Abstract 2986: Germline signatures predicting toxicity and response to CTLA4 inhibitors

Abstract

Abstract Background: Because germline microRNA-based variants predict toxicity risk to anti-PD-1/PDL-1 treatment, and CTLA4 inhibitors are often used in combination with them, we wanted to see if we could also predict toxicity and/or response to anti-CTLA4 therapy. Methods: The study included 69 patients from 3 institutions treated with anti-CTLA4 therapy alone. Patients were 23-89 years old with median age of 62. “Toxicity” was defined as grade 3 or higher toxicity and “Response” as a complete response, partial response, or stable disease. A panel of previously tested variants (n=139), age, sex and cycles of anti-CTLA4 were used to predict endpoints of interest. To balance model dimensionality we pre-screened variants and included them if they were marginally associated with our outcomes of interest via Fisher’s Exact Test or Jonckheere-Terpstra Test at a p-value threshold of < 0.2. We fit Elastic Net (EN), Random Forest (RF), and Boosted Tree (BT) models with up-sampling to predict outcomes of interest. Leave-one-out cross validation (LOOCV) was used to evaluate model performance, and the model with the highest LOOCV AUC (between all EN, RF, and BT) was chosen as the final model and re-fit using all observations. We also performed a stratified gene ontology (GO) analysis to assess biological differences between variants involved in the signatures. Our GO analysis compared gene clusters across groupings with an adjusted p-value cutoff of 0.05 against a universal genomic background. Results: 15 patients had toxicity and 21 patients had a response, with 27/139 and 25/139 variants marginally associated with these endpoints, respectively. Our toxicity model had a LOOCV AUC of 0.811 with the top variables being age, TNNT2_rs3729843, REV3L_rs465646, miR34b_c_rs4938723, and XRCC3_rs861539. Our response model had an LOOCV AUC of 0.716 with top variants being IL2RA_rs2476491, REV3L_rs465646, and IL10RB_rs2834167. Including anti-CTLA4 cycles improved the prediction accuracy for Response with an AUC 0.764. GO analysis revealed 117 pathways enriched in both the Toxicity and Response signatures, which were predominantly involved in pri-miRNA transcriptional regulation. Top GO terms in the Response signature (333 unique) were related to apoptosis and cell death. Top GO terms in the Toxicity signature (425 unique) were related to DNA replication and cellular signaling and response, including the interleukin-35-mediated signaling pathway. Conclusion: We identified germline signatures predicting toxicity and response following anti-CTLA4 therapy. These signatures differed from each other as well as from a previously identified signature of toxicity to anti-PD1/PDL-1 therapy, and GO analysis identified plausible underlying signaling pathways. Our findings continue to validate the potential of these variants to predict meaningful endpoints in cancer therapy, and thus further research in larger and more diverse cohorts is warranted. Citation Format: Joanne B. Weidhaas, Kristen McGreevy, Nicholas Marco, Nora Sundahl, Christopher R. Cabanski, Christine Spencer, Piet Ost, Donatello Telesca. Germline signatures predicting toxicity and response to CTLA4 inhibitors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2986.