Patient characteristics description

The study involved 87 participants, including 29 male patients with BPH and 58 BCa patients. The benign group consisted exclusively of male patients. In the BCa cohort, 29 patients provided paired pre- and post-BCG samples, resulting in 29 matched sets. The other 29 BCa patients contributed unpaired post-BCG samples. Table 1 details the clinical characteristics of these patients. The average age in the benign group was 73.0 years, which was not significantly different from the BCa group's average age of 72.6 years (p=0.6373). However, there was a notable difference in gender composition, with the BCa group comprising 79.3% male patients (p=0.0070). In terms of antibiotic usage within a month prior to sample collection, 20.7% of the benign group and 10.3% of the BCa group reported usage, with no significant difference observed (p=0.6373). The prevalence of smoking was identical in both groups at 48.3%, showing no statistical disparity (p>0.9999). In addition, none of the patients in either group had a urinary tract infection within a month of sample collection.

Within the BCa group, clinical follow-up averaged 668 days, with a high-grade disease prevalence of 91.4%. The response rate to intravesical BCG treatment was 67.2%, and the recurrence rate stood at 29.3%. The median RFS time was 509.5 days. Additionally, the progression rate was 5.2%, with a median PFS time of 650 days.

Compared to benign group, BCa group has a lower toluene degradation capacity

According to the Chao1 index, there was no significant difference in microbial richness between the benign and pre-BCG groups; however, a significant difference was observed between the benign and post-BCG groups (p<0.0001), as well as between the pre-BCG and post-BCG groups (p<0.05; Figure 1B). The Simpson index indicated a statistically significant difference only between the benign and pre-BCG groups (p<0.05; Figure 1B). Beta diversity analysis based on Bray-Curtis dissimilarity also revealed a considerable overlap in microbial distribution between the benign and pre-BCG groups, distinct from the post-BCG group (p=0.001, R²=0.159, Figure 1C). Upon synthesizing the analyses on diversity, it was evident that the benign group and the pre-BCG group displayed some differences yet shared similarities. This distinction was further validated at the species level, as depicted in Figure 1D. The benign group showed a more consistent species-level composition within its samples compared to the greater dissimilarity observed among BCa samples (Figure 1D). For a detailed comparison between groups, the pre-BCG group was considered as the BCa group. This decision was guided by the logic that a comparison between the benign group and the BCa group should be made prior to BCG treatment to exclude the treatment's impact. The heatmaps reveal divergent microbial composition patterns between the groups at both genus and species levels, as illustrated in Figures 1E and S1A. Specifically, the genera Anaeroplasma, Tetragenococcus, Thauera, and Sporotomaculum were significantly enriched in the benign group (Figure 1E). Similarly, Treponema sp5, Desulfovibrio mexicanus, and Lactobacillus agilis showed significant enrichment in the benign group (Figure S1A).

In addition to microbial composition, our study also explored the enrichment of microbial metabolic pathways within each group. Figure 1F presents a volcano plot that highlights the significant enrichment of specific pathways, including PWY-6145 (Superpathway of CMP-sialic acids biosynthesis), PWY-5789 (3-hydroxypropanoate/4-hydroxybutanate cycle), and PWY-5184 (Toluene degradation VI (anaerobic)), in the benign group. Among them, we focused on the toluene degradation pathway because toluene is well-known for its carcinogenic properties [35]. As the genus Thauera and Desulfovibrio mexicanus, both known for their toluene-degrading capabilities [36], were found to be enriched in the benign group, we deduced that the toluene degradation pathway may play a significant role in preventing bladder carcinogenesis. Toluene degradation pathway is schematically depicted in Figure S1B.

To establish the relevance of toluene to bladder carcinogenesis, we utilized an open dataset [30]. From this dataset, we identified metabolites that were significantly enriched in both cancer patients and controls (Table S1). We then performed ORA using these selected metabolites. The ORA revealed that the metabolites enriched in the BCa group were closely associated with the category of organic acids and derivatives (p<0.001, FDR=0.012; Figure S1C). Metabolites enriched in the control group were closely related to organic acids and derivatives (p<0.001, FDR<0.001), benzenoids (p<0.001, FDR<0.001), organoheterocyclic compounds (p<0.001, FDR=0.002; Figure S1D), and exposure to volatile organic compounds (VOCs) (p=0.004, FDR=0.221; Figure S1E). Furthermore, hippuric acid—a degradation product of toluene in the human body [37] — was found at significantly increased levels in urine from benign control patients in this dataset and other studies [38], reinforcing our hypothesis that an impairment in toluene degradation is linked to bladder carcinogenesis.

