Clinico-genomic Characterization of ATM and HRD in Pancreas Cancer: Application for Practice

Pancreatic ductal adenocarcinoma (PDAC) is a lethal malignancy, characterized by nonspecific symptoms, frequent early metastasis, genomic complexity, an immunosuppressive tumor microenvironment, and inherent relative therapeutic resistance, which collectively contribute to a high mortality rate (1–3). As understanding of cancer genomics continues to advance, key genetic variants will further define the biology of individual subgroups and inform targeted therapy choices (4, 5).

The ataxia-telangiectasia mutated (ATM) gene was first identified through studies of ataxia-telangiectasia syndrome, an autosomal recessive disorder (6, 7). Heterozygous germline ATM variants (gATMm) have been associated with a 2- to 3-fold increased risk of many types of cancer and contribute to familial PDAC (8–12). ATM is one of the most frequently altered somatic genes across various sporadic cancers (6). However, the prognostic value of ATM variants appears to be disease specific with improved overall survival (OS) reported in endometrial and colorectal cancers in contrast to putative worse prognosis reported in breast cancer, prostate cancer, and B-cell chronic lymphocytic leukemia (13–19).

gATMm or somatic ATM variants (sATMm) are observed in approximately 3%–4% of patients with PDAC (12, 20). The ATM serine/threonine protein kinase is known to play a critical role in homologous recombination (HR). As a main signal transducer, ATM receives an activation signal from the MRN (MRE11-RAD50-NBS1) complex, which senses and responds to double-strand DNA damage (21). This phosphorylation activation signal is transduced through CHEK2 and activates overlapping downstream effectors including BRCA1 for HR DNA repair, as well as TP53 for apoptosis and senescence, which is counterregulated by MDM2 (21).

Further defining ATM variants and their contribution to HR deficiency (HRD) in PDAC is of particular importance due to the increasing relevance of HRD in treatment selection (22). ATM, along with other candidate genes such as BAP1, BARD1, BLM, CHEK2, FAM175A, FANCA, FANCC, NBN, RAD50, RAD51, RAD51C, and RTEL1, are considered non-core HR genes (22, 23). The role of platinum and PARP inhibitor (PARPi) therapies for carriers of non-core HR variants is under active investigation (NCT04666740, NCT04123366).

In contrast, BRCA1, BRCA2, and PALB2 are well-established core-HR genes with DNA-damage signal effector function further downstream, and carriers of pathogenic BRCA1/2 or PALB2 variants derive enriched benefit from platinum and PARPi therapies (24–27). In a randomized phase II trial of gemcitabine and cisplatin with or without veliparib in patients with germline BRCA/PALB2 variants, the 2-year OS rate for the entire cohort was 30.6% [95% confidence interval (CI): 17.8–44.4], and 3-year OS was 17.8% (95% CI: 8.1–30.7), underscoring the benefit of platinum-based therapies in BRCA/PALB2-mutant (core-HRD) patients (26). In the Pancreas Olaparib Ongoing study, patients with germline BRCA variants and metastatic PDAC had extended progression-free survival (PFS) on olaparib in comparison with the placebo (7.4 vs. 3.8 months; HR: 0.53; 95% CI: 0.35–0.82; P = 0.004; ref. 25).

Carriers of core-HR variants frequently display genomic signatures of HRD including large-scale state transition (LST), mutational signature 3 (Sig3), telomeric allelic imbalance (TAI), and loss of heterozygosity (LOH; refs. 22, 23). Across various malignancies, including breast and ovarian cancers, patients with a BRCA1/2 variant have been observed to have a higher genomic instability score (GIS), calculated as the sum of LOH, LST, and TAI (28–30). In an analysis investigating the association of BRCA1/2 variants, HR status, and platinum response in triple-negative breast cancer, Telli and colleagues observed that BRCA1/2 variants were significantly associated with high HRD scores (GIS ≥ 42; ref. 28). Using genomic analyses to identify additional patients beyond those with germline BRCA/PALB2 variants that may benefit from platinum and PARPi therapies is of distinct clinical relevance in PDAC. For instance, PARPi trials in ovarian cancer have investigated the role of GIS as a predictive biomarker for response in exploratory analyses (31).

In PDAC, a large study of 391 patients with whole-genome sequencing identified 29 (7%) patients with HRD in the absence of a germline core-HR pathogenic variant (23). Notably, there were six individuals with gATMm of which none displayed any structural or mutational indicators of HRD (23). While these data suggest that ATM-mutant PDAC does not confer HRD in comparison with core HR-gene variants, the role of HR in ATM variants in PDAC as well as its association platinum and PARPi response has yet to be established (22, 23). Furthermore, epistatic relationships between ATM and other genes of significance, such as CHEK2, TP53, and MDM2 as well as other canonical PDAC oncogenic drivers such as KRAS, CDKN2A, and SMAD4, remain to be characterized. Herein, we evaluate a cohort of patients with PDAC and ATM variants and report comprehensive clinico-genomic analyses to define the association of ATM variants respective to HRD and inform clinical implications.

