An expanded metabolic pathway for androgen production by commensal bacteria

Abstract

Commensal bacteria have been implicated in the modulation of steroid hormones, including circulating androgen levels in the host. However, the microbial genetic pathways involved in androgen production have not been fully characterized. Here, we identify a microbial gene encoding an enzyme that catalyzes the conversion of androstenedione to epitestosterone in the gut microbiome member Clostridium scindens, and named this gene desF. We demonstrate that epitestosterone impacts androgen receptor-dependent prostate cancer cell proliferation in vitro. We also demonstrate that stool desF levels are elevated in prostate cancer patients unresponsive to abiraterone/prednisone therapy. Bacterial isolates from urine or prostatectomy tissue produced androgens and 17β-hydroxysteroid dehydrogenase activity encoded by the desG gene was detected in strains of urinary tract bacterium, Propionimicrobium lymphophilum. Further, we demonstrate that urinary androgen-producing bacterial strains can promote prostate cancer cell growth through metabolism of cortisol and prednisone. Abiraterone, which targets host desmolase (CYP17A1), a rate limiting enzyme in adrenal steroidogenesis, does not inhibit bacterial desmolase (DesAB), whereas the conversion of prednisone to androgens by DesAB, DesF, and DesG drives AR-dependent prostate cancer cell line proliferation in vitro. Our results are a significant advance in steroid microbiology and highlight a potentially important role for gut and urinary tract bacteria in host endocrine function and drug metabolism.

Introduction

The human body is composed numerically of equal bacterial and mammalian cells¹, and that 99% of the functional genes are microbial². Accordingly, the functional capacity of the human-associated microbiota to influence human health and disease is immense. A growing body of literature implicates the gut microbiota in the modulation of circulating androgen levels in the host^3,4 For example, fecal microbiota transplant (FMT) of gut microbiota from adult male to immature female nonobese diabetic mice altered the recipient’s microbiota, resulting in elevated circulating testosterone (T), and robust type 1 diabetes protection³. Another study reported that FMT from metastatic castration resistant prostate cancer (mCRPC) donors promotes castration-resistant tumor growth in mice⁵, which is related to the microbial metabolism of host steroids to androgen precursors such as dehydroepiandrosterone (DHEA) and androgens (e.g., T and dihydrotestosterone (DHT)); however, the bacterial genes encoding the enzymes involved in androgen production were not determined^3–5. Elucidation of these pathways as well as their demonstrated relevance to androgen-mediated disease could offer novel approaches for therapeutic targeting.

We previously discovered a gut bacterial enzymatic pathway (desABCD operon) that is involved in the metabolism of cortisol derivatives in Clostridium scindens^6,7. We demonstrated that desC encodes steroid 20ɑ-hydroxysteroid dehydrogenase (20ɑ-HSDH) involved in side-chain oxidoreduction^6,8, while desAB encodes steroid-17,20-desmolase involved in side-chain cleavage of cortisol derivatives to 11-oxy-androgens^6,7. 11-oxy-androgens are now regarded as potent androgen receptor (AR) ligands (e.g., 11-keto-T and 11-keto-dihydrotestosterone (11-keto-DHT)) on par with T and DHT⁹. Androgen biosynthesis in the gut via bacterial species that carry the des operon may contribute to disease etiology, with prostate cancer as one exemplary paradigm. mCRPC is largely incurable and characterized by progressive metastatic cancer growth despite treatment that blocks androgen synthesis (e.g., gonadotropin-releasing hormone agonist/antagonist) in the testes. Second-line treatments that block adrenal androgen synthesis (e.g., abiraterone acetate (AA) given with the replacement glucocorticoid prednisone (P)), and/or directly antagonize AR (e.g., enzalutamide) are likewise not curative, and resistance invariably develops¹⁰. Despite castrate levels of circulating T in individuals with mCRPC, intratumoral levels of androgens remain high^11,12. Current research on the source(s) of these intratumoral androgens is almost entirely focused on host enzymatic biosynthesis and intracrine pathways through which androgen-precursors are synthesized and become AR-ligands^13–16. The capacity for androgen production by microbiota colonizing other host epithelial sites to contribute to disease etiology and therapeutic response is completely unexplored.

Herein, we report an expansion of the bacterial desmolase pathway in desAB-harboring taxa to include host-associated bacteria expressing 17ɑ-hydroxysteroid dehydrogenase (17ɑ-HSDH; desF) or 17β-HSDH (desG) involved in the formation of epitestosterone (epiT) and T, respectively. We also demonstrated that, contrary to the current dogma, epiT is an AR agonist that drives prolonged expression of AR-responsive genes and proliferation of prostate cancer cell lines. Moreover, we found that fecal desF-carrying bacteria are present in the gut microbiota of individuals with advanced prostate cancer, and that fecal desF levels are elevated in individuals with disease progression on androgen-deprivation therapy combined with AA/P versus individuals with hormone sensitive prostate cancer (HSPC) who are not undergoing treatment. We previously demonstrated that the genomes of urinary strains of Propionimicrobium lymphophilum harbor desAB¹⁷. In this study, we isolated additional strains of P. lymphophilum from human prostatectomy tissue and urine and determined that these isolates are capable of converting cortisol to 11β-hydroxy-androstenedione (11OHAD) as well as 11β-hydroxy-testosterone (11OHT) that drive hormone-responsive prostate cancer cell proliferation in vitro using a unique microbe-encapsulation technique. These findings significantly expand our knowledge of human steroid microbiology and the expanded steroid-17,20-desmolase pathways should be further investigated.

Results

A novel pathway for epiT formation by gut microbiota

C. scindens ATCC 35704 (Csci35704) expresses steroid-17,20-desmolase encoded by the desAB genes^6,7. Three decades ago, C. scindens VPI 12708 (Csci12708) was reported to convert androstenedione (AD) to epiT, indicating the presence of a gene encoding the enzyme 17ɑ-HSDH¹⁸. We reasoned that co-culture of both strains in the presence of 11-deoxycortisol (11DC) would yield epiT with a stable AD intermediate. As predicted, co-culture of Csci35704 and Csci12708 yielded the expected conversion of 11DC to both AD and epiT after 24 h (Fig.1a, b). Indeed, 11DC was depleted in 24 h, yielding 6.20 ± 0.22 μM AD and 27.42 ± 0.64 μM epiT. We confirmed the formation of epiT from AD in pure cultures of Csci12708 by a combination of high-resolution LC-MS, and NMR (Extended Data Fig. 1, Supplementary Table 1). The biosynthesis of AD and 11OHAD from cortisol and its derivatives by C. scindens strains is likely unique to host- associated bacteria¹⁹, and distinct from known adrenal biosynthetic pathways. Endocrine pathways in steroidogenic tissues (e.g., adrenal gland and gonads) convert pregnenolone via the 17,20-lyase activity of CYP17A1 coupled with P450 oxidoreductase (POR) and cytochrome b5 (CYB5) to DHEA, some of which is converted to AD via HSD3B2²⁰. CYP11B1 coupled with adrenodoxin (ADX) hydroxlates AD at C11 to yield 11OHAD¹⁹. While the metabolic pathway for epiT formation by human steroidogenic pathways is unknown, it is speculated to derive from 5-androstene-3β,17ɑ-diol rather than AD²¹.

