EpiGuide：通过追踪循环肿瘤DNA中的表观遗传可塑性来监测肿瘤进展

Cell state transitions such as epithelial-to-mesenchymal plasticity (EMP) play an important role in the progression of triple-negative breast cancer (TNBC) and are strongly associated with therapy resistance. These transitions are hypothesized to be largely regulated by DNA methylation (DNAm) changes, which are retained in tumor-derived plasma cell-free DNA (cfDNA). Therefore, detecting EMP-related DNAm signatures in cfDNA offers a minimally invasive strategy to monitor tumor dynamics in liquid biopsies. However, estimating the tumoral cfDNA EMP fractions is challenging because liquid biopsy samples contain a mixture of cfDNA originating from both cancer and healthy cells. As a solution, DNAm deconvolution algorithms can be adopted to estimate tumoral cfDNA fractions in liquid biopsy samples. In the last decade a high number of DNAm deconvolution tools have been developed. However, despite the need to identify the most effective deconvolution tools for tumor fraction estimation, no benchmarking study has specifically focused on this task. Therefore, we developed DecoNFlow, an automated Nextflow pipeline including 12 DNAm deconvolution tools and 3 differential methylation analysis tools. This is the most comprehensive pipeline for DNAm deconvolution to date, which allowed us to perform a benchmarking of 12 deconvolution tools using 3.5K in silico mixtures spanning multiple tumor types, sequencing depths, marker-selection strategies and profiling technologies (paper in review). We show that CelFiE is the overall top-performing tool across multiple evaluation criteria. In a subsequent proof-of-concept study, we assessed EMP monitoring in vivo using the MMTV-PyMT mouse model, which develops TNBC-like tumors which spontaneously undergo EMP. First, several cell lines have been derived from primary tumors of this model and characterized. These cell lines exhibited distinct EMP states (epithelial or mesenchymal) or stably co-existing EMP states. Following orthotopic injection of these cell lines into mice, we performed methylation profiling on both tumors and plasma cfDNA. Tumor EMP state fractions were estimated using CelFiE and a DNAm EMP atlas of MMTV-PyMT tumors as reference, consisting of both single-cell and bulk EMP DNAm markers. We demonstrated that EMP states can be differentially detected in cfDNA, and that mice injected with mixed cell lines show significantly higher tumoral cfDNA fractions than those injected with epithelial-only lines, suggesting that coexistence of multiple EMP states may promote higher tumor burden. In summary, this study presents a robust analytical and computational pipeline that allows to monitor EMP in a minimally invasive way through cfDNA DNAm analysis. In future, clinical application of this approach is expected to be instrumental for timely identification of cancer patients at risk for therapy resistance through epigenetic plasticity.