Genomic and Clinicopathologic Characterization of ATM-deficient Prostate Cancer

The ataxia telangiectasia mutated (ATM) gene encodes a PI3K-related serine/threonine kinase involved in the maintenance of genomic integrity. ATM plays a central role in the sensing and cellular response to DNA damage, and acts as a key signal transducer in the double-strand break repair process (1). Biallelic germline mutation of the ATM gene leads to ataxia telangiectasia syndrome, characterized by neurodegeneration, immune deficiencies, and increased cancer susceptibility (2). Heterozygous deleterious germline mutations in ATM occur in around 0.5%–1% of the population and are also associated with tumor predisposition, including breast, pancreatic, lung, thyroid, and prostate cancers (2). In prostate cancer, tumoral ATM mutations may be either germline or somatic in origin, and are present in 5%–8% of castration-resistant tumors overall, an enrichment of approximately twofold over the frequency in localized prostate cancers (3–5). This enrichment suggests an association of ATM mutations with aggressive disease. Indeed, germline ATM mutations are significantly more common among patients with lethal compared with indolent prostate cancer (6) and more common among patients with high-grade disease (7), although whether this association persists for somatic ATM alterations is unknown.

Beyond acting as a potential prognostic biomarker in prostate cancer, ATM status may also be predictive of response to novel targeted therapies. Initial trials of the PARP inhibitor, olaparib, in metastatic castration-resistant prostate cancer (mCRPC) showed an impressive response rate among patients with mutations in the homologous recombination repair pathway, apparently including patients with ATM deficiency (8, 9). However, retrospective series and recent findings from prospective phase II and phase III trials examining a variety of PARP inhibitors (olaparib, rucaparib, niraparib, and talazoparib) in patients with mCRPC have shown that tumor responses, as well as progression-free survival estimates, are relatively modest in patients with the ATM mutation compared with patients with the BRCA1/2 mutation who appear to derive the majority of the benefit (10–14). This suggests that PARP inhibitor treatment may not induce synthetic lethality, or perhaps not to the same extent, in ATM-mutated prostate cancers compared with BRCA1/2-mutated cases (15, 16). However, the recent emergence of potent ATR (ataxia telangiectasia and Rad3-related protein) inhibitors has renewed interest in interrogating ATM status in prostate cancer and other malignancies (16–19), because initial phase I trials have shown favorable responses to ATR inhibitors that may be specific to ATM-deficient tumors (20). Preliminary studies also suggest a potential sensitivity of ATM-mutated prostate cancers to anti-PD-1 immunotherapy (21, 22), radium-223 treatment (23), and bipolar androgen therapy (24, 25). Therefore, the availability of an assay to detect ATM deficiency/loss would have great clinical utility in prioritizing such patients for clinical trials exploring these unique therapeutic vulnerabilities.

Before targeted therapies can be deployed for prostate tumors with ATM inactivation, numerous open questions about the clinical, pathologic, and genomic characteristics of ATM-deficient tumors will need to be resolved. For example, it remains unclear what fraction of patients with ATM germline or somatic mutations have detectable second-hit alterations and/or protein loss suggesting complete ATM inactivation and implying likely response to targeted therapy. In addition, how frequently are ATM alterations subclonal? What is the sensitivity of DNA-sequencing assays for detection of biallelic inactivation, and what assays should be used to screen patients for ATR inhibitor trial eligibility? Given that ATM inactivation and aberrations in other DNA damage repair pathway genes generally occur in the primary tumor rather than during metastatic evolution (26), many of these questions can be resolved by studying primary prostate cancers, which are more readily available compared with metastatic samples. Herein, we developed and genetically validated a clinical grade IHC assay to detect ATM protein loss and used it to assess ATM status in primary prostate cancers with known germline ATM mutations, as well as to screen for ATM inactivation in a cohort of more than 1,000 primary tumors. We comprehensively compared ATM protein status with DNA-sequencing data and characterized the largest group of ATM-deficient primary prostate cancers reported to date.