Figure 1. Identification of the desF gene encoding steroid 17ɑ-hydroxysteroid dehydrogenase.

a, A proposed biochemical pathway by which Csci35704 and Csci12708 convert 11DC to AD and epiT. SEM images of Csci35704 and Csci12708 are included. b, LC-MS chromatographs of 11DC, AD, and epiT at time 0 and 24 h (left) and quantification of metabolites (right). Data are presented as mean ± SD (n=3 biological replicates). c, Venn diagram summarizing comparative genomic analysis between Csci35704 and Csci12708. d, Scatterplot of RNA-Seq analysis to identify genes upregulated by 50 μM 11OHAD in Csci12708. Significantly upregulated genes (3 log₂fold=3.069, FDR=0.012) in red, downregulated genes in blue, and not differentially regulated in black. e, Gene organization of 17ɑ-hydroxysteroid dehydrogenase candidate (“desF”) identified in RNA-Seq analysis. f, SDS-PAGE of Streptavidin-purified recombinant DesF and LC-MS chromatographs showing NADPH-dependent conversion of AD to epiT and NADP⁺-dependent conversion of epiT to AD at 0 h and 24 h. g, Ribbon diagram of AlphaFold 2 structural prediction of DesF (NADP⁺ in red and epitestosterone in yellow). h, Space-filling model of both the DesF (gray) and ligands NADP⁺ (red) and epitestosterone (yellow). i, Molecular dynamic trajectory analysis revealed strong interactions between ligands and catalytic triad.

The gut microbial desF gene encodes a novel 17ɑ-HSDH

We next sought to identify the gene(s) encoding 17ɑ-HSDH in Csi12708 responsible for catalyzing the conversion of AD to epiT. We performed comparative genomics between Csci35704 and Csci12708 to identify reductases unique to Csci12708. The strains share 35% of their genes (1916 ORFs) while 33% of genes (1,800 ORFs) are unique to Csci12708 (Fig. 1c). We narrowed this list down to three protein families known to include HSDH enzymes: 25 belonging to the short chain dehydrogenase/reductase (SDR) family; 23 to the medium chain dehydrogenase/reductase (MDR) family; and 2 to the aldo-keto reductase (AKR) family²². Of these, 18 SDR, 18 MDR, and 2 AKR proteins are unique to Csci12708 (Extended Data Fig. 2).

Given this relatively large number of candidates and the reported steroid-inducibility of 17ɑ-HSDH¹⁸, we opted to utilize genome-wide transcriptomics to identify candidates after the growth of Csci12708 in the presence of 11OHAD vs. uninduced controls. The addition of 11OHAD significantly upregulated the expression of a single gene (GGADHKLB_00774) (Fig. 1d, Supplementary Table 2). This observation was reproducible since repeat transcriptome analysis yielded the same overall result (Supplementary Fig. 1, Supplementary Table 3). GGADHKLB_00774 is predicted to encode a 27-kDa short chain dehydrogenase/reductase (SDR) superfamily protein (Fig. 1e). Importantly, GGADHKLB_00774 is in the list of reductases found in Csci12708 but not in Csci35704 (Extended Data Fig. 2). We overexpressed GGADHKLB_00774 and observed this enzyme partially converted AD to epiT in the presence of NADPH (but not NADH), and epiT was converted to AD in the presence of NADP⁺ (not NAD⁺) (Fig. 1f, Supplementary Figure 2), establishing that GGADHKLB_00774 from Csci12708 encodes a novel NADPH-dependent 17ɑ-HSDH in the SDR superfamily²². We propose the name desF for this gene.

To investigate the predicted molecular mechanism of ligand binding in the enzyme DesF we utilized AlphaFold 2²³ through its QwikFold interface in VMD²⁴. The predicted structure was used to fit an NADP⁺ molecule into the binding pockets of DesF monomer. The ligand structure was minimized in the pocket along with nearby protein residues, while most of the protein structure remained static. The structure of epiT in DesF was also fitted to the most probable binding site to minimize the structure of the complex⁸. The complex was solvated and subjected to a 100-ns-equilibrium MD simulation. AI-based predictions and MD simulations showed that both epiT (yellow) and NADP⁺ (red) occupy a mostly open cleft (Fig. 1g, h). Here, we observed that epiT and NADP⁺ remained stable in the predicted catalytic cleft. Analysis of the MD trajectory revealed strong interactions between the ligands and serine, tyrosine, and lysine residues in DesF. SER144 and TYR157 formed stable interactions with the epiT molecule (Fig. 1i). Additionally, LYS161 was identified by our network analysis as a key residue in stabilizing NADP⁺ in its position.

Phylogenetic analysis indicates that DesF may be unique to C. scindens strains, with these strains in a cluster well separated from the other sequences analyzed (Supplementary Figure 3). Accordingly, amino acid identity drops precipitously to at most ~50% identity in SDR family proteins in other taxa. DesF shares only 20.2% amino acid identity with Mus musculus 17ɑ-HSDH, which is in the aldo-keto reductase family (AKR)²⁵. We thus sought to determine the proportion of C. scindens strains that harbor desABC and desF genes. We sequenced the genomes of 14 C. scindens strains and obtained an additional 20 C. scindens genomes from NCBI^26–28. Interestingly, two clades of C. scindens strains were apparent in whole genome phylogeny (Extended Data Fig. 3a) with the two most studied strains, Csci35704 (Clade 1) and Csci12708 (Clade 2), being distinct. In Clade 1, 10 strains harbored desABC genes, with desF being absent. In Clade 2, 16 of 19 strains encode desF, and two strains, SL.1.22 and S076 had both desABC and desF genes. To verify that CsciS076 is indeed capable of both steroid-17,20-desmolase and 17ɑ-HSDH activities, we cultivated this strain in the presence of 11DC. As predicted, we observed the formation of both AD and epiT after 24 h incubation (Extended Data Fig. 3b). We then constructed metagenome-assembled genomes (MAGs) from publicly available human metagenomes resulting in 225 C. scindens MAGs. In the C. scindens MAGs, 20 had desABC, 97 had desF, and 4 had both desABC and desF (Supplementary Table 4).

EpiT functions as an AR agonist of prostate cancer cells

EpiT is regarded as an “antiandrogen” that is expected to bind to and antagonize AR and reduce prostate cancer growth²⁹. However, this dogma has been challenged with a recent study indicating that epiT serves as a partial AR agonist in a reporter cell line³⁰. Circulating epiT is measured in the low nanomolar concentrations, with epiT/T ratios of 0.1 for females and up to 1 for males³¹. However, little evidence in the literature has examined epiT for its potential to alter cell physiology via nuclear AR²⁹. We compared the growth of androgen-sensitive prostate cancer cells (LNCaP) in the presence of 10 nM AD, T, and epiT and vehicle control (VC; 0.5% v/v methanol) overtime. Both T and epiT caused significant proliferation relative to VC and AD. Interestingly, epiT caused significantly higher proliferation than T overtime (Fig. 2a-c). To determine whether epiT-induced proliferation requires AR agonism, we treated LNCaP cells with 2.0 μM enzalutamide, an AR competitive inhibitor and prostate cancer drug³². Treatment with enzalutamide caused consistent and significant growth inhibition of LNCaP cells in the presence of 10 nM T and epiT at all time points, indicating that proliferation in all cases was AR-dependent (Fig. 2a-c). We confirmed AR-dependent gene expression through measurement of the AR downstream gene target prostate specific antigen (PSA)³³. PSA gene expression is elevated to 31.60 ± 4.07-fold in the presence of T at 24 h, but dropped significantly to 4.88 ± 1.15-fold at 48 h and to 1.68 ± 0.33-fold by 96 h. In contrast to T, epiT caused prolonged PSA gene expression throughout the time course. In all cases, PSA gene expression is significantly reduced by enzalutamide treatment (Fig. 2d-f). Similar results were observed with expression of another AR-regulated gene, transmembrane protease, serine 2 (TMPRSS2, Fig. 2g-i).

