Urine specimens were collected from two cohorts of patients. The first cohort of 76 patients had urothelial carcinoma and was enrolled into a prospective clinical trial at Fox Chase Cancer Center to assess the accuracy of cystoscopic evaluation in predicting tumor stage compared to stage at the time of radical cystectomy (NCT02968732, “The pT0 Trial” (19)). Patient characteristics have been listed in Table 1 . All participants gave their informed consent for collection of urine samples and the subsequent analyses for research purposes. Inclusion/exclusion criteria have been previously described.

The second cohort was enrolled from outpatient clinics from Albert Einstein Medical Center. Patients provided informed written consent to an IRB-approved specimen collection protocol (2019-88). Urine was collected from patients undergoing benign, nonendoscopic general urological procedures such as vasectomy, circumcision, hydrocelectomy, wart removal, etc., at 4 time points: at the time of consultation, at the time of procedure (prior to administration of any antibiotics), at the time of wound check (typically ~1 month), and at the time of follow up (typically ~3 months). Not all patients followed up at all 4 time points.

Clean-catch midstream urine was collected in sterile containers and kept at 4°C until processing within 1-2 h of sampling as per below.

Urine samples were processed immediately upon receipt. Urine was centrifuged at 4°C at 4000 rpm for 5 min and separated into supernatant and pellet. The pellet was then washed with phosphate buffered saline (PBS) at >10,000 rcf in a microfuge for 3 min. The pellet was frozen at -80°C for DNA extraction at a later time.

DNA was extracted from thawed pellets using a combination of mechanical, thermal and enzymatic disruption techniques. Briefly, the urine pellets were resuspended in a lysis buffer containing 500mM sodium chloride, 50mM Tris-HCl (pH 8.0), 50mM EDTA, and 4% sodium dodecyl sulfate. To this, 50 µl lysozyme (10 mg/ml, Sigma-Aldrich), 6 µl mutanolysin (25 KU/ml, Sigma-Aldrich), and 3 µl lysostaphin (4000 U/ml, Sigma-Aldrich) were added along with 0.4g of sterile zirconia beads (Benchmark Scientific Inc, Sayreville, NJ). Samples were then homogenized in a Mini-BeadBeater™ (BioSpec Products, Bartlesville, OK) at maximum speed for 3 min. Homogenized samples were further incubated at 37°C for 45 min with periodic gentle shaking every 5 min. Subsequently, the samples were heated for 30 min at 70°C. Samples were then centrifuged at 4°C for 5 min at 16,000 rcf. The resulting supernatant was collected, and the pellet was resuspended in lysis buffer for a second round of lysis. The pooled supernatant was then subjected to acetate/isopropanol DNA precipitation with glycogen, washed with 70% ethanol, air dried, resuspended, and stored at -20°C. We found that this approach gave the most consistent 16S profile when compared head-to-head with two commercial kits when applied to homogenize mouse stool pellets.

5-10 micron-thick sections of diagnostic TURBT formalin-fixed paraffin-embedded (FFPE) tissue specimen were prepared. These slides were stored at 4°C until DNA extraction for 16S analysis. Using a fresh sterile blade for each sample, FFPE sections were cut. Mineral oil was used to deparaffinize the tissue. QIAamp DNA FFPE kit (Qiagen) was used for DNA isolation. The aqueous phase (lower, uncolored phase) was transferred to a tube containing sterile zirconia beads and the solution was homogenized in the Mini-BeadBeater™ for 10 minutes. To this, an enzyme cocktail consisting of 50 µl lysozyme (10 mg/ml, Sigma-Aldrich), 6 µl mutanolysin (25 KU/ml, Sigma-Aldrich), and 3 µl lysostaphin (4000 U/ml, Sigma-Aldrich) was added and the tubes were transferred to a water bath set at 37°C for 30 minutes. The supernatant from this step was then taken through the rest of the manufacturer’s protocol to obtain DNA, which was then quantified and used for 16S.

