Role of Gut Microbiome in Neoadjuvant Chemotherapy Response in Urothelial Carcinoma: A Multi-institutional Prospective Cohort Evaluation

Treatment with neoadjuvant chemotherapy (NAC) in muscle-invasive bladder cancer (MIBC) is associated with clinical benefit in urothelial carcinoma. Cisplatin-based NAC prior to extirpative surgery confers a 6%–8% overall survival (OS) benefit compared with surgery alone (1). Historically, the two most commonly utilized regimens include dose-dense methotrexate, vinblastine, doxorubicin, and cisplatin (dd-MVAC) or gemcitabine and cisplatin (GC), with pathologic complete response (pT0) after NAC at the time of surgery being widely adopted as a surrogate endpoint for improved OS (2). Conversely, patients with residual disease or nodal involvement are at a high risk of recurrence and progression of disease.

While extensive research evaluating role of tumor mutational expression profiles and clinicopathologic factors into chemotherapy response has been published, the role of gut microbiome (GM) in bladder cancer in chemotherapy response has not been thoroughly evaluated. Indeed, the GM is increasingly seen as important and modifiable aspect of anticancer therapeutic response (3, 4). We hypothesized that GM composition associates with response to therapy in bladder cancer. To test this hypothesis, we prospectively collected fecal microbiome samples from patients with MIBC starting treatment with NAC across two American institutions. In addition, samples from partners living with patients during NAC were collected to identify changes caused by chemotherapy versus confounding environmental effects.

Furthermore, to investigate the change in GM during carcinogenesis using the N-butyl-N-(4-hydroxybutyl)-nitrosamine (BBN) murine model, stool pellets were collected for GM profiling throughout the BBN exposure period and at the time of sacrifice. This research provides the largest assessment of GM and its association with response to NAC and facilitates specific microbial special investigation and functions associated with response.

The study was conducted in accordance with recognized ethical guidelines and approved by Case Western Reserve/University Hospitals Cleveland Medical Center (CW) and Fox Chase Cancer Center (FCCC) under the Institutional Review Board (IRB)# STUDY20200350 and IRB #18-4001, respectively. Patients with MIBC undergoing cystectomy were enrolled prospectively across both institutions, written informed consent was obtained.

Stool samples were collected prior to cystectomy between July, 2018 until March, 2023. Patients were excluded from participation if they have any antibiotics within 6 weeks of collection or history of Clostridium difficile infection and treatment within 2 months. All patients were > 6 weeks out from last cystoscopy requiring antibiotics or transurethral resection of bladder tumor. Presurgical stool samples were collected at the time of cystectomy prior to initiation of antibiotics with rectal swabs, and all collections were performed prior to the use of preoperative antibiotics for surgical preparation (Supplementary Fig. S1). Sample sizes were determined on the basis of availability of eligible participants and the feasibility of sample collection within the study timeline.

A total of 142 patients with bladder cancer were recruited for the study after meeting inclusion criteria (Supplementary Fig. S2). The swabs were immediately placed into 1.5 mL microcentrifuge tubes containing 1 mL of PBS. Swabs were stored at 0°C during transport to the laboratory for processing within 2 hours of collection. Samples were then resuspended and stored in sterile PBS at −80°C until analysis. Partners and changes during chemotherapy: stool samples were collected prior to NAC (dd-MVAC), after cycle 2 (of 3), and prior to cystectomy. Stool samples from partners were collected at the same time and stored. Samples were stored at room temperature for <24 hours prior to being split into aliquots and stored at 80°C until use. All patients and partners were provided an OMNIgene GUT kit (OMR-200; DNA Genotek) for outpatient fecal sample collection, which maintains DNA stability at room temperate for up to 60 days. The final analysis consisted of 190 patients (142 bladder cancer, 48 Controls) (Supplementary Appendix; Supplementary Table S1).