Figure 2. Epitestosterone and gut microbial desF in prostate cancer.

a-c, Time-dependent proliferation (MTS assay) of LNCaP cells in the presence (2 μM) or absence of AR-antagonist, enzalutamide. Proliferation was measured at 24, 48, or 96 h in the presence of 10 nM steroids or VC (n=3 independent biological replicates (each with 6 technical replicates)). Time-dependent qRT-PCR quantification of KLK3 gene (d-f) and TMPRSS2 gene (g-i) after treatment with 10 nM T, epiT, or VC at 24, 48, 96 h in the presence or absence of 2 μM enzalutamide Data are presented as mean ± SD (n=3 independent biological replicates (each with 3 technical replicates)). Cell proliferation and gene expression data for each group at each time point are normalized to VC. P values (a-f) were calculated by unpaired t-test (two-sided) and Benjamini-Hochberg correction. j.Schematic representation of clinical study examining fecal desF quantification (qPCR) in 25 individuals with hormone sensitive prostate cancer (HSPC) who were not undergoing any treatment and 44 individuals with advanced prostate cancer undergoing treatment with AA/P either stable (blood PSA trend stable, n=28 samples) or progressing (blood PSA trend increasing, n=28 samples). k-l. Scatterplots of desF % normalized to total fecal 16S rRNA gene (k) or desF copy number/10 ng DNA(l) in HSPC controls and stable vs. progressing AA/P treated patients. Significance in desF was observed between HSPC controls and the progressing group (desF %, P=0.021; desF copy, P=0.007). m-n. Fecal desF quantification (n=12 samples) of donor samples taken both while they were stable on AA/P and when they were progressing. Three of these donors were negative for desF in both the stable and progressing samples. P values (k-n) were calculated by Kruskal-Wallis one-way ANOVA (two-sided) and corrected for multiple comparisons using Dunn’s test. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001.

To confirm the proliferation observed in LNCaP cells from the MTS assay, we also employed live cell imaging analysis with LNCaP cells. Indeed, we observed significant AR-dependent growth in the presence of T, epiT, and 11OHT over time (Supplementary Figure 4). Since LNCaP cells express a mutant AR with a broadened steroid-binding capacity³⁴, we repeated the MTS assay with androgen-responsive VCaP cells that express wild type AR³⁴. Significant proliferation was also observed with 10 nM AD, T, epiT, and 11OHT after 2, 4 and 8 days of exposure (Extended Data Fig. 4). Taken together, our results strongly indicate that epiT exerts androgenic function not recognized previously. A study published recently indicated that 5α-dihydroepitestosterone (epiDHT) was capable of AR agonism³⁰. Since LNCaP cells express steroid-metabolizing enzymes (e.g., 11β-HSD1/2 and SDR5A1,2)³⁵, it is possible that epiT may be converted to a more potent agonist such as epiDHT. To test for this possibility, we repeated treatment of LNCaP cells with 10 nM T, DHT, epiT, and epiDHT in the presence or absence of the 5α-reductase inhibitor, dutasteride³⁶. In all cases, dutasteride had no measurable effect on proliferation. Interestingly, epiT caused significantly higher proliferation than epiDHT (Supplementary Fig. 5).

Fecal desF gene levels in prostate cancer patients

We previously reported that prednisone (a replacement glucocorticoid commonly given in combination with AA) can be converted to 1,4-androstadiene-3,11,17-trione (AT) by DesAB¹⁷. In the context of AA/P treatment, prednisone (rather than cortisol) may be the more relevant substrate for microbial androgen biosynthesis. We thus sought to determine whether desF expressing bacteria convert AT to epiAT, and whether treatment with epiAT would lead to prostate cancer cell proliferation. We cultivated Csci12708 in the presence or absence of 50 μM AT, diluted the 0.2 μm-filtered, 72-h spent culture medium, and added it to LNCaP cell culture (Extended Data Fig. 5a, c). We demonstrated that spent medium in which Csci12708 converted AT to epiAT resulted in proliferation of LNCaP cells. We also found that proliferation was ablated by enzalutamide treatment, indicating that epiAT triggered proliferation through AR-dependent signaling (Extended Data Fig. 5d).

We next determined whether bacteria carrying the desF gene are present in the gut microbiota of individuals with hormone sensitive prostate cancer (HSPC) versus those undergoing treatment for advanced prostate cancer with AA/P therapy, and whether quantitatively, fecal desF levels correlate with response to AA/P (Fig. 2j). Quantitative PCR (qPCR) primers were designed to target the desF gene. We found that 76% of the HSPC donors (19/25) and 84.1% of the AA/P donors (37/44) had detectable fecal desF. Interestingly, both the percentage of desF normalized to total 16S rRNA (as a surrogate for total bacterial load) as well as the absolute copy number of fecal desF were significantly elevated in samples taken while individuals were progressing on AA/P versus the HSPC controls (Fig. 2k, l). This relationship was not observed in samples taken during AA/P response (stable). Twelve of the donors had samples taken both while they were stable on AA/P and when their disease was progressing. Three of these donors were negative for desF in both the stable and progressing samples. We found that a subset of the remaining nine individuals had substantial increases in fecal desF levels during PSA progression on AA/P (Fig. 2m, n). These results may suggest a potential desF-mediated androgen biosynthesis by the gut microbiota (e.g., epiT) and/or metabolism of a replacement glucocorticoid (e.g., prednisone).

Formation of androgens by P. lymphophilum strains

We previously determined that some strains of P. lymphophilum, a normal inhabitant of the urinary tract, can harbor desAB genes¹⁷. We therefore sought to further explore microbial steroid biotransformations by urinary tract isolates. After screening pure cultures established in a prior collection of bacterial isolates from prostatectomy tissue³⁷, we identified and cultured a strain of P. lymphophilum (strain API-1) capable of steroid metabolism (Fig. 3a, b; Extended Data Figure 6). We speculate that strain API-1 could have derived from prostatic fluid, or it could have been present in the prostatic urethra. Strain API-1 was found to generate products from cortisol consistent with 11OHAD and 11β-hydroxytestosterone (11OHT) that co-migrated with authentic standards (Fig. 3c).

Figure 3. Androgen-producing bacterium isolated from human prostatectomy tissue expresses a multi-step pathway for conversion of glucocorticoids to derivatives of testosterone.

a-c, Propionimicrobium lymphophilum API-1 was isolated from a post-surgical prostatectomy specimen. Strain API-1 was placed in a tube of thioglycolate broth, and cultured anaerobically at 37°C. LC-MS analysis showed that this strain converted cortisol to 11OHAD and 11OHT. SEM image of the strain was obtained and the whole genome was sequenced. d, A gene predicted to encode “3β-hydroxycholanate dehydrogenase” was identified in the genome sequence and selected as a potential 17β-HSDH candidate. e, We cloned JGFJDHCB_01426 and overexpressed the recombinant protein for enzyme assay. f, LC-MS analysis of enzyme assay steroid products observed that 11OHAD was converted to 11OHT. We proposed the name desG for JGFJDHCB_01426. g, Pathway for conversion of cortisol to 11OHAD and 11OHT by P. lymphophilum API-1. h, Ribbon diagram of AlphaFold 2 structural prediction of DesF (NADP⁺ in red and testosterone in blue). i, Space-filling model of both the DesG (gray) and ligands NADP⁺ (red) and testosterone (blue). j, Molecular dynamic trajectory analysis revealed strong interactions between ligands and catalytic triad. k, The product of cortisol metabolism by strains API-1, 11OHT, causes significant LNCaP cell proliferation relative to vehicle control (VC; methanol 0.5% v/v). Data are presented as mean ± SD (n=3 independent biological replicates (each with 6 technical replicates)). Cell proliferation data for each group at each time point are normalized to VC. P values were calculated by unpaired t-test (two-sided) and Benjamini-Hochberg correction. * P < 0.05, ** P < 0.01, *** P < 0.001.