Memorial Sloan Kettering Cancer Center (MSK) databases were queried with Institutional Review Board oversight and in accordance with the Declaration of Helsinki to identify patients with PDAC and gATMm or sATMm. Written informed consent for genomic profiling was obtained (IRB# 12-245, NCT01775072) from all participants with PDAC included in this analysis.

We limited the time course of the search to the period when somatic and germline sequencing was routinely conducted at MSK and with the last person enrolled in early 2021 to ensure adequate follow-up (September 2015 to January 2021). Patients with PDAC were identified with billing codes and tumor registry query. Germline and somatic ATM variants were identified by evaluating pathology and genomic reports. MSK Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT), an institutional protocol all patients had previously consented to, yielded genomic data and profiling (32). Patients with pathogenic or likely pathogenic ATM variants as adjudicated by OncoKB (www.oncokb.org) were included in this analysis; patients with neutral ATM variants, variants of unknown significance (VUS), or with BRCA1/2 or PALB2 covariants, were excluded (Supplementary Fig. S1; ref. 33). Clinical, pathologic, radiographic, and demographic data were abstracted through medical record review. Somatic variant frequencies observed in our ATM-mutant PDAC cohort were compared with frequencies observed in a non–ATM-mutant PDAC cohort at MSK and to an independent PDAC cohort previously published by Queensland Centre of Medical Genomics (QCMG) through cBioPortal (34, 35).

Allelic copy-number states for g/sATMm were determined using previously described methodology (36). Allele-specific copy-number inferences were adjudicated using the FACETS algorithm (37). To determine zygosity, the observed variant allele frequency of germline or somatic alterations in the tumor were compared with their expected values to confirm copy-neutral LOH (CN-LOH) event targeting the wild-type allele. In addition to CN-LOH, other types of somatic variants were also considered for biallelic loss. For example, patients with a germline and a somatic ATM variant or those with two or more somatic ATM variants affecting the same gene, were interpreted as harboring biallelic inactivation.

Genome-wide copy-number segmentation and allele-specific calls were determined using FACETS v0.5.14. LST, loss of heterozygosity (HRD-LOH), and TAI were computed using facets-suite v2.0.6 package (https://github.com/mskcc/facets-suite). GIS was calculated as the unweighted sum of LST, HRD-LOH, and TAI scores (28, 38). GIS was derived for core HR–mutant PDAC (cHRm: BRCA1/2, PALB2) from an independent dataset to evaluate performance and accuracy (22). We also queried all ATM-mutant cancers with GIS available from the MSK-IMPACT database without clinicodemographic restrictions, and evaluated HRD-enriched cancers (ovary and breast) and select gastrointestinal cancers (esophagogastric, colorectal, hepatobiliary, and pancreas) for a disease-specific GIS comparison. Samples with available GIS data (N = 348) were included in analysis.

Patient demographics and treatment information were characterized using descriptive statistics. The data were summarized using the frequency and percentages for categorical variables and the median and range for continuous features. Wilcoxon rank-sum test and Fisher exact test were used to compare covariate distributions between groups. OS was calculated from date of diagnosis to death or last follow-up. PFS was restricted to advanced disease patients and calculated from the date of first-line treatment until date of first progression of disease (POD) or death, whichever occurred first. Patients who discontinued frontline treatment due to reasons other than POD and received other therapy were censored at the end of frontline therapy. OS and PFS were estimated using the Kaplan–Meier method and compared between subgroups using a log-rank permutation test (39). GIS comparisons, detailed zygosity analyses, and germline and somatic groups were evaluated by the Wilcoxon rank-sum test. A linear regression model was constructed to exam whether zygosity status was associated with GIS in all ATM-mutant cancers combined. The model was controlled for different cancer types and GIS was log transformed. Statistical analyses were performed using R Version 3.6.0 (R Foundation for Statistical Computing Vienna). All P values were two sided and P < 0.05 was considered to indicate statistical significance. FDR was adjusted for q < 0.05 for false positives (40).

All key data generated and analyzed for this study are available in article and Supplementary Data, with the exception of identifiable information. Any additional inquiries should be referred to the corresponding author.

