Comparative analysis of urinary, bladder tissue, and stool microbiota in bladder cancer using shotgun metagenomics

Urine has been investigated as a less invasive biomarker source for bladder cancer (BCa) because of its direct contact with the bladder mucosa. Urine is assumed to reflect tissue microbiota, but most prior studies relied on 16S rRNA gene sequencing and contamination-prone voided urine. We aimed to determine whether catheterized urine microbiome profiles reflect bladder tissue microbial composition and metagenomically inferred functional potential, and to develop a model predicting tissue microbiota from urine. Matched bladder tissue ( n = 79), catheterized urine ( n = 79), and rectal-swab stool ( n = 17) samples were collected from patients with BCa under sterile conditions and analyzed by shotgun metagenomic sequencing to characterize microbial composition and predicted functional pathways. Patients were grouped by specimen availability: M-TUS (tissue, urine, and stool; n = 17) and M-TU (tissue and urine only; n = 62). Diversity, compositional similarity, and predicted functional pathway concordance were compared, and associations with sex or invasion status were tested by PERMANOVA with PERMDISP. Ridge regression with PCA-based dimensionality reduction predicted paired tissue relative-abundance profiles from urine at each taxonomic level, training on M-TU and testing on the independent M-TUS cohort. Tissue and urine microbiomes showed broadly similar richness and evenness and were more similar to each other by Bray-Curtis dissimilarity than either was to stool. Species-sharing and predicted functional pathway analyses further supported tissue–urine concordance and the biological distinctness of stool. Class- and genus-level profiles were broadly similar between sexes and showed minimal differences between invasion groups; the urine sex-associated PERMANOVA result was accompanied by significant dispersion heterogeneity and was interpreted cautiously. The Ridge regression model performed well on the held-out test set at the phylum (R² = 0.907) and class (R² = 0.834) levels, but performance declined at finer resolutions and did not outperform the null model at the family and genus levels. At broad taxonomic levels, catheterized urine closely reflects bladder tissue microbiota in BCa, supporting its use as a practical, less invasive proxy for tissue microbiome profiling and longitudinal monitoring. Fine-resolution prediction will require validation in larger, multi-institutional, multi-ethnic cohorts including healthy controls.