We then sequenced and assembled the 2.3 Mb genome of P. lymphophilum API-1 (Fig. 3d; Supplementary Table 5-7). The desABE operon was located, providing an enzymatic basis for the conversion of cortisol to 11OHAD. We then searched for candidate 17β-HSDH genes in P. lymphophilum API-1, and a single gene (JGFJDHCB_01426) in the SDR family appeared most probable based on genomic context (monocistron), as we have observed for HSDHs previously³⁸. This gene is annotated as a “3β-hydroxycholanate dehydrogenase” involved in bile acid metabolism, suggesting that the enzyme may metabolize steroids. This gene was cloned into pET51b+, overexpressed and affinity purified from E. coli BL21(DE3)RIPL. The Strep-Tactin affinity purified 26.8 kDa recombinant enzyme was incubated with NADPH and 11OHAD with 11OHT formed (Fig. 3e, f). The capacity of this urinary bacterial 17β-HSDH to convert 11OHAD to 11OHT differs from human 17β-HSDH isomers which do not recognize 11OHAD as a substrate^35,39,40. This 17β-HSDH gene is part of the steroid-17,20-demolase pathway in P. lymphophilum API-1, so we propose to name this gene desG (Fig. 3g).

The structure of DesG was predicted in a manner similar to DesF, allowing us to minimize the ligand structure of NADP⁺ into the predicted binding pocket. The structure of testosterone in DesG was also fitted to the most probable binding site. In contrast to the open pocket observed with DesF, testosterone (blue) and NADP⁺ (red) occupy a pocket that closes over the catalytic region in DesG (Fig. 3h, i). In DesG, LYS161 is predicted to play a role in stabilizing NADP⁺, while SER144 and TYR157 formed stable interactions with the testosterone molecule (Fig. 3j). Simulations indicated that the predominantly hydrophobic clefts in both DesF and DesG are crucial for maintaining the complex stability, with steroid molecules fluctuating minimally in the cleft. We propose that the serine and tyrosine residues stabilize the steroid to initiate the enzymatic reactions, while the lysine residue helps hold NADP⁺ in place in both DesF and DesG. Although both enzymes share the same catalytic triad, we demonstrated the specificity of DesF for 17α-oxidoreduction and DesG for 17β-oxidoreduction (Supplementary Fig. 6).

RNA-Seq analysis was performed in the presence of cortisol or 11OHAD vs. DMSO vehicle control with P. lymphophilum strain API-1. We did not observe differential gene expression between cortisol treatment and vehicle control with respect to desA (JGFJDHCB_01528; 0.70 log₂FC; 0.16 FDR), desB (JGFJDHCB_01529; 0.78 log₂FC; 0.099 FDR), or desG (JGFJDHCB_01426; 0.26 log₂FC; 0.63 FDR) (Supplementary Table 8). These results indicate that expression of steroid-17,20-desmolase genes in P. lymphophilum is not regulated by steroids, as in the GI tract⁶, but rather the genes are constitutively expressed. This form of regulation may be important in the urinary tract due to the nanomolar levels of urinary steroids¹⁸ whose quantities are likely insufficient for sensitive inducible systems to evolve.

Protein phylogeny of the amino acid sequence of DesG revealed other taxa isolated in the urinary tract, but whose steroid metabolism remains unknown (Supplementary Fig. 7). Such sequences, which display >77% ID with the DesG from P. lymphophilum, include Arcanobacterium urinimassiliense (BQ7117_RS04815), Vaginimicrobium propionicum (CZ356_RS01445), and Propionimicrobium sp. BV2F7 (HMPREF1255_RS00895). Phylogeny and SSN analysis of DesAB from the gut bacterium Csci35704 led to identification of A. urinimassiliense as a desAB-harboring urinary bacterium¹⁷. We expressed the synthesized genes predicted to encode 17β-HSDH in these species in E. coli BL21(DE3) (Extended Data Fig. 7a-c). Only CZ356_RS01445 from V. propionicum which shares 88% amino acid identity with DesG had 17β-HSDH activity (Extended Data Fig. 7d). To determine androgenicity of 11OHT, we measured time-dependent proliferation of LNCaP cells. Compared to VC, 11OHT caused significant and prolonged (96 h) growth of LNCaP cells (Fig. 3k). Similar proliferation of VCaP cells was observed in the presence of 11OHT relative to VC (Extended Data Fig. 4).

A urine sample from the same male patient from whom we isolated P. lymphophilum API-1 from prostatectomy tissue was obtained to determine long-term colonization by strains similar to or evolved from P. lymphophilum API-1. This urine sample was collected approximately 17 years after prostatectomy (Fig. 4a). Upon isolating colonies from the urine sample, we screened for cortisol metabolism and obtained a bacterium capable of producing both 11OHAD and 11OHT (Fig. 4b). We sequenced and completed the 2.1 Mb genome of this isolate and named it P. lymphophilum API-2 (Supplementary Table 5-7). Comparative genome analysis indicated an average nucleotide identity (ANI) of 98.9% to API-1 with both strains sharing 72.0% of their predicted protein coding genes (Fig. 4c, d). These results indicate that some individuals experience long-term colonization of androgen-producing urinary tract bacteria.

Figure 4. Urinary tract isolates of Propionimicrobium lymphophilum drive prostate cancer cell growth through androgen biosynthesis.

a, Strain API-1 was isolated from prostatectomy tissue, and strain API-2 was isolated ~17 years later from a urine sample collected from the same patient. b, LC-MS analysis confirmed the conversion of cortisol to 11OHAD and 11OHT by strain API-2.c, d, Synteny and comparison of gene content between the two strains indicate a high degree of similarity. e, Schematic representation of urinary tract isolates encapsulated in microgels which are co-cultured in culture medium with LNCaP cells in the presence of cortisol. f, Micrographs of calcium alginate microgels at Day 0 and Day 2 display dense growth of API-2 in DMEM medium under aerobic conditions. g, LC-MS analysis confirmed the conversion of cortisol to 11OHAD and 11OHT in encapsulated beads. Data are presented as mean ± SD (n=3 independent biological replicates (each with 6 technical replicates)). h, LNCaP proliferation in the presence/absence of API-2 and/or 10 nM cortisol. Data are presented as mean ± SD (n=3 independent biological replicates (each with 6 technical replicates)) Cell proliferation data for each group at each time point are normalized to VC. P values were calculated by unpaired t-test (two-sided) and Benjamini-Hochberg correction, * P < 0.05, ** P < 0.01, *** P < 0.001.