Specific pathogen-free C57BL/6 female and male mice were purchased from Taconic Laboratory. The animal experiments were carried out following FCCC Institutional Animal Care and Use Committee guideline and approved (IACUC 19-03). BBN was purchased as a single batch from TCI America (Batch ODW3F-FH). BBN was administered at a concentration of 0.05% in water given ad libitum and replenished weekly to half of the cages, with the other half receiving regular drinking water from the same source. Half of the mice in each treatment arm were male. Mice began BBN administration via drinking water at 8–10 weeks of age and continued for 12 weeks. After 12 weeks of BBN exposure, mice were given regular drinking water until the end of the study. Presence of tumors was evaluated via excretory µCT urography following retroorbital injection of dilute Visipaque contrast (Supplementary Fig. S3). Stool pellets were collected at pretreatment, 6 weeks, 12 weeks, and between 16 to 22 weeks (when tumors were visible as filling defects on µCT scan). A total of 23 mice were exposed to BBN (14 males, 9 females), with 18 water only controls (10 males, 8 females; Supplementary Fig. S3). When sacrificing mice, one mouse from each cage was sacrificed at each timepoint rather than sacrificing all mice from one cage to avoid confounding factors that might be present in one cage and not another. Stool pellets were collected from each mouse if it spontaneously defecated or with gentle abdominal pressure. Pellets were resuspended in PBS and stored −80 °C until use.

Fecal samples were stored in 250 µL aliquots at −80°C for DNA extraction. DNA was extracted from the thawed fecal samples using the DNeasy blood and tissue kit (Qiagen) with minor modification. IBriefly, to 250 µL fecal sample, 180 µL Buffer ATL was added and mixed carefully. To this mixture, 59 µL of enzyme cocktail containing 50 µL lysozyme (10 mg/mL, Sigma-Aldrich), 6 µL mutanolysin (25 KU/mL, Sigma-Aldrich), and 3 µL lysostaphin (4,000 U/mL, Sigma-Aldrich) was mixed and the resuspended sample was transferred in to tubes containing 0.4 g sterile zirconia beads. The samples were homogenized in a Mini-BeadBeater at maximum speed for 1 minute and then incubated at 37°C for 30 minutes. A total of 20 µL of proteinase K was added and thoroughly mixed followed by incubation at 56°C for 1–2 hours to allow complete lysis. The samples were then centrifuged at maximum speed for 5 minutes to pellet debri. The supernatant was then used for DNA extraction using the DNeasy blood and tissue kit according to manufacturer's guidelines.

16S rRNA gene sequencing methods were adapted from the methods developed for the NIH-Human Microbiome Project (5). Filtered sequences with >97% identity were clustered into operational taxonomic units (OTU) and classified at the genus against the SILVA 16S ribosomal RNA sequence database (release 138.1; ref. 6). The relative abundance of each OTU was determined for all samples. A step-by-step description of our analysis pipeline has previously been published and is available (7). QIIME 2 platform was utilized for processing and visualizing microbiome data (8). All processed data are available as Supplementary Index S1.

Complete response (CR) was defined as pT0N0M0 pathologic assessment, partial response (PR) as (pTis, Ta, T1N0M0) and all other stages defined as CR. Response data were combined for CR and PR into single category to denote chemotherapy response with downstaging of the disease. Differentially abundant OTUs were identified using LEfSe [linear discriminant analysis (LDA) effect size] for each pairwise comparison of clinical groups [healthy vs. bladder cancer, bladder cancer vs. partner, CR vs. no response (NR)].

LEfSe first uses a nonparametric factorial Kruskal–Wallis rank-sum test to identify differentially abundant OTUs (9). This is followed by a set of pairwise tests among clinical groups to ensure biologic consistency using the Wilcoxon rank-sum test. LDA is then used to estimate the effect size of each differentially abundant OTU. LEfSe statistics were ranked to identify the greatest differences in microbial relative abundance across patient groups mentioned above. Associations between individual taxa and clinical variables were examined using generalized linear models as implemented in the Microbiome Multivariable with Linear Models (MaAsLin2) package (10).