Post-chemotherapy urine samples from 7 patients on the RETAIN cystectomy avoidance trial (NCT02710734) (20) were used to isolate DNA as described. The takara Thru Plex DNA HV kit was used to generate low pass shotgun sequencing data using 100PE Illumina chemistry. Whole-Genome Sequencing (WGS) Analysis was performed on samples to characterize microbial composition and validate findings from 16S rRNA sequencing. Raw sequencing reads in.fastq format were subjected to quality filtering and adapter trimming using Fastp to remove low-quality bases and short reads. High-quality reads were classified using Kraken2, a k-mer-based taxonomic classifier, against the Standard-Full RefSeq database, which includes archaea, bacteria, viruses, plasmids, human, and UniVec_Core sequences. Kraken2 reports were processed using KrakenTools to extract relevant taxonomic assignments and generate summarized classification reports. Multiple Kraken2 reports were combined to create a final aggregated report, ensuring comprehensive microbial classification across all samples. Species-level identification was performed to determine the relative abundance of microbial taxa. Decontamination was conducted to remove potential human-derived sequences and environmental contaminants from the final dataset.

To account for bacterial contamination from the laboratory environment, we included negative controls at each stage of the procedure. Contamination during DNA extraction was accounted for by including extraction from a water blank. Impurities introduced at the PCR stages were accounted for by including non-template controls. As positive control for DNA extraction, we collected 12 fresh mouse stool pellets, pooled/rehydrated/homogenized/aliquoted them, into microcentrifuge tubes, and stored at -80°C. One aliquot was used during each batch of DNA extraction. As a second positive control at the PCR stage, we included DNA extracted from the first mouse stool pellet. Both positive controls were carried through all subsequent stages of analysis to ensure reproducibility and identify significant deviations (i.e. batch effects). For the microbiome analysis of TURBT samples, to control for contamination introduced during the paraffin embedding and formalin fixing process, we processed empty paraffin from the edges of the FFPE specimens as negative controls. These were carried through the entire analysis process and any reads detected from these specimens were identified as contamination and excluded from the reads originating from the test samples.

Mice were studied under an IACUC approved protocol (20-09). Specific pathogen free C57B6 mice (n=56; 33 male, 23 Female) were purchased from Taconic and housed in a specific pathogen free environment. At age 8 weeks, mice were treated with 0.1% BBN in drinking water or regular drinking water from the same source ad libitum. Cages were randomized as to which treatment the cage received, and male and female mice were housed separately. At 6 and 12 weeks, mice were randomly chosen from each cage for euthanasia by C0₂ inhalation. Bladder tissue was collected and split into two parts. One part was subjected to DNA extraction (described below) and the other part was kept at -80°C. Mice were imaged using µCT excretory urography for the presence of tumors starting at week 14, and when tumors were identified (between 16-22 weeks), mice were sacrificed for bladder collection.

DNA was isolated from bladder tissue first using proteinase K digestion, then with enzymatic cocktail (mutanolysin, lysostaphin, lysozyme) as described above. DNA was then isolated using phenol chloroform extraction/ethanol precipitation. Pelleted DNA was quantitated as below.

The quantity of the purified genomic DNA from both urine and fecal samples was estimated with the Qubit dsDNA high-sensitivity assay (ThermoFisher Scientific, Waltham, MA) using the Qubit 4 Fluorometer.

DNA isolated from urine or fecal samples was subjected to 16S PCR amplification using primers for the V3-V4 regions of the bacterial 16S rRNA in two stages. To ensure accurate base-calling on Illumina platforms and reduce the need for phi X DNA, we introduced sequence diversity using a series of frameshifts in the 16S amplicons between the adapter and the V3 or V4 annealing sites in the first stage. This concept was described and validated previously (21). List of 16S PCR amplicon primers and indexing primers can be found in Supplementary Table 1 . Briefly, 10x PCR buffer, AccuPrime™ Pfx DNA Polymerase (ThermoFisher Scientific), nuclease-free water, a unique pair of frameshift Stage I primers, and 20 ng template DNA were mixed together and amplified in using the conditions listed in Table 2 Stage I. Following bead-clean up at a ratio of 1.25:1 (PCR product: AMPure XP beads), stage 2 indexing PCR was carried out under the conditions listed in Table 2 Stage II. Bead-clean up using beads at a ratio of 1:1.12 (PCR product: Ampure beads) was performed after which the purified libraries were quantified using the Qubit 4 Fluorometer and validated on a Tapestation 2200. Libraries were pooled together at equimolar ratio to create a 4nM pool which was then denatured with NaOH following Illumina’s 16S Metagenomic Sequencing Library Preparation protocol for Library Denaturing and MiSeq Sample Loading. Library was spiked with 1-2% PhiX (10 pmol) before sequencing on Illumina MiSeq platform with paired-end 300bp reads. Sequencing was performed at the Johns Hopkins Sequencing and Synthesis facility.