P. lymphophilum promotes prostate cancer cell proliferation

We recently developed and reported a microbe-encapsulation technique using calcium alginate beads to co-culture bacteria and host cells allowing metabolic interaction without direct contact (Fig. 4e)^41,42. We first established the growth of P. lymphophilum API-2 in this platform (Fig 4f). LNCaP cell proliferation was determined in the co-culture platform in the presence of the bead encapsulated strain API-2 +/− 10 nM cortisol. To confirm the metabolism of cortisol by strain AP1–2, steroids were extracted from the spent medium and quantified (Fig. 4g). Significant proliferation was observed only in the presence of strain API-2 and cortisol (Fig. 4h). We then aimed to determine if the side-chain cleavage product of prednisone (AT) would be converted to the 17β-reduced metabolite (Δ^1,4-testosterone; ATT). We observed almost complete conversion of AT to ATT by strain API-2 after 24 h incubation (Extended Data Figure 5b). Treatment of LNCaP cells with ATT led to significant proliferation, and enzalutamide was less effective in preventing cell growth (Extended Data Figure 5e). These results indicate that P. lymphophilum generates androgens that drive proliferation of prostate cancer cells using both endogenous (e.g., cortisol) and therapeutic (e.g., prednisone) glucocorticoid sources.

Culturing of androgen-forming taxa in male human urine

To identify additional cortisol side-chain cleaving urinary microbial taxa, we collected clean catch urine from 25 patients during a pre-biopsy visit to Carle Hospital Oncology, and clean catch urine from 14 age-matched healthy controls. We screened urine samples for conversion of cortisol to 11OHAD (desAB activity) or 11OHT (desAB and desG activity), and identified 11OHAD and/or 11OHT biosynthesis in cultured urine from 8 of 25 pre-biopsy samples, 4 of which were subsequently diagnosed with prostate cancer, and 2 of 14 healthy control samples. Of the samples collected, 10 urine samples tested positive for 11OHAD and/or 11OHT biosynthesis and were used for P. lymphophilum strain isolation. Colonies were cultivated in the presence of 11DC in order to screen for desAB function, or 11OHAD to screen for desG function (Fig. 5a, Supplementary Table 9). We obtained 9 P. lymphophilum isolates from urine samples and sequenced their genomes (Extended Data Figs. 6, 8). We identified the desABE genes encoding bacterial desmolase (desAB) and steroid 20β-HSDH (desE)^38,43 in all strains of P. lymphophilum. Strikingly, 6 of 9 strains of P. lymphophilum also had 17β-HSDH activity involved in conversion of 11OHAD to 11OHT. The desABE and desG genes and their metabolic activities in strain API-2 are consistent with that in strain API-1 (Fig. 5b-d).

Figure 5. Isolation and characterization of androgen-producing bacteria from human male urine samples.

a, Schematic of isolation and screening approach from human male urine. b, Biochemical pathway and DesABEG enzyme functions proposed for urinary tract isolates. c-d, Organization of desABE and desG genes corresponds with the formation of 11OHAD and 11OHT in pure cultures incubated with 50 μM cortisol.

Abiraterone does not inhibit androgen biosynthesis by DesAB

Host steroid-17,20-desmolase (CYP17A1) is an NADPH and O₂-dependent P450 monooxygenase that facilitates adrenal corticosteroid and androgen biosynthesis through the 17α-hydroxylation and 17,20-lyase reactions that convert pregnenolone to DHEA and AD²⁰. AA is used to treat prostate cancer through the inhibition of CYP17A1 which blocks adrenal androgen synthesis. By contrast, bacterial steroid-17,20-desmolase (DesAB) functions under anaerobic conditions, through a predicted vitamin B1-dependent manner⁷, and may continue to function in the presence of AA. The efficacy of AA/P therapy may be further reduced as prednisone is side-chain cleaved and converted to epiAT and ATT. We therefore tested abiraterone for inhibition of the bacterial desmolase by pre-treating early log-phase cultures of C. scindens ATCC 35704 or P. lymphophilum API-1 in the presence of cortisol. After 72 h of growth, Csci35704 depleted cortisol with an equimolar amount of 11OHAD produced in both the absence and presence of abiraterone. P. lymphophilum converted cortisol to 11OHAD and 11OHT in both the absence and presence of abiraterone, and no differences were found between the groups (Extended Data Fig. 9). These results demonstrate that the side-chain cleavage of glucocorticoids by desmolase-positive bacteria in the GI and urinary tracts is not inhibited by the drug that blocks host androgen formation, indicating that bacterial metabolism of endogenous (e.g., cortisol) and pharmaceutical (e.g., prednisone) glucocorticoids is a potential source of androgens during androgen deprivation therapy combined with AA.

Discussion

This study significantly advances our understanding of the genetic potential of host-associated microbiota to produce androgens. We discovered desF and desG genes that expand the steroid-17,20-desmolase pathways that already include the side-chain cleavage enzyme (DesAB) and the side-chain oxidoreductases 20ɑ-HSDH (DesC) and 20β-HSDH (DesE)^6,7,38,43. Specifically, our results reveal that (1) a novel bacterial pathway for conversion of AD (or cortisol side-chain cleavage products) to epiT is encoded in the gut microbiome; (2) the end-product, epiT, activates AR-dependent growth of LNCaP cells to a greater extent than T, indicating that epiT is a currently unrecognized AR agonist. Our results indicate that conversion of epiT to epiDHT is not required for the observed proliferation. Moreover, mechanisms that inactivate AR-ligands such as T may not be targeting 17ɑ-reduced steroids. Previous studies have shown that UGT2B17 and UGT2B15 knockout in LNCaP cells significantly prolongs androgen signaling⁴⁴. While beyond the scope of this study, it is possible that glucuronidation of epiT by UGT2B17 and UGT2B15 is greatly reduced relative to T. Moreover, we cannot rule out intracrine mechanisms that convert added steroid substrates to more potent AR ligands in prostate cancer cell lines. Controversy exists among endocrinologists regarding 11OHT as an AR ligand^35,45,46, and it is possible that 11OHT was converted to 11KT intracellularly through the action of 11β-HSD1/2 isoforms³⁵ (3) abiraterone inhibits host biosynthesis of adrenal androgen precursors (e.g., AD, DHEA, and 11OHAD) but not bacterial desmolase (desAB genes); (4) the desF gene is enriched in the fecal microbiota of individuals with advanced prostate cancer with disease progression on AA/P; (5) enzymes in the steroid-17,20-desmolase pathway may represent potential therapeutic targets for the treatment of prostate cancer in the same manner as the host steroidogenic enzymes are for drug targets, which needs to be further confirmed in the future.

Moreover, we demonstrate that urinary tract bacteria, including a prostate tissue isolate, encode both desAB and the newly discovered desG gene that convert glucocorticoids (including P) to T derivatives that promote prostate cancer cell proliferation. Urine is the main route of glucocorticoid excretion in humans, and glucocorticoids are measured in urine on the order of hundreds of nanomolar²⁰. Based on this, we predict that it is possible that bacterial androgen biosynthesis occurs locally in the prostatic urethra. How or whether this androgen biosynthesis can influence the prostate is beyond the scope of the current study, but suggests the need for future research. Intriguingly, studies have shown that P. lymphophilum abundance in urine is associated with prostate cancer^47,48. We speculate that long-term colonization of the urinary tract by androgen-producing bacteria may be an underrecognized promoter of the development and/or progression of prostate cancer in some individuals (Extended Data Fig. 10). Further clinical and mechanistic microbiome studies examining the role of androgen producing bacteria in the urinary tract in primary prostate cancer is warranted. The role of androgen producing bacteria in other human diseases driven by androgens in males such as benign prostate hyperplasia and in females such as breast cancer⁴⁹ and polycystic ovary syndrome (PCOS)⁵⁰, deserves to be examined.

Methods

Bacteria, cell cultures, and chemicals

Bacteria.