To create a genus-level OTU table, OTUs with the same genus name were merged into one genus. We calculated the relative abundance of each genus. Samples within patients with bladder cancer were grouped into two clusters based on beta diversity using k-means clustering (11). The number of clusters (k = 2) was determined heuristically.

Rectal swab samples with ample stool and peripheral blood collected in Ethylenediaminetetraacetic acid (EDTA) tubes prior to cystectomy and administration of antibiotics was analyzed for short chain fatty acid (SCFA) composition in blood and stool was performed by Microbiome Insights. Briefly, SCFA extraction was replicated on the basis of that of Zhao and colleagues (12). Material were resuspended in MilliQ-grade H₂O, and homogenized using MP Bio FastPrep, for 1 minute at 4.0 m/s. A total of 5 mol/L HCl was added to acidify fecal suspensions to a final pH of 2.0. Acidified fecal suspensions were incubated, and centrifuged at 10,000 RPM to separate the supernatant. Fecal supernatants were spiked with 2-Ethylbutyric acid for a final concentration of 1 mmol/L.

Extracted SCFA supernatants were stored in 2-mL GC vials, with glass inserts. SCFA were detected using gas chromatography (Thermo Fisher Scientific Trace 1310), coupled to a flame ionization detector (Thermo Fisher Scientific). Our SCFA column is “Thermo TG-WAXMS A GC Column, 30 m, 0.32 mm, 0.25 um,” which is very similar to the instrument and method we have taken from the literature.

Additional SCFA methodology can be found in Supplementary Methodology.

Alpha diversity comparison between bladder cancer versus healthy, CR versus NR, bladder cancer versus partner and enterotype were compared using Wilcoxon rank-sum or Mann–Whitney test (used for comparison between binary variables) and Spearman rank was used to compare continuous variables. Fisher exact test was used when proportions were compared between binary variables. Adjustments for multiple comparisons were done using the FDR method at an α level of 0.05. Hypothesis testing was done using both one-sided and two-sided tests as appropriate at a 95% significance level. Correlation analysis to investigate the relationship between SCFA and bacterial taxa and clinical response was assessed via Spearman correlation coefficient. All analyses were conducted in R (13) and Python (14). All visualization was performed via R, Python, with CorelDraw.

Random forest prediction for response-prediction classifier was utilized to test prediction performance of identified microbiome genera (15). AUC was used to assess the performance of the classifier for CR.

Raw data can be provided upon request from the corresponding author. Analyzed data can be found in Supplementary Appendix.

We first analyzed the GM differences between all available patients with bladder cancer (n = 142) and controls (CTR; n = 48). We observed a higher prevalence of Prevotella (4.35% vs. 0.17%, P = 0.001) and Porphyromonas (1.37% vs. 0.02%, P = 0.001) in patients with bladder cancer, along with a reduced abundance of Faecalibacterium (4.66% vs. 6.19%, P = 0.04) compared with controls. In addition, there was an overall increase in alpha diversity indices in the bladder cancer group (Shannon P = 0.00029, Simpson P = 0.0029, Chao P = 0.001, Observe P = 0.00051; Supplementary Index S1), although the differences were minor (Fig. 1A–C).

Principal coordinates analysis (PCoA) for beta diversity (Fig. 1D) demonstrated distinct microbial community separation between the groups, associated with disease presence (bladder cancer vs. CTR; Fig. 1D).

Using these data, we also identified differential bacterial abundance between the GM of healthy controls and patients with bladder cancer, reaffirming the previously noted increased levels of Prevotella and Porphyromonas in the bladder cancer group (LDA > 3.5, P < 0.01 for both; Fig. 1E and F).

In addition, we explored the influence of factors like gender on microbial composition, comparing the abundances of predominant bacteria such as Bacteroides (32.01% vs. 27.8%), Prevotella (4.2% vs. 5.5%), and Escherichia coli (4.0% vs. 6.9%) between males and females, respectively. Our analysis showed no statistically significant differences in these bacterial abundances between genders (P > 0.05; Supplementary Fig. S4A).