Detailed description of this analysis pipeline has been previously published (25). Filtered sequences with >97% identity were clustered into Operational Taxonomic Units (OTUs) and classified at the genus against the SILVA 16S ribosomal RNA sequence database (release 138.1). The relative abundance of each OTU was determined for all samples. QIIME 2 platform was utilized for processing and visualizing microbiome data.

Alpha diversity metrics between the various groups were compared using Wilcoxon rank-sum or Mann-Whitney (MW) test. Adjustments for multiple comparisons were done using the false-discovery rate (FDR) method at an α level of 0.05. Differentially abundant OTUs were identified using LEfSe (linear discriminant analysis (LDA) effect size) for each pairwise comparison of groups. All analyses were conducted in R and Python.

J82, SW780, and UMUC3 cell lines were used, as well as a cell line that we derived from a mouse that was treated with BBN and developed a tumor. The tumor was dissociated and grown as a monolayer in DMEM+10% FBS. No antibiotics were used in maintenance media unless otherwise specified. Cells were grown in humidified 5% CO₂ and were checked for Mycoplasma approximately every 3 months using PCR-based detection as described previously (22).

RFP-labeled UTI89 is a uropathogenic strain of E.coli (UPEC) and was a kind gift from Molly Ingersoll PhD (Institut Pasteur, Paris France). Fresh cultures were grown in LB and CFU/mL were quantitated by streaking agar plates, while storing the undiluted liquid culture at 4°C. Once the titer was calculated, the culture was diluted to 1M CFU/mL and stored in 1 mL aliquots 15% sterile glycerol/LB at -80°C. One week later, the number of CFU was assessed by streaking agar plates in triplicate to adjust for any loss of viability during cryopreservation. This approach provided a true “expected” live bacterial count to use in later experiments.

For all infections, cell lines were infected at a multiplicity of infection of 500. Plates were centrifuged in a swinging bucket rotor at 1000 rcf for 10 min at room temperature to increase bacterial uptake by concentrating them at the bottom surface of the dish. Following centrifugation, the bacteria were allowed to grow intracellularly for 30 min at 37°C. Then, media was exchanged for fresh media containing gentamicin at 50µg/mL, which is bactericidal in the media but does not penetrate into the cells.

Cells were also imaged using the Leica confocal microscope in the Biological Imaging Facility of Fox Chase using Hoechst 33342 to localize the nuclei and phalloidin to image cell cytoplasm. Briefly, cell lines were seeded in the wells of 8-well chamber slide at a density of up to 10,000 cells per well and left to attach for 24 hours. Next day, cells were infected as described previously. Post infection cells were fixed with 4% PFA. This ensured that the bacteria were actually inside the cells and not simply attached to the cell membrane.

Cell TiterGlo was used to estimate the concentration of gemcitabine that inhibits growth by 50% (GI₅₀). Briefly, 5,000 to 10,000 cells were plated into each well of a black 96 well dish the night before treatment with 0 to 1000 nM concentrations of gemcitabine. Cell TiterGlo was added to the media per manufacturer’s instructions 3 days later and the GI₅₀ was estimated using Graphpad Prism. For determining the effect of UPEC on gemcitabine toxicity, Cell TiterGlo was performed with cells infected with UPEC as described earlier.

Additionally, a colony formation assay was performed to test the toxicity of gemcitabine in the presence of UPEC. 10,000 cells were plated in wells of 24-well plates and allowed to attach for 24 hours prior to infection. On day 2, half the plate was infected with UPEC as shown earlier. Cells were then treated with 0.5x, 1x, and 2x gemcitabine IC50 dose and plates were incubated for 4 hours (J82,BBN,SW780) or 24 hours (UMUC3) after which, gemcitabine-containing media was removed, and fresh media was added. On day 5, media was removed, and cells were washed with PBS. Wells were then stained with 0.25% crystal violet stain (20% methanol) for 30 minutes. The stain was removed, and wells were thoroughly rinsed with water. Plates were allowed to dry o/n then imaged.