With Johns Hopkins Institutional Review Board (Baltimore, MD) approval, and in accordance with the U.S. Common Rule, three patient sets were included in this study. As this study involved only retrospective analysis of previously collected tissue samples, it was performed under a waiver of consent. (i) The first set included a cohort of all radical prostatectomies from 2004–2014 with primary Gleason pattern 5 and available clinical follow-up data (n = 77), of whom 52 had previously described targeted next-generation sequencing data from the UW OncoPlex platform available (27); these cases were used for genetic validation of the ATM IHC assay described below. (ii) The second patient set included 20 primary prostate tumors with available radical prostatectomy tissue with known pathogenic germline mutations in ATM. Of these cases, 16 had ATM mutations detected during sequencing of benign seminal vesicle or leukocyte DNA performed as a part of previously described studies (6), while the remaining four had ATM mutations detected using clinical grade germline sequencing platforms (Invitae, Color Genomics) from saliva samples (12). (iii) The third patient set included seven partially overlapping tissue microarray (TMA) cohorts from radical prostatectomies performed at Johns Hopkins (Baltimore, MD). These cases were used to screen for the frequency of ATM protein loss using the IHC assay. All of these TMA cohorts have been described previously, and were selected on the basis of risk factors. Some focused on high-risk features that enriched for adverse oncologic outcomes, so they do not represent an unbiased survey of the overall radical prostatectomy population at Johns Hopkins (Baltimore, MD). In brief, these consisted of: (i) a set of control patients who lacked a HOXB13 germline mutation, not selected by outcome (n = 99; ref. 28); (ii) a set of high-grade (Gleason score 9/10) tumors at radical prostatectomy from 1998 to 2005, designed for comparison with high-grade urothelial carcinomas, not selected by outcome (n = 38; ref. 29); (iii) a set of grade-matched African-American and European-American radical prostatectomy samples from 1995 to 2005, enriched for Gleason score ≥7 pathology, not selected by outcome (n = 292; refs. 30, 31); (iv) a set of grade-matched African-American and European-American radical prostatectomy samples from 2006 to 2010, all Gleason score 4+3 = 7 and higher, not selected by outcome (n = 163; refs. 30, 31); (v) a set of patients who all developed metastatic disease after radical prostatectomy at Johns Hopkins (Baltimore, MD) from 1995 to 2011, who were evaluated for response to abiraterone or enzalutamide (n = 34; ref. 32); (vi) a cohort study of men undergoing radical prostatectomy from 1992 to 2009 with intermediate- and high-risk disease who were followed for development of metastatic disease (n = 325; ref. 33); and (vii) a set of men with biochemical recurrence following radical prostatectomy from 1992 to 2009 (n = 240; ref. 34).

ATM protein IHC was performed using the Y170 rabbit monoclonal antibody (Abcam, 32420) on the Ventana Discovery XT Autostaining System (Roche/Ventana Medical Systems). The ATM Y170 clone was used to develop a similar IHC diagnostic assay (35) and was deployed in the phase III GOLD study, which looked at combination of olaparib and paclitaxel in patients with advanced gastric cancer who have progressed following first-line therapy (36). Slides were incubated with primary antibody (1:100 dilution) after antigen retrieval in CC1 buffer, and primary antibody incubation was followed by detection with the OptiView HQ System (Roche/Ventana Medical Systems). Each TMA spot or standard histologic section containing tumor cells was visually dichotomously scored for presence or absence of nuclear ATM signal by a urologic pathologist (T.L. Lotan). The pathologist was blinded to sequencing results when reading the IHC. A spot was considered to show ATM protein loss if any tumor cells in any tumor spot showed loss of nuclear ATM, with intact staining in admixed benign prostate glands and surrounding stromal cells, endothelial cells, or lymphocytes. Spots without internal control staining were considered ambiguous and were not scored. All samples were initially screened for ATM loss by scoring TMA spots; however, for all cases with ATM protein loss on TMA, confirmatory ATM immunostaining was also performed on standard histologic tissue sections.

ERG IHC was performed on the Ventana Platform (Ventana Discovery Ultra, Ventana Medical Systems) using a previously reported genetically validated staining and scoring protocol (37, 38). In brief, this assay utilized a rabbit anti-human ERG antibody (EPR3864). ERG was scored as positive if any tumor glands showed nuclear ERG expression, negative if no sampled tumor glands showed ERG expression, or ambiguous if endothelial nuclei did not show ERG expression within the evaluated tumor core as an internal positive control.

Isogenic CWR22Rv1 cell lines with and without biallelic ATM or BRCA2 inactivation were derived via CRISPR/Cas9 genome editing as described previously (39). Cells were pelleted and formalin-fixed, paraffin-embedded (FFPE) as described previously (40). Mycoplasma testing was conducted every 3 months (latest date October, 2019) and authentication of genome editing was described previously (39). Cells were passaged no more than twice before experiments were performed.