A total of 68 (3.7%) patients with ATM variants were identified among 2,379 patients with PDAC between September 2015 and January 2021. For the cohort of patients with pathogenic ATM variants, 46 (2.5%) patients were confirmed to have a gATMm (N = 24, 52%) or sATMm variant (N = 22, 48%; Supplementary Fig. S1). Twenty-four (52%) patients had advanced disease (N = 1 stage III, N = 23 stage IV) and 22 (48%) had early-stage disease (N = 1 stage I, N = 21 stage II) at diagnosis (Table 1). There were 26 (57%) males and 20 (43%) females. Forty-two (91%) patients had adenocarcinoma and 4 (9%) with early-stage disease had colloid carcinoma.

The median age at diagnosis for gATMm was 64 years (range, 24–84) and 68 years (range, 44–88) for sATMm, whereas the median age of diagnosis for unselected PDAC is 71 years (3). The distribution of age at diagnosis between gATMm and sATMm was not statistically different when compared in the early and advanced disease cohorts separately (respectively, P = 0.09 and P = 0.62). In the gATMm cohort, 4 (17%) reported a family history of PDAC and in the sATMm cohort 2 (9%) reported a family history of PDAC. Of the 6 patients with a family history of PDAC, 5 were diagnosed with early-stage disease and 1 patient was diagnosed with metastatic disease. In addition, 6 patients reported a personal history of another cancer type. One patient with a gATMm variant had a history of breast cancer. Five patients with an sATMm variant had a personal history of cancer, including 1 patient with appendiceal carcinoma, 1 patient with B-cell lymphoma and melanoma, 1 patient with breast cancer, 1 patient with lung adenocarcinoma, and 1 patient with squamous cell carcinoma of the tonsil. Of the 6 patients with a personal history of cancer, 4 were diagnosed with early-stage disease and 2 were diagnosed with metastatic disease. Detailed clinicodemographic descriptors are included in Table 1.

Among 46 patients with PDAC harboring pathogenic ATM variants, 44 patients (96% of cohort) underwent both germline and somatic genetic testing and 2 (4%) had only somatic testing (Fig. 1). Twenty-four (52%) patients had a pathogenic gATMm: 11 (46%) advanced disease and 13 (54%) had early-stage disease at diagnosis. Twenty-two (48%) patients had an oncogenic somatic ATM variant only (sATMm). Zygosity determination was available in 42 (91%). Twenty-seven (64%) patients had biallelic variants and 15 (36%) had monoallelic variants. Seventeen patients had gATMm and somatic LOH, 8 had gATMm with a different somatic ATM variant, and 2 had multiple somatic ATM variants (Supplementary Fig. S2). For patients with gATMm: 18 had biallelic (75%), 2 monoallelic (8%), and 4 indeterminate (17%) zygosity status. For patients with sATMm only: 9 biallelic (41%) and 13 monoallelic (59%). The median tumor mutation burden (TMB) for the entire cohort (N = 46) was 3.4 mt/Mb (range, 0.9–7.9). The median TMB for those with a gATMm and sATMm were 2.6 mt/Mb (0.9–5.3) and 3.5 mt/Mb (1.0–7.90), respectively. The median TMB in the entire MSK pancreas cohort (N = 2,379) from cBioPortal for Cancer Genomics (http://cbioportal.org, accessed February 28, 2021) was 3.5 mt/Mb.

The four common canonical somatic driver variants for PDAC were observed less frequently in the ATM-mutant PDAC group (N = 46) compared with non–ATM-mutant PDAC cohort (N = 2,311) and compared with an independent QCMG cohort (N = 456), ATM-mutant PDAC (N = 46):non–ATM-mutant PDAC (N = 2,311):QCMG (N = 2,311), respectively: KRAS (83%:90%:90%), CDKN2A (24%:36%:16%*), and SMAD4 (15%:23%:18%) (*Structural variants not included; Fig. 1; Supplementary Figs. S3A and S3B). Interestingly, 5 of 8 patients (63%) with wild-type KRAS had a GNAS variant, 1 (13%) had a SMAD4 variant, and none of these patients had a TP53 or a CDKN2A variant (Fig. 1). From the MSK PDAC cBioPortal cohort (N = 2,357), TP53 variants were mutually exclusive with ATM variants (q < 0.001 and P < 0.001), whereas CHEK2 variants were rare and tended to co-occur (q = 0.005 and P = 0.003) with ATM variants. For TP53, which is mutated in about 70% of all PDAC (N = 2,357), only 5 (11%) patients in this cohort with a pathogenic ATM variant (N = 46) had a TP53 covariant. All of these TP53 covariants were different. Four of the five tumors had monoallelic ATM variants and one had an indeterminate zygosity status, suggesting less oncogenic dependency on ATM with TP53 covariants. One patient with wild-type TP53 had MDM2 amplification. This finding of mutual exclusivity between ATM and TP53 was confirmed from an external dataset (QCMG, N = 456, q < 0.001, P < 0.001; ref. 41). Interestingly, TP53 variant remained mutually exclusive for other ATM-mutant cancers (N = 348), though a higher percentage of TP53 covariants was observed in other cancers relative to PDAC (Supplementary Fig. S4).