To assess the effect of UPEC on gemcitabine metabolism, we performed HPLC and mass spectrometry analysis. Briefly, UPEC was grown to an OD₆₀₀ of about 1 in LB media. Gemcitabine was added to 200 µl of bacterial sample at a concentration of 5x GI₅₀ in 5ml of cell culture media. UPEC and gemcitabine were incubated for 1 h at 37°C, after which the samples were centrifuged for 15 mins. The resulting supernatant was filtered using a 0.2µM PES filter. 200 µl of this supernatant was mixed with 800 µl methanol in a pre-chilled tube, vortexed and incubated at -80°C overnight. The next day these samples were centrifuged at 13,000 rpm at 4°C for 15 mins. The resultant supernatant was stored in cryovials at -80°C till the time of analysis at the Wistar Metabolomics and Proteomics Institute.

Microbial sequences were filtered out from TCGA data from the BLCA cohort and analyzed using Kraken (23), which provides comprehensive classification of metagenomics sequences for identification of microbial and viral members by aligning short reads to bacterial genomic databases. We analyzed 116 tumors, 22 adjacent normal tissues, and 99 blood samples. Mean relative abundance of each OTU in tumor tissue was normalized against all blood samples or all matched adjacent normal tissues to identify taxa that were enriched or depleted. The same analysis was performed to compare adjacent normal vs blood. Several known bladder pathogens were enriched compared to blood including Enterococcus spp, Escherichia coli, and Prevotella melananogenica, in addition to viruses reported by TCGA (24). There were 27 taxa in all, engendering confidence in the approach. The most relevant species have been listed in Figure 1A . Thirty-two species were enriched when considering the tumor plus adjacent normal tissue compared to blood denoting a possible ‘microbial field defect’ in the malignant bladder. Notably, bacterial species such as B. thetaiotamicron, A. mucinaphila and Bifidobacterium spp. that have been previously associated with a positive response to anti-CTLA4 (25) and anti-PD-L1 (26) therapies when found in the stool of cancer patients and preclinical models were enriched in tumors. Further, the presence of Alphapapilloma viruses HPV16 and HPV18, the causative agents of cervical cancer (27, 28), were enriched in tumors but not normal bladder. Differential expression of the TP53 (p=0.56), RB1 (p=0.07) and CDKN2A (p=0.72) was seen between HPV positive and negative tumors, however, due to the low number of HPV positive tumors, this difference was not pronounced ( Figure 1D ).

Because TCGA data includes replicate sequencing data for some tissue samples, available replicates were cross validated to each other as a test of reproducibility. A total of 22 tumors had duplicate whole genome sequencing. If a taxon was present with at least 5% abundance in any replicate, we determined its concordance in replicates. Some tumors had triplicate samples, in which three case pairs were evaluated. In total, 119 of 287 taxa were concordant (R=0.33, p=1x10^-8, Pearson; Figure 1B ), suggesting there is a true underlying microbiome in this data, and it can be reproducibly detected.

As described in previous reports the urinary microbiome differs between males and females (6). We therefore determined if there was a prevalence of sex-specific taxa in the TCGA data from the bladder cancer patients. We found that Lactobacillus and Prevotella were enriched in female bladder cancer patients (15% vs 4%, p=0.02; 16% vs 4%, p=0.045 respectively, Fisher’s test). These genera are prevalent in the normal female genitourinary tract (6), supporting the validity of the approach. We also saw that A. prevotii and F. magna were enriched in females (15% vs 4%; p=0.045).

We leveraged RNAseq data from TCGA samples from which we made tumor microbiome calls to determine if specific genera were associated with intrinsic bladder cancer subtypes (24, 29–31). We found that Prevotella was more prevalent in TCGA subtypes III and IV (0%, 3%, 16% and 18%, in subtypes I, II, III, and IV, respectively, p=0.045, Fisher’s test). Finally, RNAseq data was used in single sample gene set enrichment analysis (32) using the Hallmark (33) and KEGG gene sets to show the association of Mycoplasma with the Hallmark DNA repair gene set and with the KEGG pyrimidine metabolism pathway ( Figure 1C ). Mycoplasma and other Gammaproteobacteria found in pancreatic cancers was recently shown to metabolize gemcitabine, a pyrimidine nucleoside, using a long isoform of cytidine deaminase (17), and it is possible that this enzyme has additional unrecognized effects on DNA repair or pyrimidine metabolism.

