Urine cell-free DNA methylation-based deconvolution identifies tumor-specific cell types in localized urinary tract cancers

Background: Urine is a promising liquid biopsy source for non-invasive detection and monitoring of cancers of the urinary system, which currently lack screening tools for early diagnosis. Cell-free DNA methylation immunoprecipitation sequencing (cfMeDIP-seq) is a well-established tool to profile enriched 5-methylcytosine (5mC) signals in cell-free DNA (cfDNA). Recently, the decemedip computational framework was developed to enable cell-type deconvolution from cfMeDIP-seq data, providing a strategy to infer tumor- and immune-derived cfDNA contributions. Methods: We designed a customized urinary cancer cfMeDIP-seq deconvolution panel and integrated it with the decemedip framework to estimate the proportions of nine cell types, including immune subsets (B cells, CD4⁺ T cells, CD8⁺ T cells, monocytes, neutrophils, and natural killer cells) as well as tumor-specific signatures for prostate, kidney, and urothelial cancers. Urine cfDNA from 54 patients at Dana-Farber Cancer Institute was profiled by cfMeDIP-seq, comprising 19 localized bladder cancers (BLCA), 18 localized renal cell carcinomas (RCC), 9 localized upper tract urothelial carcinomas (UTUC), and 8 non-cancerous kidney disease cases. Deconvolution was applied to infer cell-type composition across these groups. Results: RCC samples exhibited significantly higher kidney cancer-derived cfDNA signal than BLCA and UTUC (p = 3e-06), while urothelial cancers showed significantly higher urothelial cancer-derived signal compared to RCC (p = 1e-05). The kidney-to-urothelial cancer cfDNA signal ratio distinguished RCC from urothelial cancers with an AUC of 0.984. In the bladder cancer cohort, five patients with low urothelial cancer signal had received neoadjuvant therapy, suggesting that deconvolution captures treatment-related changes. Excluding these cases further improved the discriminatory performance (AUC = 0.987). RCC patients also demonstrated significantly higher kidney cancer-derived signal than those with non-cancerous kidney disease (p = 0.03), supporting the utility of this approach for benign-malignant differentiation. Conclusions: Urine cfDNA deconvolution via cfMeDIP-seq enables detection of tumor-specific signals and profiling of immune cell composition across localized urinary cancers. This noninvasive liquid biopsy approach demonstrates potential for early detection, disease subtyping, minimal residual disease assessment, and treatment response monitoring and warrants validation in larger prospective cohorts.