GIS was computed for the ATM-mutant PDAC cohort (N = 46) and was successfully derived for N = 33 (72%) patients (limited by low purity and low variant numbers). From the literature regarding breast and ovary cancers, GIS ≥ 42 is adjudicated as HRD (23, 28). However, in pancreas cancer, the GIS for core HR–mutant PDAC (cHRm, BRCA1/2 and PALB2) is generally lower (22). In the previously published study from our group, the median GIS derived from cHRm PDAC was 26, which was higher than the median GIS for non-core HRm PDAC (ncHRm, other HR candidate gene alterations, e.g., ATM, CHEK2, RAD51, etc.; 12, P < 0.001). Notably, GIS was higher in PDAC with biallelic loss of HR-genes compared with PDAC with monoallelic loss (24 vs. 10, P = 0.004; ref. 22). For GIS derived for N = 33 ATM-mutant PDAC in our cohort, the median GIS was 11 (range, 2–29). No differences were observed between gATMm and sATMm groups [median of 8 (range, 3–28) vs. median of 14 (2–29), respectively, P = 0.17], nor between monoallelic and biallelic ATM loss groups [median 7 (2–24) vs. median 14 (4–29), respectively, P = 0.20; Fig. 2]. We also examined GIS per detailed zygosity classification. Because of the small number of samples for zygosity calling, statistical comparison was not feasible; however, no apparent trend in GIS differences was observed among ATM zygosity groups: 11 for heterozygous (monoallelic), 7 for germline and somatic variants, 13 for somatic LOH and germline variant, and 1 for multiple somatic variants.

GIS evaluation in selected ATM-mutant cancers with a high frequency of HRD (N = 348) was performed to evaluate whether the low likelihood of HRD in PDAC was cancer-specific (Fig. 3A). This analysis included ovary (N = 27), breast (N = 80), colorectal (N = 137), esophagogastric (N = 34), and hepatobiliary (N = 37) cancers. GIS distributions were significantly among the different cancer types. Median GIS was 11.0 (range, 2–29) for pancreatic, 25 (3–56) for ovarian, 18 (3–55) for breast, 18 (1–34) for hepatobiliary, 13.5 (1–49) for esophagogastric, and 11 (1–39) for colorectal cancer. Median GIS was higher in HRD-prone ATM-mutant cancers: ovary [25, (3–56)] and breast [18, (3–55)] cancers compared with ATM-mutant PDAC [11, (2–29); P < 0.001 and P = 0.003, respectively]. From all pan-cancer ATM-mutant cohort combined (N = 348), biallelic ATM-mutant cancers had higher GIS [18, (1–55)] compared with monoallelic ATM-mutant cancers [10, (1, 56); P < 0.001] and biallelic ATM variants were more common in cancers without TP53 covariants, which is consistent with previous studies of HRD in other cancers (Fig. 3B; refs. 22, 42). The significant finding remained after controlling for the variability of GIS across different disease types (monoallelic vs. biallelic: beta = −0.5, P < 0.001). ATM-mutant cancers with wild-type TP53 and MDM2 amplification were rare among the pan-cancer ATM-mutant cohort (N = 18, 0.5%) and had a higher median GIS relative to cancers without MDM2 amplification (23.5 vs. 14, respectively; Fig. 3C). No formal comparison was done due to limited sample size.

There were 24 patients with advanced disease at diagnosis and they received a median of 2 lines of treatment (refs. 1–5; Table 2). Of these 24, 16 (67%) received platinum-based therapy as first-line (frontline) treatment for metastatic disease with 5-fluorouracil, leucovorin, irinotecan, and oxaliplatin (FOLFIRINOX). Eight patients received frontline non–platinum-based therapy: 7 gemcitabine and albumin-bound paclitaxel and 1 single-agent gemcitabine. The best overall response to frontline platinum (N = 16) was 7 (44%) partial response (PR), 3 (19%) stable disease (SD), 5 (31%) progressive disease (PD), and 1 (6%) unevaluable. Of 8 patients with gATMm and advanced PDAC treated with frontline platinum therapy, 6 (75%) had a PR, 1 (13%) SD, and 1 (13%) was unevaluable. Of 8 patients with sATMm treated with frontline platinum therapy, 1 (13%) had a PR, 2 (25%) SD, and 5 (63%) PD. Of 9 patients with biallelic ATM variants, 5 (56%) had a PR, 2 (22%) SD, 2 (22%) PD, and 1 (11%) patient was unevaluable. Of 7 patients with monoallelic ATM variants, 2 (29%) had a PR, 2 (29%) SD, and 3 (43%) had PD. The best overall response to frontline non-platinum therapy (N = 8) was 2 (25%) PR, 3 (38%) SD, and 3 (38%) PD. Because of small cohort size, PFS analysis was limited. Exploratory analyses are included in Supplementary Figs. 5A and 5B).