Furthermore, we examined alpha diversity, measured by the Shannon index, and found no significant differences between males and females (P = 0.47). This was further supported by the lack of discernible clustering between the two groups, as visualized in Supplementary Fig. S4B, S4C, and S4E). While there was a total of 64 unique OTUs identified in males, the majority of these were overlapping with those found in females (157 OTUs shared; Supplementary Fig. S4D). Our results indicate that, at least in the context of the abundances of the selected bacteria, alpha diversity, and clustering patterns, there were no notable differences observed between males and females.

We aimed to determine which demographic or outcome factors had the strongest associations with GM profiles by utilizing the MaAsLin2 analysis. This thorough analysis identified sex, smoking status, and cohort as significant covariates that were statistically associated with variations in bacterial profiles (Fig. 2A–C, Supplementary Appendix).

Specifically, smoking status had a noticeable effect on the bacterial abundance within the Anaerostipes, Eggerthella, and Fusicatenibacter taxa (Fig. 2A). The differences in bacterial abundance were especially marked when comparing current smokers with former smokers. In addition, age was linked to changes in the UBA1819 and UCG0002 taxa, with older adults showing a higher association and prevalence of these bacterial profiles (Fig. 2B and C). The stage at the time of cystectomy and the institution were also associated with distinct microbiome profiles, notably affecting Pseudomonas and Escherichia-Shigella populations. Interestingly, a decreased abundance of Desulfovibrio was associated with a pT0 (no evidence of disease) status at cystectomy, even after adjusting for NAC, age, body mass index (BMI), sex, and variant histology.

Given the potential influence of compositional differences on both cancer development and therapeutic response, we investigated how specific GM compositions affect the response to NAC, as indicated by CR versus NR at the time of surgery. We analyzed GM OTU abundance in CR versus NR cases (n = 57), comparing the enrichment of OTUs in CR (n = 23) versus NR (n = 34) cases (Fig. 3A). No significant differences were observed in alpha diversity (Ace P = 0.35, Chao P = 0.34, Observe P = 0.35, Shannon P = 0.17) or beta diversity (P = 0.383) between the two groups (Fig. 3B and D). CR cases exhibited both overlapping and unique OTUs compared with NR, including taxa such as Proteus, Fastidiosipila, and Tolumonas (Fig. 3E). In contrast, NR cases showed a higher abundance of Bacteroides (26.95% vs. 18.93%, P < 0.01) and Pseudomonas (2.18% vs. 1.67%, P = 0.03) compared with CR (Fig. 3A and C). High-dimensional class comparisons using LEfSe, adjusted for overall abundance, identified genus GCA-900066575 (an unidentified member of the Lachnospiraceae) as the only bacteria differentiating CR from NR, being enriched in NR subjects (LDA > 3, P < 0.01; Fig. 3F). In contrast, an increased abundance of Lachnospiraceae, known for producing SCFAs, was associated with the responder status (LDA > 3, P < 0.01; Fig. 3F and G).

Because of significant differences in composition and associations with responses to NAC regimens, we investigated whether collective abundances within specific treatments correlated with CR, beyond individual bacterial genera. We grouped all identified OTUs using hierarchical clustering without incorporating response data. This analysis revealed that the bladder cancer cohort divided into two groups, with the majority of patients with bladder cancer falling into community Group I (n = 115, 80.9%), which included many patients with a history of NAC (n = 48, 84.2%; Supplementary Fig. S5A). As expected, significant differences in alpha and beta diversity between the groups emerged because of the nature of the analysis (Supplementary Fig. S5B and S5C; Supplementary Appendix). Despite Group I containing a larger proportion of nonresponders compared with Group II (60.4% vs. 55.6%, P = 0.78), this difference was not statistically significant (Supplementary Fig. S5E and S5F).