When comparing male BCa patients with male benign controls, more definitive results were obtained (Figure S2A). A significant difference in the Chao1 index was observed between the benign group and both the pre-BCG and post-BCG groups (p<0.05 and p<0.0001, respectively; Figure S2B). Additionally, there was a statistically significant difference between the pre-BCG and post-BCG groups (p<0.05, Figure S2B). For the Simpson index, a significant difference was noted only between the benign group and the pre-BCG group (p<0.05; Figure S2B). Beta diversity also varied between the groups (p=0.001; Figure S2C). Differential abundance testing revealed distinct microbial composition patterns between the benign and BCa groups, as illustrated in Figures S2D and S2E. Microbial metabolic pathways such as PWY-6145, PWY-5789, and PWY-5184 (Toluene degradation VI) were found to be enriched in the benign group, mirroring the patterns observed in the BCa group that included female patients, as shown in Figure 1F (Figure S2F).

Responders exhibit changes in urinary microbiome and metabolite production after intravesical BCG treatment

To uncover the characteristics of patients who responded to BCG therapy, we conducted comparative analyses between matched pre-BCG and post-BCG samples, with each group comprising 29 patients (Figure 2A). Among these patients, 23 were identified as responders, while 6 were non-responders. Figures S3A and S3B demonstrate no significant differences in diversities between the paired pre-BCG and post-BCG groups in the ‘Paired cohort,’ which includes both responders and non-responders. Figure 2B compares the microbial composition at the species level based on the response to BCG treatment and the timing of sample collection relative to BCG therapy in the paired cohort. Our analysis revealed that while pre-BCG and post-BCG samples from non-responders were almost identical, those from responders displayed significant differences. Following, we proceeded with a more detailed comparison at both the genus and species levels. This revealed variations in microbial composition associated with the response to BCG (Figures 2C and 2D; Figures S3C and S3D). In responders, the post-BCG group had more Klebsiella oxytoca, Morganella morganii, and Salmonella enterica, but fewer Anoxybacillus kestanbolensis and Bacillus flexus (Figure 2D). In the paired cohort, the post-BCG group exhibited an increased presence of Klebsiella oxytoca, Morganella morganii, Salmonella enterica, and Trabulsiella farmeri compared to the pre-BCG group (Figure S3D). However, non-responders did not exhibit a significant difference in microbial composition between the pre-BCG and post-BCG groups. We concluded that BCG treatment primarily induces compositional changes in the urinary microbiome, observed exclusively in responders and not in non-responders.

Next, we compared metabolic pathways between the groups (Figure 2E; Figure S3E). The post-BCG group in responders showed increased activity in pathways PWY-7002 and PWY-4361 (Figure 2E). The S-methyl-5-thio-α-D-ribose 1-phosphate degradation I pathway, also known as PWY-4361, was significantly enriched exclusively in responders (Figures 2E; Figure S3E). This pathway degrades S-methyl-5-thio-α-D-ribose 1-phosphate into 2-oxoglutarate and L-methionine, both of which are known for their anticancer properties [39-44]. PWY-4361 is schematically depicted in Figure 2F.

Lastly, we identified key driver microbes, revealing significant bacterial shifts from the pre-BCG to post-BCG microbiome. Figures 2G and S3F display the common sub-network at the genus level in the responder and paired cohorts, respectively. Enlarged red nodes highlight important driver microbes that transition from pre-BCG to post-BCG status. Tindallia_Anoxynatronum, Lutispora, Marinococcus, and Parvimonas were identified as critical driver microbes in responders with the highest NESH scores, underscoring their significant role in this transformation (Figure 2H).