Three patients with gATMm received olaparib single-agent (PARPi) therapy, and all had PD as best response; 2 of the 3 received a PARPi in combination with another experimental agent as part of a clinical trial and one received olaparib as a third-line therapy in a platinum-resistant setting.

The median follow-up for the cohort is 28.6 (range, 2.1–55.9) months. The median OS (mOS) for the entire cohort (N = 46) was 44 months (95% CI: 23–69). The 1-year OS rate for patients with advanced disease (N = 24) was 73% (95% CI: 56–94; Table 2; Supplementary Fig. S6). The mOS for the advanced-stage cohort (N = 24) was 19.7 months [95% CI: 12.3–not reached (NR)]. Patients with gATMm had an improved mOS compared with patients with sATMm in advanced PDAC [27.1 (95% CI: 22.7–NR) and 12.3 (95% CI: 11.9–NR) months, P = 0.02; Fig. 4A]. For patients with gATMm, the 1-year OS rate was 82% (95% CI: 62–100) versus 64% (95% CI: 41–100) in patients with sATMm. There was no statistical difference in mOS in patients with a biallelic (N = 16) or a monoallelic variant (N = 8) in advanced PDAC; the 1-year OS rate was 73% (95% CI: 54–100) versus 71% (95% CI: 45–100), respectively. Patients with wild-type TP53 were observed to have improved OS compared with patients with a TP53 variant (P = 0.04) though statistical significance could not be determined because of small sample size (Fig. 4B). These findings need to be interpretated with caution given the small numbers within the subgroups. In the pan-cancer analysis of ATM-mutant cancers, patients with wild-type TP53 had a significantly improved mOS (61.9 months; 95% CI: 56.1–78.4) compared with patients with TP53 covariants (47.4 months; 95% CI: 40.9–54.2, P < 0.001; Supplementary Fig. S7; cBioPortal accessed 3/8/2022).

The characterization and application of therapeutic targeting of non-core HR-gene variants in PDAC is an important area that remains under investigation with unknown prognostic impact or predictive benefit (22, 23, 43). BRCA1, BRCA2, and PALB2 are core HR-gene variants known to induce genomic instability, as supported by indicators including structural variant burden, signature 3, LOH, TAI, and LST scores (22–26, 44). An enriched response to platinum and PARPi therapies is well established for carriers of core HR-gene variants, while the role of these therapies for patients with germline ATM variants has yet to be defined (25, 27). ATM carriers constitute approximately 4% of all patients with PDAC when including VUS, with about 2.5% having pathogenic alterations, thus representing an important and distinct minority subgroup. Delineating the clinico-genomic consequences of ATM variants in relation to treatment response and survival is necessitated to refine therapy selection and optimize outcomes in patients with PDAC. The analysis herein presents a comprehensive review of a sizeable cohort of patients with ATM variants and PDAC and includes zygosity curation, HR signature evaluation, somatic covariant analysis, along with detailed treatment and survival outcomes.

The median age at diagnosis for the ATM-mutant cohort was 65 years, which is younger than the onset of disease observed in patients with unselected PDAC, typically being 71 years of age (45). The median age at diagnosis for patients with gATMm was 64 (range, 24–84) years and 68 (range, 44–88) years for sATMm (P = 0.09). Similar to our cohort, Hannan and colleagues observed a median age at diagnosis of 63.4 years in an ATM cohort of 22 patients (46). Likewise, Hsu and colleagues observed an earlier onset of disease in a cohort study aimed at estimating the age-specific risk of PDAC for patients with gATMm (9). In contrast, the median age at diagnosis for the gATMm cohort presented by Gower and colleagues was 67.8 years (9, 47). However, the later onset of disease observed by Gower and colleagues may be partially attributed to the small sample size (N = 10) and the inclusion of ATM VUS (40%, N = 4; ref. 47). While a younger age of diagnosis is consistently observed in patients with gBRCA1/2m and PDAC, further evaluation will determine as to whether gATMm also contributes to an earlier onset of disease (27, 48, 49).

