Abstract 1951: Inferring immune and tissue cell type contributions to cell-free DNA (cfDNA) with a DNA methylation assay

Background: In tissue-localized disease, dying cells release DNA into the bloodstream, enabling disease detection and characterization via blood draws. Circulating tumor DNA (ctDNA) is a key biomarker for cancer screening and monitoring. Quantification of immune cell- and tissue-derived DNA in plasma can expand the range of conditions identifiable through liquid biopsies. Here, we demonstrate the ability to infer cell type contributions to plasma cell-free DNA (cfDNA) from a clinically validated differential methylation-based cfDNA assay. Methods: We first implemented and validated a non-negative least squares cell type deconvolution method that uses single-site methylation (SSM) data. Accuracy was assessed by comparisons to gold-standard CyTOF and RNA-seq-based deconvolution on blood samples, as well as in silico simulations in blood and plasma. Next, we developed a model to estimate immune and tissue cell types contributing to plasma based on data derived from a clinically validated differential methylation-based cfDNA assay. Using molecule counts from >2700 regions, we trained a model to optimize alignment with our validated SSM deconvolution method run on the same samples. Training and evaluation was done using leave-one-out cross validation. Results: Within the cellular blood compartment, SSM deconvolution accurately estimated T cell, B cell, and neutrophil proportions compared to CyTOF (r ≥ 0.82; N=13) and RNA-seq-based deconvolution (r ≥ 0.78; N=70). Within the plasma compartment, we used SSM deconvolution to infer, in 106 healthy plasma samples, the frequencies of cell types commonly contributing to cfDNA: T cells, B cells, neutrophils, monocytes, megakaryocytes, erythroid cells, hepatocytes, and endothelial cells. In silico simulations showed the limit of detection (LOD90) for these cell types in plasma to be <=1.1% (median=0.28%). Inferred abundances aligned with distributions in published literature (Loyfer et al., Nature 2023). We compared the hypermethylation assay-derived cell type estimates to our validated SSM deconvolution approach on 106 healthy plasma samples run on both assays. We observed Pearson correlations of 0.66-0.98 for all evaluated cell types except for monocytes (r = 0.3; P<0.003), indicating accurate estimation of the main cell types contributing to cfDNA in plasma. Conclusions: We developed an algorithm using a clinically validated cfDNA assay to accurately infer immune and tissue cell type contributions to plasma cfDNA. This technology has the potential to improve disease detection and management by identifying biological processes involving immune and/or tissue shedding into plasma. Further development of this technology will include application and evaluation on additional sample cohorts, including various disease states. Citation Format: Michael Gloudemans, Elena Zotenko, Tam Banh, Kimberly Zhu, Gleb Martovetsky, Sara Wienke, Min Woo Sun, Alan Selewa, Samantha Liang, John Connolly, Stefanie Mortimer, Noam Vardi, Emily Tsang, Drew Kennedy, Meromit Singer. Inferring immune and tissue cell type contributions to cell-free DNA (cfDNA) with a DNA methylation assay [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1951.