Responders exhibit enriched quinolone biosynthesis post-BCG treatment

Next, we performed subgroup analyses for both the pre-BCG and post-BCG groups. Initially, we evaluated 29 pre-BCG samples, comprising 23 responders and 6 non-responders, and compared them based on factors such as gender, grade, and response to BCG (Figure 3A). No significant difference was observed between responders and non-responders in terms of antibiotic use within the month prior to sample collection (p>0.9999) or smoking history (p=0.6693). There were no cases of urinary tract infection within a month before sample collection. While non-responders exhibited a higher age range, this difference was not statistically significant (p=0.0684). However, they did show significantly higher recurrence rates (p<0.0001; Table S2). Since there were no cases of disease progression in the pre-BCG group, the analysis of recurrence essentially mirrored that of the response to BCG treatment. Alpha and beta diversity assessments revealed no significant differences across these variables. Differential abundance testing, however, identified gender-based variations in microbial composition (Figure S4A) and differences associated with recurrence (Figure S4B) and BCG response (Figure 3C). Notably, responders exhibited higher abundances of Campylobacter ureolyticus and Bifidobacterium bifidum species compared to non-responders (Figure 3C; Table S3). Furthermore, survival analyses focusing on the presence of C. ureolyticus and B. bifidum revealed no significant extension in RFS (p=0.5536 and 0.2915, respectively; Figures 3E and 3F). As previously noted, the Bifidobacterium genus is renowned for its health benefits, prompting a further examination of its abundance. However, our analysis indicated that the abundance of the Bifidobacterium genus did not differ significantly based on treatment response (p=0.8837; Figure 3D). No significant differences were detected in metabolic pathways when categorized by BCG response. In summary, the comparison of pre-BCG samples revealed only minor differences in microbial composition and no significant changes were observed in metabolic pathways.

The analysis of post-BCG samples included 58 specimens from 39 responders and 19 non-responders (Figure 3A). No significant differences were observed between responders and non-responders in antibiotic use within the month prior to sample collection (p=0.2475), smoking history (p=0.5827), or age (p=0.1832). No cases of urinary tract infection were reported within a month before sample collection. However, non-responders exhibited significantly higher rates of recurrence and progression (p<0.0001 and 0.0314, respectively; Table S4). We compared these groups based on gender (Figures S5A and S5B), grade (Figures S5C and S5D), response (Figures 3B, 3G~3K, and 3M), recurrence (Figures S5E and S5F), and progression (Figure S6). Although no significant differences were observed in diversities, there were distinct differences in microbial composition and metabolic pathways based on these parameters. The heatmap revealed distinct microbial composition patterns at the family and genus levels according to the response to BCG (Figure 3B). Significantly, responders exhibited a prevalent presence of the Bifidobacterium genus, previously mentioned as beneficial bacteria (Table S5; Figure 3H). Within this genus, various species including B. adolescentis, B. breve, and B. longum, were notably more prevalent among responders (Table S6; Figure 3G). Specifically, B. breve and B. longum were associated with prolonged RFS (p=0.0203 and 0.0423, respectively; Figures 3I and 3J), but Bifidobacterium genus and B. adolescentis were not (p=0.4816 and 0.3012, respectively; Figures S4C and S4D). Cox regression analysis of post-BCG urine samples identified the presence of B. breve as an independent significant factor influencing RFS prolongation (p=0.044; Table 2).

The analysis of metabolic pathway abundances showed an increased presence of PWY-6660 and PWY-6662 pathways in responders, both linked to quinolone biosynthesis (Figure 3K). When comparing enzyme abundance, higher levels of quinolone synthases, such as 2-heptyl-3-hydroxy-4(1H)-quinolone synthase and 2-heptyl-4(1H)-quinolone synthase, were observed in responders compared to non-responders (Figure 3M). These enzymes contribute to the production of 2-heptyl-3-hydroxyl-4-quinolone, a variant of quinolone (Figure 3L).

Development of Predictive Models for Malignancy and Response to BCG Treatment

To develop a predictive model for malignancy and response to BCG treatment, we incorporated genera and species identified in differential abundance tests. Initially, we utilized data on the presence and relative abundance of genus Thauera and Desulfovibrio mexicanus. These microbes, identified to play a key role in the toluene degradation pathway [36], were suggested to be associated with the prevention of bladder carcinogenesis. This model exhibited excellent predictive accuracy for malignancy, achieving an AUC of 0.913 (Figure S7A).

We then aimed to develop a model predicting the response to BCG treatment. Initially, we included data on the presence and abundance of C. ureolyticus and B. bifidum from pre-BCG urine samples (N=29), resulting in a model with good predictive performance (AUC of 0.717; Figure S7B). Subsequently, we incorporated data on the presence and abundance of the Bifidobacterium genus and various species including B. adolescentis, B. breve, and B. longum from post-BCG urine samples (N=58). This model demonstrated good predictive accuracy for BCG treatment response, as evidenced by an AUC of 0.753 (Figure S7C).