A notable finding in our cohort was that 22 (48%) patients presented with early stage (stage I/II) at diagnosis and underwent surgical resection of their primary pancreas tumor, which contrasts with an unselected sporadic PDAC population, in which approximately 15%–20% of patients present with stage I/II disease at diagnosis (Supplementary Fig. S8; ref. 2). This observation was recently reported for gastroesophageal junction (GEJ) adenocarcinoma, where El Jabbour and colleagues observed that patients with GEJ cancer and a gATMm had a numerically higher frequency of early-stage disease at diagnosis (N = 10, 55.6%), compared with patients with GEJ and wild-type ATM (N = 106; 36.1%; P = 0.13), suggesting that gATMm may lead to a more indolent disease course and earlier stage at diagnosis (50). Furthermore, in our cohort, 4 (18%) of the patients with early-stage disease had a colloid carcinoma histology, which is generally associated with an improved prognosis relative to ductal carcinomas (51, 52). Similarly, in a histomorphology analysis of 23 patients with resectable PDAC and a germline ATM variant, Hutchings and colleagues observed that 3 (13%) patients had a colloid carcinoma, a significantly higher frequency of colloid carcinoma compared with what has been previously observed in patients with sporadic pancreas cancer (P < 0.01; ref. 53). In addition to the underlying disease biology, a family history of PDAC may increase the likelihood of an early-stage diagnosis due to measures such as increased surveillance (54). However, although 5 of the 6 patients that reported a family history of PDAC in our cohort were diagnosed with early-stage disease, none of the patients were identified through screening efforts.

The mOS for the entire (N = 46) ATM and advanced stage (N = 24) cohort were significantly longer compared with ATM wild-type PDAC populations (55, 56). The mOS for patients with advanced disease at diagnosis was 19.7 months (95% CI: 12.3–NR). Similarly, Hannan and colleagues observed a mOS of 24.7 months for patients with advanced PDAC with ATM variants (46). In a cohort of patients with resectable PDAC and a gATM variant, Hutchings and colleagues observed a median PFS of 51 months, further supporting that ATM variants may favorably impact prognosis (53). In our cohort, patients with advanced disease and gATMm had a longer mOS compared with patients with sATMm [27.1 (95% CI: 22.7–NR) months versus 12.3 (95% CI: 11.9–NR) months, P = 0.02]. This finding was also observed by Gower and colleagues, as the gATMm cohort (N = 10) had a mOS of 21.5 months and the sATMm cohort had a mOS of 11.3 months (N = 16; P = 0.001; ref. 47). In our advanced disease cohort, 9 of 11 patients with gATMm (81%) were biallelic and 7 of 14 sATMm (50%) were biallelic, which may contribute to the observed difference in platinum sensitivity between the subgroups. Although limited by small cohort size, response rate was higher in gATMm (75%) compared with sATMm (13%).

Currently, published data regarding enriched response to platinum in patients with ATM variants and PDAC are discordant. In the Know Your Tumor Program study by Pishvaian and colleagues, patients with ATM or ATR variants treated with platinum (N = 15) had a longer OS compared with HR-proficient tumors [N = 113; 2.05 vs. 1.45 years, HR = 0.35 (95% CI: 0.14–0.85), P = 0.021], while patients with an ATM or ATR variant treated without platinum (N = 7) did not [0.97 years, HR = 1.65 (95% CI: 0.6–4.59), P = 0.34; ref. 43]. Durable responses to platinum have also been observed in smaller cohort studies (57, 58). In a retrospective review of 11 patients with DNA damage repair–related PDAC (pathogenic germline BRCA1/2, PALB2, FANCC, or ATM variants), Macklin and colleagues observed extended PFS on platinum in 2 ATM patients (14 and 8 months) and in a study from our group by Lowery and colleagues, 5 of 9 patients with a gATMm who received platinum had a partial response to therapy (58). In contrast, Hannan and colleagues observed that improved OS in ATM patients remained significant following adjustment for receipt of platinum, suggesting that an ATM variant improved mOS regardless of therapy selection and may in fact be a prognostic biomarker (46). In addition, Gower and colleagues observed an improved OS in gATMm (N = 10) compared with wild-type patients (N = 180), though the majority of gATMm patients received frontline gemcitabine plus nab-paclitaxel instead of a platinum therapy (N = 8, 80%; ref. 47). While our data suggest that patients with gATMm may have improved outcome on platinum compared with patients with sATMm, this observation should be interpreted with caution given our small sample size.

