Abstract BACKGROUND: Pathogenic POLE exonuclease domain mutations (pPOLE) undermine mismatch error correction during DNA replication, causing somatic ultramutation and response to immunotherapy. We examined the pan-cancer landscape of POLE mutations and applied a POLE-specific variant classification model. METHODS: Comprehensive genomic profiling was performed during clinical care. Mutational signature calling was performed via decomposition using the 96-feature single-base substitution COSMIC reference signatures. A POLE-specific classification model encompassing mutation position in the exonuclease domain, TMB, presence of POLE signature, absence of other signatures, germline frequency, and other features was applied to identify pPOLE mutations causative of ultramutation. RESULTS: POLE mutation status was evaluated in 458,437 samples (425,520 tissue biopsies (TB) and 32,917 liquid biopsies (LB)). One or more POLE alterations, including pathogenic alterations and variants of unknown significance (VUS), were detected in 3.8% of samples. 19,470 total alterations were identified, 84.8% of which were missense substitutions. Application of the POLE-specific classification model identified 35 unique pathogenic variants, many of which were VUS prior to this study. 749 samples harbored a pPOLE, more than half (56.6%) of which were either p.P286R (n=245) or p.V411L (n=179). pPOLE were found in 1.4% (199/13,688) of endometrial cancers (EC) and 0.5% (270/55,981) of colorectal cancers (CRC) and were rarer in a long tail of other malignancies. The overall pPOLE rate was significantly lower in LB than TB (0.02% vs 0.17%, P<0.001) in the context of different clinical ordering patterns for EC (0.9% of LB cohort vs 3.1% of TB cohort; P<0.001) and CRC (7.5% of LB cohort vs 12.6% of TB cohort; P<0.001). Median TB TMB for pPOLE+ samples was 157.5 mut/mb, compared to 3.5 for the cohort overall (P<0.001). Similarly in LB, median pPOLE+ bTMB was 165.6 vs 2.5 overall (P<0.001). MSI-H or an MMR-associated signature was found in 17.5% of samples with pPOLE, most commonly in neurologic malignancies (75%, 27/36). Median TMB of samples with both pPOLE and MMRD was 2.4-fold higher than those with pPOLE alone (337.6 vs 139.4; P<0.001). Notably, 6.8% of pPOLE+ cases had TMB<10, which was associated median pPOLE VAF of 3.5%, compared with 25.7% among pPOLE samples with TMB≥10 (P<0.001). This pattern suggests that TMB is underestimated when tumor purity is near the limit of detection for the assay. CONCLUSIONS: pPOLE were seen in both TB and LB across cancer types. The high rate of passenger mutations underscores the utility of this POLE-specific variant classification model. Because TMB can be underestimated when tumor purity is near the limit of detection for the assay, accurate detection and classification of pPOLE is critical for identifying patients who may benefit from immunotherapy. Citation Format: Rachel B. Keller, James Haberberger, Tyler Janovitz, Alexa B. Schrock, Hanna Tukachinsky, Lei Zhong, Douglas A. Mata, Lyle V. Lopez, Zoe Fleischmann, Radwa Sharaf, Ethan S. Sokol, Garrett M. Frampton, Nimesh R. Patel, Douglas I. Lin, Geoff R. Oxnard, Erik A. Williams, Julia A. Elvin, Brennan Decker. POLE-specific variant classification strategy is critical for identifying patients who may benefit from immunotherapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 305.