Several GU tract pathogens are Gammaproteobacteria, such as E.coli, P. mirabilis, P. aeruginosa, Klebsiella spp, and E. cloacae. Gemcitabine is a frequently used systemic or local/intravesical therapy in patients with bladder cancer. We therefore set out to determine if gemcitabine metabolism by Gammaproteobacteria would affect gemcitabine toxicity in bladder cancer. We first established that E.coli would enter cells by infecting SW780, J82, UMUC3, and our BBN-induced tumor derived cell line (‘BBN’; Figure 2A ). Spinfected cells were then imaged using a confocal microscope to ensure that bacteria were able to enter into the cells. We next exposed 4 bladder cancer cell lines to increasing doses gemcitabine in 96 well dishes with or without exposure to the RFP-labeled UTI89 strain of E.coli to calculate the 50% growth inhibitory concentration (GI₅₀) using Cell TiterGlo. This showed that uropathogenic strains of E.coli did indeed result in lower efficacy of gemcitabine ( Figure 2B ).

Once GI₅₀ were set, each of the 4 cell lines was exposed to 0.5x, 1x, or 2x the GI₅₀ with or without UTI89 for colony formation assays. Colony forming ability of the cell lines after treatment with gemcitabine was restored in the presence of UPEC infection indicating loss of gemcitabine toxicity ( Figure 2C ). HPLC analysis further confirmed gemcitabine metabolism to its less-toxic metabolite 2′,2′-difluoro-2′-deoxyuridine (dFdU) when incubated together with UPEC ( Figure 2D ).

The 16S metagenome of urine samples collected from bladder cancer patients (n=31, urine samples obtained prior to initiating chemotherapy) was compared to that of non-bladder cancer controls (N=13; Figure 3A ).There were no significant differences in typical measures of alpha diversity ( Figure 3B ). However, a clear difference in the microbial community structure as measured by principal components using weighted unifrac distances was observed between the bladder cancer and non-cancer controls ( Figure 3C ). Relative abundance of taxa were distinct between the two groups as seen in Figure 3D . As shown using the linear discriminant analysis tool LEfSe, there was an enrichment of 15 OTU including Enterobacteriales, Flavobacterium, Varicubaculum, Facklamia in bladder cancer. There was also differential abundance 29 OTU in healthy controls including Pseudoxanthomonas, Gardnerella, Bifidobacteriales, Lactobacillus, Verrucumicrobiales ( Figure 3E ).

Not much is known in terms of the effect of neoadjuvant chemotherapy on the urinary microbiome. In the pT0 Trial, some patients received neoadjuvant chemotherapy, and in these cases, urine samples were obtained both pre- and post- administration (See Table 1 for details of chemotherapy). Based on alpha diversity metrics and weighted unifrac distances using principal components, no significant difference was seen in the microbial diversity and composition post chemotherapy ( Figures 4A–C ). Relative abundance of taxa at genus level were distinct between urine samples taken pre- and post- chemotherapy (Figure 4D). Individually, the patient profiles of microbial composition pre- and post-neoadjuvant chemotherapy show differences in the types of bacterial communities present at the genus level ( Supplementary Figure 1 ). LDA scores revealed differential enrichment of Streptococcaceae in patients post chemotherapy vs Lawsonella, Varibaculum, Sanguinis prior to chemotherapy ( Figure 4E ).

Given the strong association between microbiome composition and response to immunotherapy as seen previously in other cancers like melanoma (34–36) we sought to measure differences in the urinary microbial composition between pT0 Trial bladder cancer patients who showed complete response to neoadjuvant chemotherapy and those who did not. Figure 5A shows the microbial composition in each patient from both groups. There were no statistically significant differences in alpha diversity between responders and nonresponders ( Figure 5B ). PCoA did not reveal significant differences in the beta diversity of samples collected from responders and non-responders ( Figure 5C ). At the genus level, relative abundance of taxa showed differences between the two groups (Figure 5D) with Granulicatella (p<0.0001) and Proteus (p<0.001) enriched in non-responders (Figure 5E). LDA scores revealed abundance of E. Faecalis in responders ( Figure 5F ). Not enough patients received a gemcitabine-containing regimen to determine if Gammaproteobacteria presence was associated with response.