On standard histologic sections, tumor tissue was macrodissected guided by hematoxylin and eosin (H&E)-stained section. Sections (5 × 10 μm) from FFPE tumor samples were used for DNA extraction. For the UW OncoPlex sequencing assay, DNA was extracted from FFPE material using the Qiagen FFPE DNA Extraction Kit according to the manufacturer's directions. DNA concentrations were quantified with the Qubit Fluorometer, using a Quant-iT dsDNA High Sensitivity Assay Kit (Invitrogen). For the Myriad HRD Plus sequencing assay, sections from FFPE tissue were first incubated in proteinase K followed by DNA extraction performed using the Promega Maxwell 16 LEV FFPE Plus Kit (AS1290, Promega) according to the manufacturer's instructions.

Targeted next-generation sequencing was performed in a subset of cases using the UW OncoPlex assay as described previously (27), and for all cases we performed the HRD Plus Assay (Myriad Genetics). The HRD Plus assay has been described previously in detail for BRCA1/2 sequencing (41); here, it was modified to also include the entire coding region of ATM and a panel of 106 additional genes (Supplementary Table S1). Briefly, the HRD Plus assay uses a custom hybridization capture method employing Integrated DNA Technologies' xGen Hybridization Capture Technology (Integrated DNA Technologies). Sequencing was performed on an Illumina HiSeq2500 using a 200 cycle HiSeq Rapid SBS kit v2 and a HiSeq Rapid PE cluster kit v2.

Average coverage for ATM after alignment to the target regions and removal of nonclonal reads was 687 (range, 217–1,262). Novel variants identified by tumor sequencing using the HRD Plus assay were classified using a process that is consistent with the published standards and guidelines for clinical testing from the American College of Medical Genetics and Genomics (42). Variants were classified into one of five categories: deleterious, suspected deleterious, variant of uncertain significance, favor polymorphism, and polymorphism. Both deleterious and suspected deleterious variants were considered pathogenic. Variant classifications are stored in a classification database and can be retrieved each time they are observed during routine testing.

Clinicopathologic characteristics of tumors with and without ATM loss were compared using the χ² test and the Wilcoxon rank-sum test for categorical and continuous variables, respectively. Metastasis-free survival (MFS) was estimated with Kaplan–Meier curves, both univariate and adjusted for Gleason score and TMA set (43). A proportional hazards frailty model was used to evaluate the prognostic effect of ATM loss adjusted for known prognostic factors, while accounting for clustering within TMA sets (44). All analyses were performed with SAS v9.4 (SAS Institute).

ATM IHC was initially validated using isogenic CWR22Rv1 cell lines with and without biallelic ATM inactivation derived via CRISPR/Cas9 genome editing (39). The clone with ATM knockout was previously confirmed to harbor a biallelic 33-nucleotide insertion in ATM (c.3383_3384ins33; p.Q1128_E1129insTASANSF*), resulting in a premature stop codon (39). These cells lacked detectable ATM protein by immunoblotting (Fig. 1A) and by IHC performed on a FFPE cell block with the same antibody clone (Fig. 1B). Next, we performed ATM IHC on a TMA containing 77 primary Gleason pattern 5 tumors procured from consecutive radical prostatectomies at our institution (27). Overall, 84% (65/77) were evaluable for ATM IHC, defined by adequate nuclear staining in internal control benign glands and stromal cells. Of these, 12% (8/65) had ATM protein loss. Of the 65 evaluable cases, 52 had previously generated UW OncoPlex sequencing data (27), including seven cases with ATM protein loss (13%; Fig. 1C). Of the seven cases with ATM protein loss and sequencing data available for comparison, four cases had a deleterious ATM mutation and two had shallow ATM genomic deletions (LOH implying monoallelic loss; Table 1; Fig. 1C). The final case had no detectable ATM alterations, although tumor content was noted to be low, precluding accurate copy-number calls. Of the four deleteriously mutated ATM cases, three had definitive LOH suggesting biallelic inactivation; the remaining case was uncertain due to low tumor content, however, biallelic loss was favored (Table 1). Of the remaining 45 cases without ATM protein loss, one had a pathogenic ATM missense mutation (p.R3008C) without LOH that was evident only on manual review of the sequencing data as the variant allele fraction was below the threshold of the variant calling algorithm. Of note, this mutation has been reported in clonal hematopoiesis of indeterminate potential and chronic lymphocytic leukemia and could represent a hematolymphoid clone given the low variant allele fraction (45). In summary, the negative predictive value of the ATM IHC assay for lack of underlying pathogenic ATM mutation was 98% (44/45) and the assay was 80% (4/5) sensitive for pathogenic ATM mutations and 100% (3/3) sensitive for biallelic ATM alteration.

