Inferring tissue origin from cell-free DNA in metastatic cancer

A major clinical challenge in metastatic cancer management is determining not only the presence of tumor-derived cell-free DNA (cfDNA) but also the tissue lineage and organ context reflected by circulating cfDNA signals. CfDNA supports such inference by retaining molecular signatures specific to its tissue-of-origin, including DNA methylation, 5-hydroxymethylcytosine (5hmC), fragmentomic profiles, nucleosome positioning patterns, and histone modifications. This review examines the biological principles underlying cfDNA-based tissue-of-origin determination, the computational frameworks used to interpret these signals, and the key barriers to clinical translation. We additionally compare representative deconvolution and classification strategies, including linear and probabilistic models, machine/deep learning, and single-cell-informed reference approaches. Despite substantial progress, clinical integration remains hindered by low tumor fractions, heterogeneous contributions from multiple tissues, background noise from nonmalignant cell death, and a lack of robust prospective validation. We further emphasize that source estimates in metastatic plasma should be interpreted in the context of mixed tissue contributions, differential shedding, reference design, and model uncertainty, rather than as direct measures of anatomical metastatic burden.