Clostridium scindens ATCC 35704 (Csci35704), Clostridium scindens VPI 12708 (Csci12708) and Clostridium scindens S076 (Csci076) were derived from in house 30% glycerol stock cultures and cultivated in anaerobic Trypticase Soy Broth (TSB) at 37°C. Individual colonies were picked from anaerobic TSB plates, DNA extracted, and identity confirmed by 16S rRNA gene sequence and confirmatory PCR targeting baiJ (Csci12708) or desA (Csci35704). Propionimicrobium lymphophilumstrain API-1 was isolated from prostatectomy tissue. Propionimicrobium lymphophilum strains API-2, CFH07, CFH08, CFH13, CFH17, CFH18, CFH22, ANSC4 and ANSC13 were isolated from urine.

Cell culture.

LNCaP (CRL-1740) and VCaP (CRL-2876) cells were obtained from ATCC and cultured in RPMI 1640 medium (Corning 10–040) and DMEM (ATCC 30–2002), respectively, and supplemented with 10% Fetal Bovine Serum (FBS) from GIBCO. Additionally, RPMI was supplemented with 10 mM HEPES buffer (Corning), 1 mM sodium pyruvate (Corning) and 4.5 g/L D-glucose (Sigma). For the androgen treatment experiments, cells were starved of androgens by cultivation in charcoal stripped FBS (GIBCO) medium. Cell passage number was kept below 25 for all experiments. Both cell lines were tested for mycoplasma contamination and authenticated using short tandem repeat STR profiling in TEP facility (Cancer center at Illinois).

Chemicals.

Chemicals (commercial sources) included: cortisol (Sigma); 11β-hydroxy-androstenedione (11OHAD, Steraloids, Newport, RI, USA);11β-hydroxy-testosterone (11OHT, Steraloids); Epi-testosterone (epiT, Steraloids); 17-epi-dihydrotestosterone (epiDHT, Steraloids); 11-deoxycortisol (11DC, Sigma); Androstenedione (AD; Sigma); Testosterone (T, Sigma); 1,4-androstadiene-3,11,17-trione (AT, Sigma); Dihydrotestosterone (DHT, Sigma); 11-deoxycortisol-D5 (2,2,4,6,6-D5) (11DC-D5, Sigma); Androstene-3,17-dione-2,3,4–13C3 solution (AD-13C3, Sigma); Testosterone-D3 (16, 16, 17-d3) solution (T-D3, Sigma); 17-epi-testosterone-D3 (epiT-D3, Santa Cruz); Cortisol-D4 (9,11,12,12-D4) (Sigma); 11OHAD-D4 (9,11,12,12-D4) (Sigma); Enzalutamide (Selleck); Abiraterone (A, MedChemExpress); Dutasteride (Sigma), T, and AD were purchased in solution form (Sigma), evaporated with nitrogen, and redissolved in DMSO at the required concentration. EpiT, cortisol, 11OHAD, 11OHT, P, and AT were dissolved in DMSO or methanol. Enzalutamide and dutasteride were dissolved in DMSO.

Bacterial media preparation

Brain Heart Infusion (BHI, BBL) broth was purchased and prepared based on the instructions. The Trypticase Soy Broth (TSB, BBL) was prepared as instructed with the addition of 5 g yeast extract, 1 g L-cysteine, 1 mg resazurin and 40 mL salt solution (1 L; 0.25 g CaCl₂•2H₂O, 0.5 g MgSO₄•7H₂O, 1 g K₂HPO₄, 1 g KH₂PO₄, 10 g NaHCO₃, 2 g NaCl). Peptone Yeast Glucose (PYG) broth (modified) was prepared according to the DSMZ protocol. Blood agar base (Sigma) was purchased and prepared according to the instructions with 6% (v/v) defibrinated sheep blood (Thermo Scientific) added. Schaedler agar (Sigma) was purchased and prepared as instructed. Columbia broth (BBL) was purchased and prepared as instructed with 15 g/L agar added to solidify the medium. The broth was made anaerobic by storage in Hungate tubes with 100% N₂ in the headspace. Plates were made anaerobic by storage in an atmosphere of 85% N₂:10% CO₂:5% H₂.

Whole-cell steroid conversion assay

Csci35704 and Csci12708 were precultured in TSB. Afterwards, each fresh culture of these two strains was inoculated (0.5 mL) into fresh TSB with 50 μM 11DC. P. lymphophilium strains were cultured in anaerobic PYG broth. Log-phase cultures were transferred to fresh PYG broth containing 50 μM cortisol and incubated. For sampling, 1-mL samples were collected at each time point. The collected samples were clarified by centrifugation (13,300 × g, 10 min; Thermo Scientific) and the supernatant fluid was used for subsequent analysis.

Steroid extraction

Two parts ethyl acetate and 1-part bacterial culture supernatant were thoroughly mixed by vortexing for 1 min. Next, the ethyl acetate layer was carefully collected and transferred to new tubes. The extraction process was repeated, and the collected top layers were evaporated with nitrogen gas and dissolved in 200 μL LC-MS grade methanol. For samples in which 11DC, AD and epiT were to be quantified, an internal standard mixture of 11DC-D5, AD-13C3, epiT-D3 was added before extraction. For samples where cortisol,11OHAD and 11OHT were to be quantified, an internal standard mixture of cortisol-D4 and 11OHAD-D4 was added before extraction. The concentrations of substrates and end products were normalized based on the under-curve area of the internal standards accordingly.

Liquid chromatography-mass spectrometry (LC-MS)

Samples were sent to the Mass Spectrometry Lab (University of Illinois at Urbana-Champaign, Urbana, Illinois, USA) for metabolite analysis using liquid chromatography-mass spectrometry (LC-MS). LC-MS for all samples was done on a Waters Aquity UPLC coupled with a Waters Synapt G2-Si ESI MS (Waters Corp., Milford, MA, USA). Chromatography was performed using a Waters Acquity UPLC BEH C18 column (1.7 μm particle size, 2.1 mm x 50 mm) at a column temperature of 40°C and an injection volume of 0.5 μL. For gradient elution, 2 mobile phases were used: mobile phase A contained 95% water, 5% acetonitrile, and 0.1% formic acid; mobile phase B contained 95% acetonitrile, 5% water, and 0.1% formic acid. Initially, mobile phase A was 100% for 0.5 min. Over the next 5.5 min, mobile phase B linearly increased, reaching 70% at 6 min. Then, mobile phase B increased to 100% in 1 min and maintained for 1 min. Afterwards, a steep reversal to the initial conditions occurred within 0.1 min, and the running condition was maintained until the end at 10 min. The flow rate was 0.5 mL/min. The LC eluents were introduced into the mass spectrometer equipped with electrospray ionization (ESI) with a positive ion mode for steroid analysis. The following optimized conditions were used: capillary voltage of 3 kV, desolvation temperature of 500 °C, cone voltage of 25 V, collision energy of 4 eV, collision gas helium, source temperature of 120°C, cone gas flow of 10 L/h, and desolvation gas flow of 800 L/h. The mass range was 50–2000 Da. Mass Lynx v4.1 (Waters) was used for chromatographs and mass spectrometry data analysis.