Of important note, all patients in this cohort received platinum therapy as part of the FOLFIRINOX regimen, which includes 5-fluorouracil (5-FU), irinotecan, and oxaliplatin as active agents. As a result, sensitivity to platinum could not be independently evaluated, and the improved response rates observed in patients with gATMm may be partially attributed to an enriched response to irinotecan or 5-FU. Irinotecan is best known for its role as a topoisomerase I inhibitor (59, 60). The active metabolite of irinotecan, SN38, binds to the topoisomerase I–DNA covalent complex, and prevents release of topoisomerase I. As a result, religation is inhibited, and when DNA replication forks collide with the complex, double-strand breaks are generated (61). The formation of double-strand breaks induces DNA damage response signaling, and the ATM-CHEK2-TP53 pathway is activated (59, 62). In addition, a second novel mechanism for irinotecan cytotoxicity has been supported (60). Lee and colleagues found that irinotecan also binds to MDM2, an E3 ubiquitin ligase that marks TP53 for degradation. Irinotecan inhibits the MDM2/TP53 complex, resulting in an increase in TP53 transcriptional activity, including cell-cycle arrest and apoptosis (63). Given the notably high frequency of wild-type TP53 in our PDAC cohort, this novel mechanism supporting irinotecan-induced TP53 accumulation is of particular relevance, and may contribute to the observed relative improvement in OS.

The genomic landscape of patients with ATM variants also likely contributes to enhanced survival relative to ATM wild type. In our PDAC cohort, biallelic ATM variants and TP53 variants were mutually exclusive. While TP53 is mutated in 73% of all PDAC tumors sequenced through MSK-IMPACT, only 5 (11%) patients in this cohort harbored a TP53 variant, all of whom had a monoallelic ATM variant. ATM and TP53 are tumor suppressor genes vertically located in the same axis of the G₁–S cellular checkpoint pathway. The mutual exclusivity observed in our biallelic ATM cohort appears to be somewhat PDAC specific in thatTP53 covariants were observed more frequently in the other ATM-mutant cancers. Of note, mutual exclusivity of ATM and TP53 has been reported in mantle cell lymphoma, breast cancer, and lung cancer (64–66). In contrast, ATM and TP53 co-occurrence is seen in colorectal cancer (14). Given that TP53 is mutated in approximately 70% of PDAC cases and is associated with a poor prognosis, the relationship between TP53 and ATM is of particular importance (67, 68). Consistent with prior literature, the mOS for patients with advanced PDAC and wild-type TP53 (N = 21) trended longer than the mOS for patients with advanced PDAC and a TP53 variant (N = 3). As a result, the notably low variant frequency of TP53 in this cohort likely contributed to extended survival.

The totality of the detailed clinico-genomic analyses conducted herein suggests that ATM is not a core HR gene and lacks an HRD phenotype in contrast to core HR-gene BRCA1/2 and PALB2 variants. In our ATM PDAC cohort with GIS analysis (N = 33), the median GIS was 11 (range, 2–29) and no differences were observed between gATMm and sATMm groups [8 (3–28) and 14 (2–29), respectively, P = 0.17] nor between monoallelic and biallelic ATM loss groups [7 (2–24) and 14 (4–29), respectively, P = 0.20]. No indicators of HRD were found in PDAC with gATMm from whole-genome sequencing as well as transcriptomic analysis of HRD (refs. 23, 69). GIS evaluation in other ATM-mutant cancers (N = 315, excluding n = 33 PDAC) revealed that although GIS was higher in ATM-mutant ovary (25, 3–56) and breast (18, 3–55) cancers, a strong HR signature (GIS > 42) conferring HRD is also rare in other ATM-mutant cancers.

ATM is an intermediary transducer in the DDR pathway and conducts the signal of double-strand damage to downstream effectors including BRCA1 and TP53 (21). The role of loss of function in intermediary ATM appears not to be critical for HRD. Unlike a core HR protein, for example, BRCA1, which is downstream of ATM, ATM is not an effector and other downstream effectors below ATM-CHEK2 may instead receive differential signal transduction (21). Analyzing MDM2 amplification (n = 18) and wild-type TP53, we hypothesized that higher HRD signal by GIS in MDM2-amplified cancers, which might suggest a role for MDM2 in neutralizing oscillation of TP53 and potentially affecting ATM downstream signaling directionally to BRCA1 (Fig. 3B and C). However, there were only a small number of MDM2-amplified tumors from GIS-computed pan-cancer cohort and the intimate relationship among ATM, TP53, and MDM2, in determining HRD for ATM-mutant cancers was not evaluable.

High-dose ionizing radiation, relative to cytotoxic therapy, has the potential to maximize delivery of a high burden of double-stranded DNA damage to a radiation field and may have an enhanced effect in ATM-mutant cancers with biallelic ATM loss relative to monoallelic ATM loss or ATM VUS (42). In a study by Pitter and colleagues, the efficacy difference was more apparent in cancers without TP53 covariants, supporting the observations herein. Precision radiation approaches are currently being evaluated in PDAC (NCT05182112).