Compared with Group II, Group I patients showed a higher abundance of Bacteroides (34.05% vs. 8.59%, P < 0.001) and a decrease in Prevotella (1.37% vs. 17.47%; P < 0.01; Supplementary Fig. S5A). Group I contained 136 unique OTUs, whereas Group II contained 19 unique OTUs (NR group; Supplementary Fig. S5D). Overall, the grouped communities within patients with bladder cancer did not predict NAC response status based on hierarchical clustering.

Because the CW cohort samples were collected at the time of surgery post-NAC, our study then shifted focus to identify gut compositional differences in patients with bladder cancer before any treatment, during, and after chemotherapy, compared with their partners. To account for environmental influences such as diet, housing, and exposure, we also collected stool samples from live-in partners at the same timepoints. Throughout the course of chemotherapy, no significant changes were observed in alpha and beta diversity in patients with bladder cancer (Supplementary Fig. S6A, S6B, and S6F). However, there were noticeable fluctuations in composition, particularly an increase in the prevalence of Bacteroides, Escherichia-Shigella, and Roseburia after NAC compared with before (Supplementary Fig. S6A and S6D). Although eight OTUs were unique to the post-NAC samples (Supplementary Fig. S6E), the overall diversity and distribution remained similar across all three timepoints, with 102 OTUs common to each (Supplementary Fig. S6A–S6D and S6F). LDA showed no significant differences in GM composition before, during, or after chemotherapy (Supplementary Fig. S6C).

While both patients and partners showed no significant changes in diversity throughout the NAC period (Supplementary Fig. S6A–S6D, Supplementary Fig. S7B and S7C), patients with bladder cancer experienced increases in Bacteroides, Escherichia-Shigella, and Roseburia post-NAC, in contrast to their partners, whose abundances remained stable (Supplementary Fig. S7A and S7D–S7F).

In addition to assessing GM differences in our cohorts, we expanded our analysis to include SCFA concentration differences in stool and plasma among the cohorts. We specifically measured the levels of acetic, propionic, isobutyric, butyric, isovaleric, valeric, and hexanoic acids in both the responder and nonresponder groups (n = 18), as summarized in Fig. 4.

Our results showed no statistically significant differences in SCFA concentrations between the two groups (CR vs. NR) in either stool or plasma (Fig. 4). However, we noted a significant difference in the average concentration of acetic acid, which was substantially higher in fecal samples than in plasma (14.67 ± 1.48 vs. 2.06 ± 0.48, P < 0.001). While all tested SCFAs were detectable in stool, only propionic, isovaleric, and valeric acids showed minimal presence in plasma.

In addition, when examining the impact of NAC on plasma SCFA levels, we observed an increase in detectable levels of acetic acid (1.56 vs. 0.71, P = 0.12), isobutyric acid (0.036 vs. 0.01, P = 0.01), and butyric acid (0.016 vs. 0.01, P = 0.01) in patients who underwent NAC compared with those who did not (Supplementary Appendix). Further analysis of the association between SCFA-producing bacteria (including Bacteroides, Bifidobacterium, Enterobacteriaceae, Lactobacillus, Bacillaceae, and Ruminococcus) and SCFA levels revealed no positive correlations in stool or blood (Supplementary Appendix). However, broader analysis identified a correlation between increased fecal isobutyric acid levels and higher abundances of Akkermansia (rs = 0.51, P = 0.017) and Clostridia (rs = 0.52, P = 0.018). In addition, Lactobacillus (rs = 0.49, P = 0.02) and Enterobacteriaceae (rs = 0.52, P < 0.03) showed positive correlations with elevated propionic acid levels (Supplementary Appendix).

Finally, our univariate analysis exploring the associations between SCFA concentrations in blood and stool and NAC response status did not yield any statistically significant findings (Supplementary Appendix).

