Background
The application of next generation sequencing techniques has enabled characterization of urinary tract microbiome. Although many studies have demonstrated associations between the human microbiome and bladder cancer, they have not always reported consistent results, thereby necessitating cross-study comparisons. Thus, the fundamental questions remain how we can utilize this knowledge. The aim of our study was to examine for disease-associated changes in urine microbiome communities globally utilizing machine learning algorithm. The results were further validated using our own prospectively collected urine of bladder cancer patients.
Results
Our study included 129 bladder cancer urine samples, and 60 healthy controls across four different countries. At a meta-analysis false discovery rate (FDR) of 0.01, we identified a total of 97/548 genera to be differentially abundant in the BCa microbiome compared to healthy patients. Overall, while the differences in diversity metrics were clustered around the country of origin (Kruskal Wallis, P < 0.001), collection methodology was a driver of microbiome composition. When assessing dataset from the China, Hungary and Croatia, ML data demonstrated no discrimination capacity to distinguish between BCa and healthy (AUC 0.577). However, inclusion of samples with catheterized urine only improved the diagnostic accuracy of prediction for BCa to AUC 0.995, with precision recall AUC = 0.994. Through elimination of contaminants associated with collection methodology among all cohorts, our study identified increased abundance of polycyclic aromatic hydrocarbon (PAHs) degrading bacteria Sphingomonas, Acinetobacter, Micrococcus, and Ralstonia consistently present in BCa patients.
Conclusions
The microbiota of the bladder cancer population may be a reflection of PAH exposure from smoking, environmental pollutants and ingestion. Presence of PAHs in urine of bladder cancer patients may allow for a unique metabolic niche, and provide necessary metabolic resources where other bacteria are not able to flourish. Furthermore, we found that while compositional differences associated with geography more than disease, many are driven by collection methodology.