Abstract Background: Uveal melanoma (UM) is the most common intraocular tumor of the eye in adults. About 50% of cases become metastatic, indicating poorer prognosis and survivability. There is currently no well-established blood-based method to accurately predict the risk of metastasis and disease progression of UM using proteomic analysis. This study aimed to test the efficacy and utility of proteomics in the risk assessment of UM. Methods: Proteomic analysis from serum and extracellular vesicles (EV) of 49 samples from 27 unique patients was performed to classify metastatic and non-metastatic disease. Controls were taken from 5 patients with cutaneous melanoma. Initial serum depletion was completed using the ProteoSpin Abundant Serum Protein Depletion Kit. EVs were extracted via the qEV Concentration Kit IZON system. Proteins from depleted serum and EVs were analyzed via Mass Spectrometry (MS) using a standard trypsin-based workflow. A data-dependent MS2 method on Thermo Fisher Orbitrap Lumos generated MS data. Orbitrap MS1 spectra was followed by MS2 fragmentation via ion trap using collision-induced dissociation. Machine learning was used to develop a predictive proteomic signature for metastasis. Models were trained with 3052 EV and serum protein levels and 906 serum only levels. Model sensitivity and positive predictive value (PPV) were calculated using out-of-bag bootstrap sampling conducted 500 times. Candidate models for classification of disease vs controls were used to classify holdout surveillance samples from 9 unique patients. Additionally, differential abundance analysis of each sample classification used the log(2) ratio of protein abundances with a cut off 1 and -1, indicating increased and decreased abundance, respectively. Statistical significance of abundance was set at p0.05. Results: Machine learning models revealed EVs showed higher sensitivity and PPV than serum. EV proteins with the strongest signals for disease (primary, surveillance, metastatic) vs controls include immunoglobulin (Ig) lambda variable 3-10, Ig mu heavy chain, 14-3-3 protein zeta/delta, heat shock cognate 71 kDa protein, and fibrinogen beta chain. Serum proteins include trypsin-3, mitochondrial ATP synthase complex subunit C1, type 2 cytoskeletal keratin, and polymeric Ig receptor. Dopamine beta-hydroxylase was most prominent in metastatic vs non-metastatic (primary, surveillance) in serum. Differential expression analysis of EV and serum revealed significant abundance of proteins related to metabolism, cellular organization and biogenesis, and cell proliferation in disease groups vs controls. Conclusion: This preliminary analysis demonstrates that blood-based proteomic analysis of various UM stages can be useful in providing diagnosis or prognosis of disease course for such patients. Further work with a larger sample size and repeated sampling at different disease progression stages is warranted to advance this work. Citation Format: Abarajithan Chandrasekaran, Nikhil Nayee, Jerome Lacombe, Timothy Karr, Frederic Zenhausern, Justin Moser. Blood-based proteomics analysis for uveal melanoma prognosis abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7695.
Chandrasekaran et al. (Fri,) studied this question.