Abstract Background: Combination immune checkpoint inhibitor (ICI) therapy represents a standard-of-care for patients with metastatic melanoma. However, 40% of patients do not respond to treatment and severe immune-related toxicity affects up to 60% of patients. Accordingly, we need more precise ways to select patients for combination ICIs. To address this, we developed LiquidTME, a liquid biopsy method based on Spatial EcoTyper (Cancer Res (2025) 85 (8Supplement₁): 153) that leverages a deep learning approach to noninvasively assess the tumor microenvironment from cell-free DNA (cfDNA) methylation data. LiquidTME was previously trained to predict ICI response using 78 patients with advanced melanoma treated at Yale University. Here we performed a blinded validation of LiquidTME in coordination with clinicians at Washington University (WashU). Methods: The WashU cohort consisted of pre-ICI plasma from 34 patients with advanced melanoma, each treated with ipilimumab and nivolumab (n=27) or relatlimab and nivolumab (n=7). ICI response was classified as durable clinical benefit (DCB) or no durable benefit (NDB) by a board-certified oncologist. Tumor mutational burden (TMB) was determined for 27 patients using commercial CLIA assays and analyzed as nonsynonymous mutations (mt) per megabase (Mb), using the FDA-approved 10 mt/Mb cutpoint for TMB-high and -low groups. The WashU cohort was sent to a scientific team that was blinded to all clinical and TMB data. Cell-free DNA was extracted from 2 mL plasma per patient and subjected to enzymatic methyl-seq (EM-seq) at a median depth of 15x. LiquidTME was performed on the resulting profiles, yielding a binary response prediction and a continuous response score for each sample. Test results were locked down and returned. Both assays were compared by AUC and two-sided Wilcoxon rank-sum tests for response classification, and Kaplan-Meier analysis for progression-free survival (PFS). Results: Median follow-up time of the cohort was 21. 6 months. LiquidTME performed on cycle 1 day 1 pre-ICI plasma significantly stratified PFS with a hazard ratio (HR) of 0. 27 (P = 0. 005), with LiquidTME (+) patients achieving a median PFS of 1. 7 years longer than LiquidTME (-) patients. LiquidTME distinguished DCB from NDB with an AUC of 0. 77 (P = 0. 007). Similar performance was seen across distinct combination ICI regimens. In patients with TMB data, LiquidTME maintained whole-cohort performance with a PFS HR of 0. 30 (P 0. 03; AUC = 0. 77). In contrast, TMB high vs. low subsets failed to stratify PFS (HR = 0. 49; P = 0. 17) or response (AUC = 0. 53; P = 0. 8). Conclusion: LiquidTME identified durable responders to combination ICI from pretreatment plasma in this blinded validation cohort of melanoma patients. Given its favorable performance, LiquidTME shows promise as a clinical tool to guide personalized immunotherapy decision-making. Citation Format: Aadel A. Chaudhuri, David Y. Chen, Tucker Hansen, Mirna Jarosz, Vincent A. Miller, Aaron M. Newman. Blinded clinical validation of LiquidTME, a cell-free DNA assay for predicting response to immunotherapy by noninvasively profiling the tumor microenvironment 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 1040.
Chaudhuri. et al. (Fri,) studied this question.