We utilized a random forest machine learning approach to develop a phenotype prediction classifier for CR in patients with MIBC, based on a set of microbial genera. Our analysis included 57 stool samples. After conducting five rounds of 1,000-fold cross-validation, microbial variables emerged as the optimal predictors for CR, achieving an AUC of 0.88 (95% confidence interval: 0.81–0.94; Fig. 5A) with a sensitivity of 0.70, specificity of 0.80, positive predictive value of 0.88, and a negative predictive value of 0.571 (Fig. 5C). In the machine learning model, the genera Oscillibacter, Fusicatenibacter, and Bacteroides were identified as having the highest levels of importance (0.051, 0.037, and 0.032 respectively), although no single genus exhibited a dominance, indicating a complex multigenera influence in CR (Fig. 5B).

In addition, we assessed 24 clinical variables to evaluate their potential in predicting CR. Following the same modeling process, the AUC for these clinical factors was 0.50, indicating no discriminatory power based solely on clinical and pathologic factors for predicting CR, unlike the microbiome signature (Supplementary Appendix). This suggests that predictions for NAC response are feasible based on a patient's microbiome profile, rather than clinical characteristics alone.

To assess GM changes associated with tumorigenesis, stool samples from mice given BBN (a bladder carcinogen) or regular drinking water were collected and analyzed at four timepoints: (i) baseline, (ii) during treatment, (iii) at the end of BBN exposure, and (iv) when tumors were visible radiographically, just before sacrifice. This was done to account for age-related changes. Our initial analysis of the longitudinal changes within the control and BBN-exposed mice showed no differences between the groups at individual timepoints (Fig. 6C, E, and G). While alpha diversity was not significant (Fig. 6A), beta diversity analysis did reveal clustering at time zero compared with later timepoints, suggesting changes were due to gut maturation rather than exposure to BBN (Fig. 6B).

Given the observed differences in Bacteroides abundance in patients with bladder cancer at different institutions, we monitored the abundance of Bacteroides in both BBN-exposed and control mice throughout the experiment. We observed a higher prevalence of Bacteroides in BBN-exposed mice (12.30% vs. 6.89%; Supplementary Index S1; Supplementary Fig. S8F–S8J), which interestingly normalized to control levels by the end of the 12-week BBN exposure period (Supplementary Fig. S8G). In both BBN and water control groups, there was a higher abundance of Akkermensia and Verrucomicrobiota at pretreatment, which then declined at subsequent timepoints (6, 12, 16–22 weeks), regardless of exposure (LDA 5, P < 0.05; Supplementary Fig. S8B–S8E and S8G–S8J). There were no differences in alpha diversity by exposure and week (Supplementary Fig. S8A and S8B), although week-specific clustering was present among both groups (Supplementary Fig. S8D and S8I).

Considering the potential for sex-dependent changes in microbial composition (16), we also analyzed compositional differences in the microbiome in response to BBN exposure by sex. No significant differences in alpha and beta diversity (Supplementary Fig. S9A–S9D, S9F–S9I) were observed in males or females separately throughout the experiment. Week-dependent differences in overall abundances were seen in both males and females (Supplementary Fig. S9B and S9G), although not significant with further LDA analysis (Supplementary Fig. S9E and S9J).

In this article, we report the most extensive study to date on the association between the GM and the response to NAC in MIBC. We not only explored differences between control (CTR) and patients with bladder cancer but also examined changes in the GM throughout NAC. A significant observation from our study is the link between the presence of Bacteroides and poor NAC response in patients with MIBC. This association is not limited to bladder cancer; similar reductions in the effectiveness of both immunotherapy and chemotherapy have been noted in cases of melanoma (7), pancreatic (17), and colon cancers (18). Our findings align with previous research suggesting a connection between GM and tumorigenesis, particularly noting increases in Prevotella and Bacteroides in patients with bladder cancer, which are commonly associated with both tumorigenesis and therapy response (19–21).

