for clinical trials sponsored by Celgene, Novocure, Intima Biosciences, and the National Cancer Institute, and holds a University of Minnesota membership in the Caris Life Sciences POA (no financial compensation). K.W. Mouw reports personal fees from Riva Therapeutics, other support from Pfizer, and personal fees from UroGen Pharmaceuticals and EMD Serono outside the submitted work. E.I. Heath reports other support from Caris Life Science during the conduct of the study. R.R. McKay reports advisory/consulting with AstraZeneca, Aveo, Bayer, BMS, Calithera, Caris, Dendreon, Exelixis, Lilly, JNJ, Myovant, Merck, Novartis, Pfizer, Sanofi, SeaGen, Sorrento Therapeutics, Tempus, and Telix; as well as research funding with AstraZeneca, Bayer, Exelixis, and Tempus. W.M. Korn reports other support from Caria Life Sciences and Invitae Corporation outside the submitted work. C. Nabhan reports other support from Caris Life Sciences outside the submitted work. E.S. Antonarakis reports grants and personal fees from Janssen, Sanofi, Bayer, Bristol Myers Squibb, Curium, Merck, AstraZeneca, and Clovis; personal fees from Astellas, Amgen, Blue Earth, Exact Sciences, Invitae, Pfizer, Eli Lilly, and Foundation Medicine; and grants from Novartis, Celgene, and Constellation outside the submitted work; and reports a patent for an ARV7 biomarker technology issued and licensed to Qiagen. No disclosures were reported by the other authors.

J. Hwang: Conceptualization, data curation, supervision, methodology, writing–original draft, writing–review and editing. X. Shi: Data curation, validation, writing–original draft, project administration, writing–review and editing. A. Elliott: Resources, data curation, formal analysis, methodology, writing–review and editing. T.E. Arnoff: Data curation, formal analysis, writing–original draft. J. McGrath: Resources, data curation, software, formal analysis, writing–original draft. J. Xiu: Data curation, formal analysis. P. Walker: Data curation, formal analysis. H.E. Bergom: Data curation, formal analysis, supervision, writing–review and editing. A. Day: Data curation, software, formal analysis, supervision, methodology. S. Ahmed: Investigation, visualization, writing–original draft. S. Tape: Visualization, writing–review and editing. A. Makovec: Formal analysis, visualization. A. Ali: Software, formal analysis, investigation, methodology. R.M. Shaker: Software, formal analysis, investigation, writing–review and editing. E. Toye: Validation, visualization, writing–review and editing. R. Passow: Writing–review and editing. J.R. Lozada: Writing–review and editing. J. Wang: Software, methodology. E. Lou: Project administration, writing–review and editing. K.W. Mouw: Validation, writing–review and editing. B.A. Carneiro: Validation, writing–review and editing. E.I. Heath: Validation, writing–review and editing. R.R. McKay: Validation, writing–review and editing. W.M. Korn: Supervision, validation, writing–review and editing. C. Nabhan: Resources, supervision, validation, writing–review and editing. C.J. Ryan: Conceptualization, supervision. E.S. Antonarakis: Conceptualization, supervision, writing–review and editing.

We acknowledge Dr. A. Elliott from Caris Life Sciences for facilitating our study and the Division of Hematology, Oncology, and Transplantation, University of Minnesota to provide the research funds.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.