BACKGROUND: Deleterious ATM alterations are found in metastatic prostate cancer (PC); PARP inhibition has antitumour activity against this subset, but only some ATM loss PCs respond. OBJECTIVE: To characterise ATM-deficient lethal PC and to study synthetic lethal therapeutic strategies for this subset. DESIGN, SETTING, AND PARTICIPANTS: We studied advanced PC biopsies using validated immunohistochemical (IHC) and next-generation sequencing (NGS) assays. In vitro cell line models modified using CRISPR-Cas9 to impair ATM function were generated and used in drug-sensitivity and functional assays, with validation in a patient-derived model. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: ATM expression by IHC was correlated with clinical outcome using Kaplan-Meier curves and log-rank test; sensitivity to different drug combinations was assessed in the preclinical models. RESULTS AND LIMITATIONS: Overall, we detected ATM IHC loss in 68/631 (11%) PC patients in at least one biopsy, with synchronous and metachronous intrapatient heterogeneity; 46/71 (65%) biopsies with ATM loss had ATM mutations or deletions by NGS. ATM IHC loss was not associated with worse outcome from advanced disease, but ATM loss was associated with increased genomic instability (NtAI:number of subchromosomal regions with allelic imbalance extending to the telomere, p = 0.005; large-scale transitions, p = 0.05). In vitro, ATM loss PC models were sensitive to ATR inhibition, but had variable sensitivity to PARP inhibition; superior antitumour activity was seen with combined PARP and ATR inhibition in these models. CONCLUSIONS: ATM loss in PC is not always detected by targeted NGS, associates with genomic instability, and is most sensitive to combined ATR and PARP inhibition. PATIENT SUMMARY: Of aggressive prostate cancers, 10% lose the DNA repair gene ATM; this loss may identify a distinct prostate cancer subtype that is most sensitive to the combination of oral drugs targeting PARP and ATR.