Abstract 6243: Mathematical modeling of cell-free DNA fragmentation to map chromatin landscapes in health and cancer

Abstract Background: Cell-free DNA (cfDNA) fragments are shed into bodily fluids as a result of apoptosis and necrosis. Profiling plasma cfDNA has emerged as a powerful tool in cancer diagnostics and research, through circulating tumor DNA (ctDNA) fragments that enable genotyping of many tumors, and epigenetic analysis of cfDNA, including fragmentomics. Fragmentomics encompasses the analysis of fragment lengths, end motifs, and central dyads, which reflect epigenetic patterns, to identify the cell of origin for each fragment. We previously showed that fragment size diversity is strongly correlated with gene expression (Esfahani et al, NBT 2022). We hypothesize that parameters inferred from a mathematical model for the fragmentation process can enhance gene expression inference from cfDNA. Methods: We followed the mechanistic model previously proposed (Esfahani et al, NBT 2022), and modified it with additional steps. For each transcription start site (TSS), we input the mean and variance of the linker DNA length, fraction of nucleosomes sites occupied, cut frequency, and DNA degradation factor among other factors. Then, in accordance with the given parameters, we generate a sequence of nucleosome units, a sample of nucleosome sites occupied, and a sequence of fragments, and degrade them. For chromatin landscaping of plasma cfDNA, we used four cfDNA samples profiled by deep whole genome sequencing (∼250X). For each of 500 TSSs, we fitted our model to the fragment size distribution from data by minimizing the Jensen-Shannon (JS) divergence between the generated fragment size distribution and the fragment size distribution from the sample. Finally, we performed regression analysis on the fitted parameters to infer gene expression. Results: We analyzed cfDNA fragmentation spectra at TSSs genome-wide in relation to transcriptome-wide gene expression levels. The regression shows that gene expression has a negative correlation (P<0.001) between fraction of nucleosome sites occupied, a positive correlation (P=0.002) with the degradation factor, and a negative correlation (P=0.009) with the mean linker length. Conclusions: We show that cfDNA fragmentation can be modeled as a multi-step stochastic process. Through surveying TSSs, we showed significant correlation between certain model parameters and gene expression. Our findings that gene expression is negatively correlated with model parameters for fraction of nucleosome sites occupied and mean linker length is consistent with previous literature suggesting that higher fraction of nucleosome sites occupied and longer linker length lead to suppressed gene expression (Valouev et al, Nature 2011). We envision this model-based approach will enhance gene expression inference that can ultimately improve cancer detection and tissue of origin classification. Citation Format: Brian Sun Liu, Mengran Zhang, Takeshi Sugio, Mohammad Shahrokh Esfahani. Mathematical modeling of cell-free DNA fragmentation to map chromatin landscapes in health and cancer [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 6243.