Abstract 3920: Precise solid tumor classification through unbiased quantification of proteoforms deep into tissue leakage

Blood is the most frequently used sample in diagnostic processes, and its liquid component is known as plasma. Plasma potentially contains molecular clues coming from all the tissues that make up an organism, thus possibly enable the holistic analysis of the health state of an individual. The major challenge in plasma protein analysis is the 1013- fold dynamic range and the strong prevalence of few, very abundant proteins. Depletion of the most abundant proteins is used to increase coverage of the plasma proteome. To be able to profile human plasma at large scale and at maximal depth, we developed and optimized two main aspects. Firstly, we automated the depletion of the top 14 most abundant proteins in human plasma to a throughput of 40 depletions per day per setup. A comparison of native and depletion workflows with a controlled quantitative experiment demonstrated that depletion led to a significant increase in the number of identifications (&gt200%) and number of true hits (&gt300% at controlled actual FDR &lt1%), while conserving quantitative accuracy. Secondly, we further optimized FAIMS-DIA methods for deep plasma proteome profiling and applied the optimized workflow to a cohort coming from the deadliest five solid tumor types. To research large cohorts’ feasibility, we analyzed a cohort comprising 30 each of healthy, lung, colorectal, pancreatic, breast and prostate cancer. Altogether, we processed 180 samples (plus 24 quality controls) and quantified 2,732 proteins, of which 1,804 in at least 50% of the runs. Based on quality control samples, we could characterize variance introduced on each level, all much lower than the inter-individual variability. We reached 8 orders of magnitude of dynamic range, within this range we covered extensively tissue leakage proteome, interleukins and signaling proteins such as Egf, Klk3 (PSA), Akt1, Cd86, Met, Erbb2 and Cd33. Using machine learning we reduced to an average of 129 (5% of the quantified proteins) biomarker candidates per cancer type, making biological interpretation more feasible while enabling classification of health and disease. The model performance was 86-100% on the 20% hold-out validation set when healthy and overall disease status were considered. Importantly, the biomarker candidates were predominantly (70%) coming from low abundance regions clearly demonstrating the need to measure deeply because they would be missed by shallow plasma profiling. Hence, we showed that our automated plasma depletion workflow has the potential to enable the unbiased and reproducible quantification of more than 2,700 proteins across very large cohorts and unlock biomarker candidate panels in different cancer types. Furthermore, with the latest analytical method iteration applied to a subset of the samples, we could quantify more than 3,500 proteins, demonstrating the full potential of the workflow. Citation Format: Marco Tognetti, Kamil Sklodowski, Sebastian Mueller, Dominique Kamber, Jan Muntel, Yuehan Feng, Roland Bruderer, Lukas Reiter. Precise solid tumor classification through unbiased quantification of proteoforms deep into tissue leakage [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3920.