To verify these sequencing results on a different platform, we reisolated DNA and resequenced all seven samples with ATM protein loss, this time using the Myriad HRD Plus assay. All four pathogenic mutations were confirmed (of which two had definitive LOH; Table 1). The additional two samples with isolated LOH by the OncoPlex assay also showed isolated LOH by the Myriad assay and the single sample that was previously low tumor content and without detectable alterations on the OncoPlex assay showed ATM LOH on Myriad resequencing (Table 1). Taking the results from both DNA-sequencing assays together, all seven cases with ATM protein loss had evidence of at least monoallelic ATM inactivation, however, only three of these cases had evidence of biallelic inactivation of ATM by one or both sequencing assays. These data suggest that targeted next-generation sequencing assays may be insensitive for biallelic deleterious ATM genomic (or epigenomic) alterations leading to protein loss and may be complemented by IHC staining assays.

Next, we assessed the prevalence of ATM protein loss in primary prostate cancer cases with known ATM germline mutations. We identified 20 cases with pathogenic germline ATM mutations and available primary tumor tissue at our institution (6, 12). Tumor tissue for all cases was resequenced on the Myriad HRD Plus platform and mutations were confirmed in 19 of 20 cases (Table 2). These cases were generally high-grade, high-stage tumors, with more than half harboring Gleason grade group 5 cancer (Table 2). Of the 19 cases with confirmed germline ATM alterations, 74% (14/19) had ATM protein loss upon immunostaining of standard histologic sections of the dominant tumor nodule. Of the cases with ATM protein loss, 79% (11/14) had homogeneous loss in the entire sampled dominant tumor nodule, suggesting that the ATM mutation was likely an early clonal event (Fig. 2). Among the cases with ATM protein loss, which were evaluable for copy-number alteration or a second pathogenic ATM alteration, 70% (7/10) had genomic evidence of potential biallelic inactivation. The term “potential” biallelic inactivation is used because in cases with two ATM alterations, the phase (cis vs. trans) of the mutations was not conclusively determined (Table 2). Of interest, two of the three cases with protein loss, but lacking evidence of LOH, had a second ATM mutation detected that was classified as a variant of unknown significance (VUS; p.W3055C; p.A1272N), suggesting the potential for biallelic ATM inactivation in these cases as well. In contrast, among five cases with germline ATM mutations and intact ATM protein, four were evaluable for copy-number alteration and none had an evidence of LOH or a second somatic ATM mutation, suggesting the development of a sporadic prostate cancer in individuals with an incidental ATM germline lesion.

Next, we assessed the prevalence of ATM protein loss in several previously described sets of TMAs prepared from genetically unselected radical prostatectomy specimens. There were a total of 1,076 patients with tumor tissue sampled in these TMAs, which included the primary Gleason pattern 5 TMA set described above. Overall, 82% (885/1,076) of these samples were interpretable for ATM staining and the subset of 831 with complete clinicopathologic data and clinical follow-up information are presented in Table 3. ATM loss was present in 3% (25/831) of these evaluable tumors overall. Patients with ATM loss underwent prostatectomy more recently (P = 0.028), had higher Gleason grade (P < 0.0001), pathologic stage (P = 0.024), and CAPRA-S score (P = 0.002). ATM loss was identified more commonly in Grade group 5 (17/181; 9%) compared with all other grades (8/650; 1%; P < 0.0001). Of the cases with ATM loss, seven were from the primary Gleason pattern 5 TMA and one case overlapped with the germline cases described above, and were already sequenced (Tables 1 and 2). The remaining 17 cases were processed for DNA sequencing and restaining for ATM using standard histologic sections to assess for ATM protein expression heterogeneity (Table 4; Fig. 2). Of these, 80% (4/5) with homogeneous ATM protein loss and 50% (6/12) with heterogeneous ATM protein loss had detectable pathogenic ATM alterations. One additional case with heterogeneous protein loss had a mutation potentially affecting a canonical splice site in ATM (c.6808-1G>C), suggesting pathogenicity (Fig. 2). Of the mutated cases, only one was inferred to have a germline pathogenic ATM mutation based on variant allele frequency in the tumor sequencing, while the remainder were inferred somatic alterations or indeterminate between germline and somatic based on tumor sequencing. Among the mutated cases with homogeneous protein loss and evaluable copy-number alteration data, two of two had potential biallelic inactivation of ATM. Among the mutated cases with heterogeneous protein loss and evaluable copy-number alteration data, three of five had potential biallelic inactivation, with one additional case harboring an additional VUS in ATM.