RNA-Seq analysis

Csci12708 was cultivated in BHI broth and P. lymphophilum API-1 was cultivated in PYG broth in the presence/absence of 50 μM of 11OHAD at 37°C for 24 h. Cultures (10 mL) were pelleted by centrifugation (4,000 x g). RNA was extracted as previously described⁶. Samples were sent to Roy J. Carver Biotechnology Center, DNA Services Laboratory (University of Illinois at Urbana-Champaign, Urbana, Illinois, United States) for library construction and sequencing. Read quality was evaluated using FastQC v0.11.8⁵¹. SeqKit v2.0.0⁵² was used to calculate the read number, sum of the read length, minimum read length, average read length and maximum read length for each sequencing file. Trimmomatic v0.39 was used to remove the adaptors and low-quality reads⁵³. SortmeRNA v4.3.6 was used to filter out ribosomal RNAs⁵⁴. Salmon v0.14.1 was used to do gene quantification⁵⁵ using C. scindens VPI 12708/P. lymphophilum API-1 genome as a reference²⁷. Gene abundance was filtered and normalized using the edgeR package⁵⁶. Differential gene expression analysis was performed using the limma package⁵⁷.

Genomic DNA isolation

Genomic DNA of C. scindens ATCC 35704, C. scindens VPI 12708 and P. lymphophilum API-1 were extracted using the QIAamp PowerFecal Pro DNA kit (Qiagen) according to the manufacturer’s instructions. The extracted DNA was used to amplify the target genes for the heterologous expression of the potential candidates.

Heterologous expression and purification

The target inserts were amplified using the primers synthesized by Integrated DNA Technologies (IDT, Coralville, IA, USA) and Phusion High Fidelity Polymerase (Stratagene, La Jolla, CA, USA). Inserts and the pET-51b(+) plasmid (Novagen, San Diego, CA, USA) were double digested using the appropriate restriction enzymes (Supplementary Table 10, 11) (NEB, Ipswich, MA, USA). The gene cloning, gene expression and protein purification were carried out as previously described⁴³. The purified proteins were assessed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). Protein concentrations were measured by Nanodrop 2000c spectrophotometer based on their extinction coefficients and molecular weights.

Enzyme assays

Purified recombinant 17α-HSDH and 17β-HSDH activities were determined by mixing 10 nM enzyme, 50 μM substrate, and 200 μM cofactor (NADPH/NADP⁺ or NADH/NAD⁺) in phosphate-buffered saline. Reactions were incubated at 25⁰C for 24 h. Samples were collected before and after the reactions. Steroids were extracted as mentioned above. Extracted samples were sent to the Mass Spectrometry Lab (University of Illinois at Urbana-Champaign, Urbana, Illinois, United States) for metabolite analysis using LC-MS.

Sample collection and prostate cancer study cohort

All specimens were studied with informed consent under Johns Hopkins Medicine Institutional Review Board approved protocol #IRB00128147 and IRB00246727. Study participants (n=69) were instructed to collect a full stool sample followed by self-collection of rectal swabs (FLOQSwabs, Copan). All of the study participants had a clinical history of prostate cancer. The participants classified as “hormone-sensitive prostate cancer” (HSPC) were not undergoing any treatment and were either undergoing active surveillance or had newly PSA recurrent disease. The participants undergoing treatment with androgen deprivation therapy (ADT) combined with abiraterone acetate/prednisone (AA/P) had advanced prostate cancer, the majority of which was castration-resistant disease. All participants were male. The age of the participants ranged from 53–94 (Source Data Figure 2). Participants were recruited in the Department of Oncology and Urology outpatient clinics at Johns Hopkins Hospital. Multiple urologists and oncologists participated in patient recruitment, and the patient population was not limited to one clinician. Participants were asked to participate in the study if they 1) had a clinical history of prostate cancer and were not undergoing treatment or 2) if they were initiating treatment with ADT+AA/P. There were no known self-selection biases or other biases in patient recruitment that would have affected the study results. Participants were not compensated.

The majority of the fecal samples used in this study were derived from rectal swabs. Eight of the fecal samples were swabs of stool. The samples were stored at −80°C until time of DNA isolation. Samples were categorized as AA/P “stable” (n=28) if the donor had circulating PSA levels at the time of sample collection that were decreasing or had not changed from the prior PSA measurement. Samples were categorized as “progressing” if the donor had circulating PSA levels that had increased at least 0.2 ng/mL from the nadir, and that continued to rise (n=27) or that was re-detectable after a prolonged period of being below the limit of detection (n=1). Twelve participants in the study had matched fecal samples collected while stable on AA/P and then while progressing on AA/P.

Fecal DNA isolation and desF quantitative PCR (qPCR)

DNA was isolated from fecal samples as previously described⁵⁸. Samples were diluted to 10 ng/μL in DNA-free water. For each 20 μL reaction, the following reagents were combined: 10 μL of iQ SYBR Green Mix (Cat No. 1708882, Bio Rad Laboratories), 2 μL of 10 μM Forward/Reverse Primer set, 6 μL DNA-free water, 1 μL of 2 ug/μL BSA (Cat No. B14, ThermoFisher Scientific), and 1 μL of 10 ng/μL DNA. Real-time PCR (qPCR) conditions and primers are outlined in Supplementary Table 12. Total copies of desF were estimated using standard curves with genomic DNA extracted from Csci12708. The qPCR efficiency of all qPCR assays was determined to be between 86–114%.

Isolation of P. lymphophilum API-1 from prostatectomy tissue

The post-surgical prostatectomy specimen was placed in a sterile container following resection and transported to the grossing room. Here, a sterile field was assembled under a vertical laminar flow module (Envirco Corporation) for collection of tissue cores. A Biopty gun and sterile, single-use Biopty needles (18 gauge x 16 cm, C.R. Bard) were used to obtain two cores from both the right and left lobes of the prostate. Biopsy needles were positioned from apex to base, sampling the posterior (peripheral) aspect of the prostate. The biopsy tissues were minced in sterile PBS, placed in a tube of thioglycolate broth, and cultured anaerobically at 37°C. Stock cultures were stored in 33% glycerol at −80°C.

Patient recruitment and urine sample collection

The urine sample from the prostatectomy patient was collected with informed consent under Johns Hopkins Institutional Review Board approved protocol #IRB00128147. To further identify androgen-forming urinary microbial taxa, we consented and recruited 25 patients and 14 healthy individuals under the IRB #22383 (University of Illinois Urbana-Champaign) and Carle Hospital #18CCC1757. All of the study participants had no history of prostate cancer. The age ranged from 50 to 90 with a BMI less than 35. The patients were excluded: having active diabetes; currently being treated for sexually transmitted infection, or urinary tract infection; taking antibiotics within the last month; currently being treated for benign prostatic hyperplasia (BPH) i.e. with drugs such as alfuzosin (Uroxatral), doxazosin (Cardura), tamsulosin (Flomax), and terazosin (Hytrin) or abiraterone/prednisone or similar drugs. Participants were given a urine collection kit and detailed instructions on how to properly collect a “clean/sterile catch”. The urine collection kit contained an alcohol swab to clean the urethra and tip of the penis. Participants were directed to catch mid-stream urine in sterile vials. At least 20 mL of urine was collected from participants. All urine samples were labelled with a unique, non-identifying code and were not derived from, or related to the participant’s personal information. The urine samples were processed immediately for culturomics work. Participants were compensated with a $10 gift card.

Urine culturomics

To test urine samples for androgen-producing bacteria, 100 μL urine from each patient was screened in PYG broth containing 11DC and 11OHAD as substrates. 11DC was used to confirm side-chain cleavage activity (desAB) followed by 17β-HSDH activities in the urine yielding AD and T whereas 11OHAD conversion to 11OHT can identify 17β-HSDH activity when desAB-encoding microbes are lacking in the sample. LC-MS was performed to analyze metabolism of the substrates in these cultures.