Antibiotic prophylaxis and therapy remain crucial in successfully managing various urological diseases, including infections, diagnostic procedures, surgical interventions, and transplants. Preprocedure antibiotics are required by the Department of Human Services in order for payment to occur to healthcare facilities and providers. As such, the urinary microbiome may be iatrogenically affected, but the effect of preoperative antibiotics has not been assessed. Further, conventional antibiotic therapy causes long-term alterations in the urinary microbiome, chronic recurring cystitis, and the development of multidrug resistant pathogens, often relating to prolonged episodes of infection (28). We sought to determine the effect of antibiotics on the urinary microbiome using samples collected from patients undergoing non-endourological procedures, as insertion of a foreign body into the urethra/bladder could confound the analysis. There was no statistically significant change in either alpha or beta diversity one month post antibiotic treatment ( Figures 6A, B ). Differential abundance was seen between taxa at the genus level pre- and post- antibiotic treatment (Figure 6C). Several alpha diversity metrics approached significant difference thresholds (5 of 6 metrics had p value of 0.051 or 0.054). This probably reflects a small sample size. LDA analysis, however, did reveal differential enrichment of taxa like Corneybacterium, Finegoldia, Varibaculum in samples obtained post antibiotic treatment (Figure 6D), which is consistent with the fastidiousness of Gram-positive organisms.

Matched diagnostic TURBT samples was also collected from the patients enrolled in the pT0 Trial, in addition to the urine samples collected pre- and post- chemotherapy. We sought to identify similarities/differences in the microbial composition between the pre-chemotherapy urine and TURBT samples from bladder cancer patients. Though the alpha diversity metrics measured by ACE, Chao1, and observed species were not significantly different between pre-chemotherapy urine and TURBT samples, we did observe a significant difference in evenness (Pielou, p=0.04) and diversity indices (Shannon, p=0.03; and Simpson, p=0.01), with the TURBT samples showing higher diversity and evenness compared to the urine samples ( Figure 7A ). PCoA analysis also revealed a clear separation between the microbial composition seen in urine and TURBT samples (p<0.001, Figure 7B ). Distribution of each genus in the prechemotherapy urine and matched tissue is illustrated in Figure 7C , with Escherichia being the most predominant single genus in urine but Bacteroides being the most predominant genus in tissue. In comparing similarities and differences, 80 taxa were found to be common between the two groups with 92 and 75 taxa unique to the pre-chemotherapy urine samples and TURBT samples, respectively ( Figure 7D ).

We also studied bladder microbial changes in mice as they undergo tumorigenesis in response to BBN (N-Butyl-N-(4-hydroxybutyl)nitrosamine) exposure. BBN is a metabolite found in tobacco smoke and is the most widely used carcinogen-induced animal model of bladder cancer. This model accurately represents histological and genetic features of human bladder tumors (37, 38). Bladder tissue samples were collected from mice at 6 weeks, 12 weeks and upon tumor formation as assessed by µCT excretory urography in mice that were given 0.01% BBN in their drinking water (Figures 8A, B). BBN was given till week 12 after which, mice were switched to regular drinking water. Age and sex-matched mice were given drinking water control from the same water source. Shannon diversity metric showed significant differences in microbial populations between baseline and 6 weeks of BBN treatment (p<0.05). However, similarities were also observed in mice that were given drinking water alone suggestive of change in background flora during this time ( Figure 8C ). Beta diversity measured by Bray Curtis dissimilarity index demonstrated that the bladder microbial composition significantly differed by timepoint (p=0.002) and treatment (p=0.004; Figure 8D ). At genus level, differences in bacterial composition between the 2 groups of mice could be observed (Figure 8E). LefSe analysis showed that increased abundance of Bifidobacterium (p<0.05) was associated with tumor presence. At the same time, genera Faecalibaculum, Acinetobacter and Bacteroides were associated with healthy bladder ( Figure 8F ). No differences were observed in alpha-diversity metrics between male and female mice within both treatment groups (data not shown). However, within the group of female mice, there was an overall trend of reduction in the diversity in the ‘BBN’ groups compared to water control while these differences do not exist in male mice (data not shown).