Our data indicate that both the baseline gut microbiota and its changes during chemotherapy are crucial predictors of CR. The microbial signatures we identified showed better predictive power for CR outcomes post-NAC than traditional clinical variables alone. However, for improved accuracy and reliability, future studies should consider integrating microbial analyses with clinical data, genomic classifiers, and circulating tumor DNA. The primary goal of our research is to identify patients unlikely to benefit from NAC, who might instead benefit from alternative systemic treatments or early surgery. By initiating the characterization of a predictive microbial signature, we encourage further large-scale studies to validate our results and promote clinical trials integrating GM modulation for therapeutic advantages, along with a combination of tumor, host, and microbial variables.

The dynamics of GM changes during NAC in bladder cancer have not yet been thoroughly examined. However, existing studies in colorectal (22) head and neck (23) and cervical (24) cancers have all documented significant microbiome fluctuations throughout treatment courses. In line with these findings, our study observed that the longitudinal changes in the microbiota were primarily marked by an increase in Bacteroides during NAC. A unique aspect of our research is the inclusion of patients’ live-in partners as a control to mitigate environmental influences on microbiome changes. Notably, while patients showed an increase in Bacteroides at the end of treatment, their partners did not exhibit similar shifts, supporting the notion of a genuine microbiota alteration rather than an incidental change. In addition, previous studies have highlighted a correlation between close social relationships and the composition of gut and skin microbiota, with the greatest similarities found between spouses (25). Our findings corroborate this, suggesting that the microbiome composition of couples exhibits notable concordance.

It is important to also highlight an interesting yet still controversial notion of existence of “enterotypes” characterized by dominant genera (Bacteroides, Prevotella, and Ruminococcus), and their co-occurring dietary traits, first published by Arumugam and colleagues (26) In particular, the Bacteroides-dominant enterotype was found to be associated with high animal protein and saturated fat, while the Prevotella enterotype was observed with high fiber/plant-based nutrition (low meat). As there is growing awareness of importance of dietary phenotype with health and disease and recognition of host-microbiome mutualism, it is difficult to say whether the dietary phenotype in our study (high fiber/plant-based Prevotella subtype) and response to NAC is result of a modifiable intervention or confounding by an environmental factor.

Our analysis of human plasma and fecal samples revealed no significant differences in concentration levels between the responder and nonresponder groups in assessing the efficacy of NAC. Although the variations in acetic acid levels in stool and plasma were not statistically significant, a consistent elevation was noted in patients who achieved a CR. This finding is in line with research on other types of cancer, such as urothelial carcinoma, highlighting acetic acid's role in promoting cellular apoptosis and its potential association with improved progression-free survival in colorectal cancer (27). Furthermore, the levels of acetic, propionic, isobutyric, butyric, isovaleric, valeric, and hexanoic acids in both fecal and plasma samples showed no correlation with known SCFA producers like Bacteroides, Bifidobacterium, Enterobacteriaceae, Lactobacillus, Bacillaceae, and Ruminococcus, contrary to what is commonly reported in literature. Previous research has suggested that propionate might boost the antitumor effects of cisplatin in human hepatocellular carcinoma cells (28), yet we found no such correlation in patients with bladder cancer. Likewise, although butyrate is reported to increase apoptosis and suppress growth, migration, and invasion in gastric cancer cell lines treated with cisplatin, our study did not find a relationship between SCFA levels and NAC response (29).

Our study was designed to track the longitudinal changes in mice exposed to a carcinogen, aiming to pinpoint crucial moments of transformation and dysbiosis. In our experiments, mice exposed to BBN exhibited a consistent increase in Bacteroides up to week 12, the point at which BBN exposure was halted, followed by a rapid decrease once the carcinogen was withdrawn. While the gut microbiotas of humans and mice share considerable similarity, with overlaps of 90% at the phylum level and 89% at the genus level, differences at the species level are significant. For example, our study noted a steady increase in Bacteroides during the 12-week BBN exposure period. In mice, this genus is primarily represented by the S24-7 family, especially Bacteroides ovatus, whereas in humans, Bacteroides vulgatus is more common (30). While subtle differences in microbiome composition exist, employing murine models for microbiome-based therapeutic research is essential for developing new treatments in a controlled setting that is not possible in human studies. Although the mice in our study did not receive treatment for their BBN-induced tumors, limiting the implications of these findings, our research does provide evidence supporting the potential for Bacteroides manipulation in murine models, particularly for future therapeutic developments.