Furthermore, the notably distinct somatic landscape supports the observation that improved OS may be driven by a lower covariant frequency of canonical oncogenic drivers in ATM-mutant PDAC compared with non–ATM-mutant PDAC (Supplementary Fig. S3). Our analysis suggests that targeting single HR genes beyond BRCA1/2 may not have a high yield clinically in PDAC, and future therapeutic strategies need to expand beyond single genes. GIS was used as a tool to evaluate the degree of genomic scar in HRD, although it is not a clinically validated tool, similar to other measures of HRD including, structural variant burden, signature 3 derived by SigMA, and HRD-RNA, and will need further validation (22, 23, 41, 69).

Important limitations of this article include the retrospective, single-center nature of this study and a patient population with a relative lack of ethnic and racial diversity. Other limitations include limited follow-up for a small number of patients, the lack of a control cohort and the limited number of patients who received a PARPi. We specifically considered a control cohort; however, the composition of that cohort is likely to be biased on the basis of the population to be included or not. Furthermore, a different genomic or transcriptomic signature that is not captured by GIS may better reflect ATM variants. Future studies could employ alternate genomic methods to identify other possible ATM-specific signatures. Major strengths of this article include the comprehensive germline and somatic sequencing of nearly the entire cohort, with mature follow-up, the comparison of germline and somatic ATM variants, somatic covariant analysis, and zygosity status curation, detailed clinico-genomic correlations of outcome and comparison with other ATM-mutant cancers.

Patients with PDAC and pathogenic ATM variants represent an important and distinct subgroup of patients that have a relatively favorable outcome, in part associated with high frequency of wild-type TP53. We did not observe an HRD phenotype using GIS in most patients from our ATM-mutant PDAC cohort or in our pan-cancer analysis. Although there were no observed differences in GIS scores between gATMm and sATMm in PDAC, patients with gATMm tended to have better outcomes on platinum compared with sATMm patients. Although limited by our small size, ATMm is not associated with enriched response to DNA repair therapies, and furthermore, the mechanism contributing to improved outcomes in gATMm remains unknown, but is likely unrelated to HRD. ATM should be considered as a non-HR gene in PDAC.

W. Park reports research funding to the institution from Merck, Astellas, and Gossamerbio, as well as consultancy (no fee) with Onconic. A.M. Varghese reports non-financial support and other support from Lilly, Merck, Silenseed, ObiPharma, and BMS outside the submitted work. K.H. Yu reports personal fees from Ipsen, as well as other support from BMS, General Oncology, and OncoC4, Inc. outside the submitted work. E.M. O'Reilly reports grants from Genentech-Roche, Celgene-BMS, Arcus, Parker Institute, and Elicio; grants and personal fees from BioNTech, AstraZeneca, and Ipsen; personal fees from Rafael Therapeutics, CytomX Therapeutics, Merck, Autem, Boehringer Ingelheim, Seagen, IDEAYA, Noxxon, BioSapien, Cend Therapeutics, Eisai, Tempus, ZeilBio, Astellas, and Agios; and non-financial support from Thetis during the conduct of the study. No disclosures were reported by the other authors.

W. Park: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. C.A. O'Connor: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. C. Bandlamudi: Conceptualization, resources, data curation, formal analysis, validation, visualization, writing–original draft, writing–review and editing. D. Forman: Conceptualization, formal analysis, validation, writing–review and editing. J.F. Chou: Conceptualization, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. S. Umeda: Formal analysis, validation, writing–review and editing. M. Reyngold: Formal analysis, validation, writing–review and editing. A.M. Varghese: Resources, formal analysis, validation, writing–review and editing. F. Keane: Formal analysis, validation, writing–review and editing. F. Balogun: Resources, formal analysis, validation, writing–review and editing. K.H. Yu: Resources, formal analysis, validation, writing–review and editing. D.P. Kelsen: Resources, formal analysis, validation, writing–review and editing. C. Crane: Formal analysis, validation, writing–review and editing. M. Capanu: Conceptualization, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. C. Iacobuzio-Donahue: Conceptualization, resources, formal analysis, investigation, visualization, methodology, writing–review and editing. E.M. O'Reilly: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

NIH K12 Paul Calabresi Award (K12 CA184746) Wungki Park Parker Institute for Cancer Immunotherapy Pilot Award (P30 CA008748) Cancer Center Support Grant/Core Grant P30 CA008748 Center for Pancreas Cancer Research.

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.