Abstract LB262: Development of a clinically feasible aqueous proteomic signature to assess the malignant potential of small uveal melanocytic tumors

Abstract Purpose: With a prevalence of approximately 5% in the U.S., uveal melanocytic tumors of indeterminate malignant potential (UMTIMP) are primarily managed by observation for growth prior to treatment. While the malignant potential of these lesions can be assessed by intraocular tumor biopsy and genetic testing, serial tumor biopsy in the operating room is not feasible for routine management of UMTIMPs. On the other hand, the aqueous humor (AH) from the anterior chamber of the eye can be easily and safely obtained by an office-based procedure that can be repeated periodically. Proteomic signatures in aqueous fluid could potentially allow early detection of UMTIMPs at high risk of having undergone malignant transformation, thus identifying patients who would benefit from a definitive tumor biopsy and earlier treatment as indicated. Methods: AH was collected via paracentesis from 79 patients with clinically diagnosed UM at three independent sites under IRB-approved protocols. Immediately thereafter, all patients underwent a tumor biopsy for the standard-of-care molecular prognostic testing with DecisionDx-UM, -PRAME, and -UMSeq. Proteomic profiling was carried out with the Olink Explorer panel, which analyzes 3072 proteins. Statistical analysis was performed in R (version 4.3.3). Algorithm development and accuracy estimation were based on 4x 10-fold cross-validation. Results: The sample distribution was representative of a typical UM cohort: mean age 64.8±14.5 years, mean tumor diameter 11.02±3.40 mm, mean tumor thickness 4.61±3.14 mm. DecisionDx-UM identified 57/79 (72%) of tumors as Class 1 and 22/79 (28%) of tumors as Class 2. Approximately 1400 out of 3072 proteins were detected in AH samples. Proteins with statistically significant differences between Class 1 and 2 (N=386, p<0.0001) were included in further analyses. The best fitting models were k-nearest neighbor, random forest, and neural networks with accuracies of 76%, 83%, and 83%, and kappa 0.25, 0.40, and 0.36, respectively. Within the Explorer3072 panel, proteins grouped into the Immune Response and Oncology subpanels achieved an average accuracy of 81-85% and 77-89%, respectively. Conclusions: Several promising risk prediction models were identified by proteomic profiling of AH from UM patients. The proteins from the highest-performing models will be utilized in the development and validation of a testing platform that is feasible for office-based clinical testing to identify high-risk UMTIMPs that would benefit from definitive risk assessment using tumor biopsy-based testing. Citation Format: Olga Zolochevska, Julia Litvinov, Kyle R. Covington, David A. Reichstein, Amy C. Schefler, Zelia M. Correa, Jason Maarsing, Kasia Wozny, Katherina M. Alsina, J. William Harbour. Development of a clinically feasible aqueous proteomic signature to assess the malignant potential of small uveal melanocytic tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_2):Abstract nr LB262.