Because many of the TMA sets that we stained for ATM had clinical follow-up information available after radical prostatectomy, we compared biochemical recurrence-free survival (BCR-FS), MFS, and prostate cancer–specific survival (PCSS) in men with and without ATM protein loss from the TMA sets. As described above, the prevalence of ATM loss was strongly correlated with Gleason grade group. In addition, the different TMA sets varied considerably both in the proportion of cases with ATM loss and with respect to survival probabilities based on their varied design and differing study populations. Thus, while ATM loss was significantly associated with an increased risk of metastasis in univariate analysis [HR, 2.32; 95% confidence interval (CI), 1.23–4.39; P = 0.009], it was not significantly associated with increased risk of metastasis after adjusting for Gleason grade and including TMA set as a random variable in a proportional hazards frailty model (adjusted HR, 1.06; 95% CI, 0.55–2.05; P = 0.857). Similar results were seen with a standard (unadjusted) and adjusted Kaplan–Meier analysis (Supplementary Fig. S1). Furthermore, ATM loss was not significantly associated with BCR-FS, MFS, or PCSS in random-effects Cox models adjusted for age, race (White vs. non-White), radical prostatectomy year, and Gleason grade group (Supplementary Table S2). Concordant results were observed if CAPRA-S replaced Gleason grade group (data not shown).

The higher frequency of ATM loss heterogeneity among cases in the TMA study (71% or 12/17) compared with those in the germline study (21% or 3/14) was interesting and suggested that ATM inactivation may occur later in tumors with inferred somatic mutations (which comprised the majority in the TMA study) compared with those with germline mutations. However, given that multifocal and multiclonal tumors are common in primary prostate cancer, it is also conceivable that the tumors with heterogeneous ATM expression represent collisions between independent clones. To resolve this, and to begin to elucidate the timing of ATM loss in primary prostate cancer, we assessed ERG status in cases with heterogeneous ATM loss. ERG gene rearrangements are present in approximately half of all prostate cancer cases in European ancestry populations (46) and are likely among the earliest genomic alterations during prostatic tumorigenesis, occurring in some cases of prostatic intraepithelial neoplasia (PIN; refs. 47–49) and before PTEN genomic deletion (50, 51). Accordingly, ERG expression (a highly validated surrogate marker of underlying ERG gene rearrangement; ref. 38) is almost always ubiquitously present in all tumor cells from a given tumor nodule if ERG is rearranged, unless the tumor represents a collision between two independent clones (51). Among cases with heterogeneous ATM loss (including three germline cases and 12 from the TMA study described above), ERG was expressed in 53% (8/15) of cases and was uniformly expressed in both ATM-positive and ATM-negative tumor cells in all cases, including two germline cases and six TMA cases (Fig. 2). These data strongly suggest that in cases with inferred somatic ATM genomic alteration (the majority in the TMA study), ATM inactivation commonly occurs subclonally and subsequent to ERG gene rearrangement, and that tumors with heterogeneous ATM expression do not commonly represent collisions of independent clones.

Because ATM loss was most commonly homogeneous and apparently clonal in primary prostate tumors with pathogenic germline ATM mutations, this raised the question of how early ATM inactivation may occur in patients with germline mutations. To begin to assess this, we screened 16 ATM protein-negative cases with known or inferred germline ATM alterations from the above cohorts for high-grade PIN, the presumptive precursor for invasive carcinoma in the prostate. We exclusively selected PIN lesions present on slides without invasive tumor by H&E, because recent studies have suggested that retrograde intraductal spread of invasive carcinoma can masquerade as PIN (52). We identified 11 cases that met these criteria, of which nine (82%) were evaluable for ATM IHC in the PIN lesions. Only one of nine (11%) showed ATM loss in the PIN (Fig. 3). Taken together, with our finding that ATM loss in germline cases is most commonly homogeneous and likely clonal, these findings suggest that ATM loss commonly occurs at or just after initial tumor invasion, rather than in precursor lesions in the prostate.