Our study has several limitations. While we collected clinical data for patients, we did not acquire concurrent tumor data such as genomic mutations and immune infiltration, which are critical for enhancing the response classifier. In addition, although we excluded patients who received antibiotics within 6 weeks prior to therapy, we did not assess nutritional data or over-the-counter medications that might influence the GM. However, our study is distinct in its inclusion of patients’ partners, which allowed us to adjust for environmental exposures that could affect microbiome composition. Moreover, because our patients received different NAC regimens—GC and MVAC—the differences in stool microbiome observed with MVAC may not apply to patients treated with GC, and vice versa. This necessitates further evaluation in a prospective setting among both GC and MVAC patient populations.

In conclusion, our findings indicate that certain microbial features in the baseline GM of patients with bladder cancer are linked to chemotherapy response. This is supported by real-world cohorts, highlighting an increased abundance of Bacteroides as a negative prognostic factor for CR. The data presented call for further validation in a large, prospective trial that also integrates immune and genomic markers from the host and tumor. We hope these preliminary results are promising and pave the way for additional ongoing trials.

L. Bukavina is an associate editor of European Urology and editorial board member of Urology Times. Not relevant to publication. D.M. Geynisman reports personal fees from Exelixis, Merck, AstraZeneca, and Bristol Myers Squibb outside the submitted work. J.R. Brown reports personal fees from Pfizer/Seagen, EMD-Serono, and AstraZeneca outside the submitted work. E.R. Plimack reports personal fees from AbbVie, Astellas, AstraZeneca, BMS, Eisai, EMD Serono, Exelixis, IMV, Merck, Pfizer, Regeneron, Seagen, Seattle Genetics, Signatera, Synthekine and grants from Merck outside the submitted work. P.H. Abbosh reports grants from Janssen and Natera outside the submitted work; in addition, P.H. Abbosh has a patent to urine biomarkers pending. No disclosures were reported by the other authors.

L. Bukavina: Conceptualization, resources, data curation, software, methodology, writing-original draft, project administration, writing-review and editing. R. Ginwala: Conceptualization, resources, data curation, software, formal analysis, writing-review and editing. M. Eltoukhi: Conceptualization, data curation, validation, methodology, writing-review and editing. M. Sindhani: Formal analysis, visualization, methodology, writing-review and editing. M. Prunty: Writing-review and editing. D.M. Geynisman: Project administration, writing-review and editing. P. Ghatalia: Writing-review and editing. H. Valentine: Project administration, writing-review and editing. A. Calaway: Writing-original draft, project administration, writing-review and editing. A.F. Correa: Writing-review and editing. J.R. Brown: Writing-review and editing. K. Mishra: Visualization, methodology, writing-review and editing. E.R. Plimack: Writing-review and editing. A. Kutikov: Project administration, writing-review and editing. M. Ghannoum: Project administration, writing-review and editing. M. Elshaer: Validation, investigation, writing-review and editing. M. Retuerto: Data curation, software, formal analysis, supervision, writing-review and editing. L. Ponsky: Writing-original draft, project administration. R.G. Uzzo: Project administration, writing-review and editing. P.H. Abbosh: Conceptualization, resources, methodology, writing-original draft, writing-review and editing.

This study is supported by Raymond Pominville MD for collection of samples.

Source of Funding: P30CA043703 Case Western Comprehensive Cancer Center Microbiome Grant, P30 CA006927 Fox Chase Cancer Center Support Grant, NIH grant no. R01AI145289-01A1 (M. Ghannoum), CA181178 Department of Defense CDMRP (P.H. Abbosh).