Pathogenic mutations in the homologous recombination DNA repair pathway, including the BRCA2, BRCA1, and ATM genes, are common in advanced prostate cancer, occurring in nearly 20% of mCRPC cases (4). Nearly half of these alterations have proven to be inherited at the germline level, comprising close to 10% of men with mCRPC (3, 4). Historically, the BRCA2, BRCA1, and ATM genes have been grouped together in most analyses, given that all encode proteins that are key components of homology-mediated DNA repair. The first large-scale sequencing studies in prostate cancer showed that there is a twofold enrichment of mutations in these genes in metastatic compared with primary cancers (4, 53), suggesting that these alterations are associated with aggressive disease. Indeed, a number of earlier studies had already confirmed this hypothesis for germline BRCA2 mutations (54–57). More recent studies have found that germline alterations in BRCA2 and ATM (considered together) are significantly more common in lethal compared with indolent primary prostate cancer (6), are associated with grade reclassification in active surveillance cohorts (58), and with high-grade disease in surgical cohorts (7). The correlation of germline BRCA2 and ATM mutations with adverse pathologic features, including more advanced Gleason scores and higher PSA levels (6), likely drives a large part of this association with adverse outcomes. Considering all homologous repair gene mutations in aggregate, our group has previously shown that aggressive histologic subsets of primary prostate cancer, such as ductal carcinomas (59), primary Gleason pattern 5 disease (27), and intraductal cancer (60) have mutation rates approaching or exceeding those in metastatic disease.

While nearly all prior studies have considered BRCA2 and ATM mutations together because both genes encode proteins functioning in the homologous DNA repair pathway, emerging evidence from clinical trials suggests that the roles of these two genes may in fact be quite distinctive. Prospective phase II and phase III trials of PARP inhibitors (including olaparib, rucaparib, niraparib, and talazoparib) in patients with CRPC have shown that the response rates are much more limited in patients with the ATM mutation compared with patients with the BRCA2 mutation (9–13, 61). ATM is a sensor of double-strand DNA damage, and is a kinase with a diverse range of substrates; these clinical data suggest that some of its key functions in prostate cancer may be independent of its role in homology-mediated DNA repair. In fact, ATM has also been shown to modulate growth factor signaling and to play a role in other stress response pathways (2). Accordingly, our study is the first to examine the clinicopathologic features and outcomes of ATM-deficient primary prostate tumors separately from those with BRCA2 deficiency. This work was largely made possible by the development of a robust clinical grade ATM IHC assay, which we genetically validated and subsequently used to screen hundreds of prostate tumors for ATM loss. Importantly, we confirm that ATM loss is highly enriched in Grade group 5 prostate cancers and is associated with poor outcomes in univariate analyses of surgically treated patients. However, this association with aggressive disease appears to be largely driven by a correlation with high-grade group, and we show here that ATM loss is not independently prognostic of outcomes in multivariate models, although these data require confirmation in validation cohorts.

Although ATM loss appears not to be prognostic in primary prostate cancer, it will still be critically important to identify ATM-deficient prostate tumors for trials of targeted therapies, including ATR inhibitors as well as other rational therapies (e.g., radium-223). Similar to ATM, ATR is a PI3K-like kinase, however, unlike ATM, which responds largely to double-strand DNA breaks, ATR responds to single-strand breaks and other types of DNA damage. Because of its atypical nature, ATR inhibitors have historically lagged in development behind other serine–threonine kinase inhibitors, such as AKT inhibitors (18). However, several potent and relatively specific ATR inhibitors have been characterized and have shown synthetic lethal activity in ATM-deficient tumors in vitro and in early clinical trials (18, 19). Given these promising initial results, larger biomarker-selected trials are currently ongoing (e.g., NCT04095273) and will require screening of large populations of patients with prostate cancer for inclusion. Combined with previous germline cohorts (7), this study demonstrates that screening of grade group 5 tumors will likely be most fruitful for trial recruitment, with 9% of such patients (and as many as 13% of primary Gleason pattern 5 tumors) harboring ATM loss.

Equally important to defining which populations to screen for these trials is determining which assay will be utilized for the screening protocol. Our data strongly suggest that if germline ATM sequencing is used for enrollment, at least a quarter of enrolled patients with prostate cancer may be unlikely to respond because they lack ATM protein loss and biallelic inactivation and most likely have sporadic cancers. We found that the presence of ATM protein loss strongly enriched for underlying potential biallelic inactivation of ATM, with 70% of cases with ATM protein loss (and none of the cases without protein loss) harboring a likely somatic “second hit” in the ATM gene by DNA sequencing. It will require careful clinical trials to resolve whether ATM may function in a haploinsufficient manner as seen with other tumor suppressor genes (62) and to discern whether patients with monoallelic inactivation may also respond to targeted therapy similarly to those with biallelic inactivation. In this setting, ATM IHC will likely prove very useful to determine whether ATM deficiency is likely to be monoallelic or biallelic, because next-generation sequencing may miss complex rearrangements of the gene or epigenomic gene silencing and germline sequencing will not detect somatic or epigenomic ATM inactivation. If ATM is not a haploinsufficient tumor suppressor gene, then screening with germline DNA assays alone will be inadequate for enrollment in biomarker-selected clinical trials and the addition of somatic sequencing with ATM IHC assays will also be important. While there have been limited preclinical studies examining ATM happloinsufficiency in prostate cancer, previously published studies in breast cancer mouse models seem to suggest that Atm can function in a haploinsufficient manner in some genetic contexts (63, 64). In contrast, other preclinical studies found no increased mutagenic response to radiation in mice with hemizygous loss of Atm in germline and somatic tissue (65, 66), implying potential context-dependent effects. Additional work in prostate cancer preclinical models will be important to evaluate for potential evidence of ATM haploinsufficiency.

Our study is among the first to shed light on the underlying heterogeneity in ATM-deficient tumors and the likely timing of ATM loss in the setting of germline and somatic pathogenic mutations. Among tumors with germline ATM mutations, only 21% had heterogeneous ATM loss in the dominant nodule of the primary tumor, suggesting subclonal inactivation. Thus, when ATM loss occurs in patients with germline ATM mutations, it is most commonly a relatively early driver event, similar to ERG gene rearrangement. Analogous to ERG rearrangement, which is only infrequently seen in PIN (49), we demonstrate that ATM is only rarely lost in the presumptive prostate cancer precursor lesion (PIN) in cases with germline and apparent clonal ATM inactivation. This suggests that similar to ERG rearrangement, ATM loss likely occurs just after the point of tumor invasion. In contrast to these germline cases, in the TMA study where most ATM alterations were inferred to be somatic (based on variant allele frequency), nearly 70% of cases had heterogeneous or likely subclonal ATM loss. In cases with heterogeneous ATM loss, we demonstrate that ATM inactivation most likely occurred after ERG gene rearrangement given that ERG was expressed in all tumor cells.

Whether the presence of subclonal loss implies that patients are less likely to respond to targeted therapy is an unanswered question that will hopefully be resolved in future trials of ATR inhibitors and other molecularly targeted approaches. Studies to date have suggested that DNA repair gene mutations are most commonly present in the primary tumor and all metastases (26, 67), implying that they are relatively early drivers in most cases. If this is the case, then even patients with heterogeneous ATM loss in the primary tumor may have metastases with homogeneous ATM loss that will respond to therapy. On the other hand, it is conceivable that some cases may have late subclonal loss in the primary tumor, potentially even after metastatic spread has occurred, in which case response to targeted therapy may not be assured and testing of the metastatic deposit will be critical. In an attempt to shed light on these possibilities, we examined the medical records of all cases in this study with heterogeneous ATM protein loss in the primary tumor to identify cases with available metastatic tissue samples. Unfortunately, no cases had available tissue from metachronous metastases, although one case had tissue sampled from a synchronous pelvic lymph node metastasis (ATM44). Interestingly, in this case, there was clear heterogeneous ATM loss in the tumor, with intact ATM in a synchronous pelvic lymph node metastasis (Supplementary Fig. S2). Notably, this case did not have a detectable ATM mutation in the primary tumor, likely due to the subclonal nature of ATM loss identified by IHC. Consistent with this, cases that had heterogeneous ATM loss in our study were less likely to have any detectable underlying ATM mutation (50%), compared with those with homogeneous ATM protein loss (80%), highlighting the challenges of utilizing DNA-sequencing assays in potentially genetically heterogeneous tumors. This is particularly true in prostate cancer, where copy-number alterations and rearrangements are common causes of somatic genomic inactivation and may be missed by sequencing in a mixed population of tumor cells (68). Thus, while IHC may be useful to screen primary tumors for subclonal loss, it is equally important to follow-up in heterogeneous cases with sequencing or IHC evaluation of metastatic deposits to further evaluate trial eligibility.

Our study has some limitations that warrant discussion. First, all sequencing was done using a panel-based approach, where there may be limited sensitivity for shallow copy-number alterations and complex genomic rearrangements which can lead to ATM deficiency. This could be a possible explanation for cases with ATM IHC loss, but lacking an apparent genomic alteration. Although less likely, epigenomic silencing of the ATM locus remains another possibility. Second, although we screened a very large cohort of primary prostatectomies for ATM loss, this study was done using selected TMAs, which may not be representative of underlying population or tumor heterogeneity. The analyses of clinical outcomes are based on the combined data from nine TMA sets, which differ with respect to design, clinical characteristics, dates of surgery, postprostatectomy treatment, and time of TMA construction. Although the data are retrospective, patients were selected for each TMA on the basis of clinical characteristics rather than outcome, reducing differences in study design. Furthermore, incorporation of TMA set as a random effect in the proportional hazards frailty models addresses the clustering within TMA. Despite the large sample size, only 25 patients exhibited ATM loss, which may have affected stability of multivariate models. Importantly, the lack of a validation cohort in the prognostic analyses is a significant limitation and additional studies in independent cohorts are required to establish the validity of our findings.

In conclusion, we present the largest study to date using a genetically validated ATM protein IHC assay to interrogate more than 1,000 primary prostate cancers for ATM deficiency. We show that ATM loss is strongly associated with the highest Gleason grades, and that the negative prognostic impact of ATM loss is primarily driven by Gleason grade. Because not all ATM-mutated prostate cancers demonstrate ATM protein loss, and not all cases with protein loss have underlying genomic ATM alterations, we propose a combined interrogation approach utilizing both ATM IHC and next-generation sequencing assays for upcoming clinical trials. These trials will make possible a head-to-head comparison of the assays to select the most predictive biomarker. We anticipate that some combination of these assays will lead to optimal detection of ATM inactivation in prostate cancer for consideration of genomically targeted strategies in this subset of patients.

D.C. Salles reports non-financial support from Myriad Genetics during the conduct of the study. C.C. Pritchard reports personal fees from AstraZeneca outside the submitted work. A.M. De Marzo reports grants from NIH, NCI and The Department of Defense during the conduct of the study, Janssen Research and Development (sponsored research project unrelated to this manuscript) and Myriad Genetics (sponsored research project unrelated to this manuscript), and personal fees from Cepheid Inc (consulting fees, for project unrelated to this manuscript) outside the submitted work. J.S. Lanchbury reports personal fees from Myriad Genetics, Inc. (employee and stockholder) during the conduct of the study. K.M. Timms reports personal fees from Myriad Genetics, Inc. (employee and stockholder) during the conduct of the study and outside the submitted work. E.S. Antonarakis reports grants and personal fees from Janssen, and Sanofi, Dendreon, Merck, Bristol Myers Squibb, and AstraZeneca outside the submitted work, personal fees from Pfizer and Clovis, Eli Lilly, and Amgen outside the submitted work, grants from Johnson & Johnson, Genentech, Novartis, and Constellation outside the submitted work, and has a patent for AR-V7 liquid biopsy technology issued, licensed, and with royalties paid from Qiagen. T.L. Lotan reports non-financial support from Myriad Genetics (provided sequencing assays used) during the conduct of the study and grants from Roche/Ventana outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

H. Kaur: Conceptualization, data curation, investigation, writing-original draft, writing-review and editing. D.C. Salles: Data curation, investigation. S. Murali: Data curation, investigation. J.L. Hicks: Methodology. M. Nguyen: Methodology. C.C. Pritchard: Conceptualization, data curation, investigation. A.M. De Marzo: Conceptualization, data curation, investigation. J.S. Lanchbury: Data curation, funding acquisition, investigation, methodology. B.J. Trock: Data curation, visualization, writing-review and editing. W.B. Isaacs: Conceptualization, data curation, investigation, writing-original draft, writing-review and editing. K.M. Timms: Conceptualization, data curation, formal analysis, investigation, writing-original draft, writing-review and editing. E.S. Antonarakis: Conceptualization, resources, data curation, formal analysis, funding acquisition, investigation, writing-original draft, project administration, writing-review and editing. T.L. Lotan: Conceptualization, data curation, formal analysis, supervision, validation, investigation, writing-original draft, project administration, writing-review and editing.

This work was supported by the Patrick Walsh Prostate Cancer Research Fund (to E.S. Antonarakis), the Prostate Cancer Foundation (to E.S. Antonarakis), NIH/NCI Prostate SPORE P50CA58236 (to A.M. De Marzo), and the NCI Cancer Center Support Grant 5P30CA006973-52 (to T.L. Lotan and E.S. Antonarakis) as well as the NIH/NCI U01 CA196390 (to A.M. De Marzo) and the U.S. Department of Defense Prostate Cancer Research Program Prostate Cancer Biospecimen Network Site (W81XWH-18-2-0015 to A.M. De Marzo).

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.