Urine (100 μL) from each individual was plated (duplicate plates for each agar) to Blood agar, Columbia agar and Schaedler agar respectively. Each plate from each agar was incubated aerobically or anaerobically at 37°C. After 4–5 days, single colonies were picked using sterilized toothpicks to a 96-well plate containing PYG broth supplemented with 50 μM 11DC and 11OHAD. After incubation in an anaerobic chamber for 5 d, each column (50 μL/well) of 96 well plate was pooled to a single 1.5 mL centrifuge tube and extracted for the steroid. Columns positive for the steroid conversion were further processed to identify the individual positive well(s) and the microbe carrying the conversion activity.

Cell proliferation assay (MTS)

Androgen-starved LNCaP and VCaP cells were trypsinized and seeded in 96-well plates at a density of 10,000 cells/well and then treated with enzalutamide (2 μM) or VC (DMSO; final concentration 0.2%). After 24 h, these cells were treated with VC (0.5% methanol) or 10 nM T, epiT, AD, and 11OHT. For dutasteride, LNCaP cells were treated with 100 nM drug or 0.2% DMSO as a VC. After 24 h cells were treated with VC (0.5% methanol) or 10 nM T, DHT, epiT and epiDHT. There were 6 replicates for each treatment. 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium (MTS) reagent (15 μL) was added to each well at the end of the experiment and incubated for 90 min. Absorbance was measured at 490 nm in a Biotek Synergy HT plate reader.

To determine if epiAT or ATT derived from AT metabolism by Csci12708 or P. lymphophilum API-2 impacted proliferation of LNCaP cells, Csci12708 was cultured in anaerobic TSB containing 50 μM AT or VC (DMSO). P. lymphophilum API-2 was cultured in anaerobic PYG containing 50 μM AT or VC (DMSO). For sampling, 1 mL cultures were collected at 0, 24, 48 and 72 h. Metabolites of AT to epiAT by Csci12708 and AT to ATT by P. lymphophilum API-2 were analyzed by LC-MS. The spent cultures (72 h) were filtered through 0.2 μm-syringe filters prior to adding to LNCaP cell cultures. The filtered (sterilized) spent culture medium was divided into the following treatment groups: 1) VC (TSB/PYG-DMSO spent culture); 2) AT spiked (TSB/PYG-DMSO spent culture with 50 μM AT added); and 3) epiAT/ATT (TSB-AT spent culture/ PYG-AT spent culture. LNCaP cells were seeded to 96-well plates as mentioned above and exposed to the equivalent of 10 nM of AT or epiAT/ATT from the above treatment groups 2 and 3, respectively or equivalent dilution of VC (group 1). After 4 d of incubation, cell proliferation assay was determined.

RNA extraction and gene expression qPCR from mammalian cells

LNCaP cells incubated in cRPMI medium for 24 h were trypsinized and seeded in 12-well plates at a density of 100,000 cells/well. Cells were treated with VC (DMSO) or 2 μM enzalutamide. After 24 h, cells were treated with T, epiT or VC and incubated overtime. Cells were collected and pelleted in 1.5-mL microcentrifuge tubes by centrifugation (500 x g). Total RNA was extracted from cell pellets using GeneJET RNA Purification kit (ThermoScientific) where the lysis buffer was supplemented with 2% of 14.3 M β-mercaptoethanol. DNA contamination was removed from the extracted RNA using RapidOut DNA Removal Kit (thermoscientific). Total RNA was measured using Nanodrop and 100 ng high-quality RNA was converted to cDNA with High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Thermofisher, Waltham, MA). Real-time PCR (StepOnePlus Real-Time PCR Systems; v 2.0 Applied Biosystems, Waltham, MA) was used to analyze differential gene expression. A total reaction volume of 20 μL in each well contained 0.5 μM forward and reverse primers (Supplementary Table 13), 8 ng cDNA and PowerUp SYBR Green Master Mix (Applied Biosystems). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH), a housekeeping gene was used as an exogenous reference to normalize transcription of target genes KLK3 and TMPRSS2 and the subsequent data was analyzed by the ΔΔCt method.

Bacterial encapsulation and host/microbe coculturing

P. lymphophilium API-2 strain was cultured in PYG broth for 3 d in an anaerobic chamber. Encapsulation processes were followed by the method as previously described⁵⁹. Hydrogel beads were immediately transferred to the anaerobic chamber and incubated in PYG for 4 d. For the coculture study, P. lymphophilium API-2 beads and control (empty beads) were washed with DI sterilized water and conditioned in cRPMI medium for 6 h. The bacterial and control beads were then transferred to the 24-well plate. LNCaP cells maintained in cRPMI were sub-cultured to 24-well plates (50,000 cells/well) and 4 wells were allocated to each treatment group. LNCaP/API-2 coculture platform was treated with cortisol (final concentration 10 nM) or VC (0.5% methanol) and incubated. After 96 h, beads were carefully removed from the cells using a 10-μl inoculation loop. Cell proliferation assay was conducted by adding 50 μl MTS reagent to each well, incubating for 90 min, and reading the absorbance (490 nm) using a Biotek Synergy HT Plate Reader.

Whole genome sequencing and de novo assembly

Urinary isolates were incubated at 37°C in anaerobic PYG broth. During log growth, cells in cultures (10 mL) were harvested by centrifugation (2576 x g) at 4°C for 15 min). High-molecular-weight (HMW) DNA extraction was extracted as previously described^26,60. Approximately 500 ng HMW DNA was sent to Roy J. Carver Biotechnology Center, DNA Services Laboratory (University of Illinois at Urbana-Champaign, Urbana, Illinois, United States) for whole genome sequencing. The HMW DNA was sheared with a Megaruptor 3 to an average fragment length of 13 kb. Sheared DNA was converted to a library with the SMRTBell Express Template Prep kit 3.0 from PacBio. The library was sequenced on a shared SMRTcell 8M on a PacBio Sequel IIe using the CCS sequencing mode and a 30-hour movie time. CCS analysis was done in an instrument with SMRTLink V11.0 (PacBio) using the following parameters: ccs --min-passes 3 --min-rq 0.99.

Read quality was evaluated using FastQC v0.11.8⁵¹. SeqKit v2.0.0⁵² was used to calculate the statistics of the sequencing files for each microbial isolate. Flye v2.9⁶¹ was used to assemble the reads (enough for 50-fold coverage) chosen using the parameters: --asm-coverage. Assembly quality and completeness were evaluated using QUAST v5.0.2⁶² and BUSCO v5.5.0⁶³, respectively. Annotations were performed using Prokka v 1.14.6⁶⁴. CGView Server was used to make the circular genome maps⁶⁵. The 16S rRNA genes were extracted by SnapGene v6.2.1 and used to determine the species by doing the BLAST in NCBI.

Statistics and Reproducibility

Statistical analyses were performed with R version 4.3.0⁶⁶. Data are shown as mean ± standard deviation (SD) when data are normalized. Data are shown in median and interquartile ranges when skewed. Categorical data are shown as counts and percentages. For the normalized data, differences between groups were analyzed by t-test (two-sided) with a 95% confidence interval. The P values for multiple tests were corrected using Benjamini-Hochberg false-discovery rate (FDR). For the skewed data, Kruskal-Wallis one-way ANOVA (two-sided) was used for statistical analysis with a 95% confidence interval. The P values for multiple tests were corrected using Dunn’s test. Differences in categorical data were analyzed using chi-square test. A P value ≤ 0.05 was considered statistically significant. The bacterial metabolite and cell line growth analysis were done with three biological replicates. The SEM for C. scindens strains were done twice. The gels for DesF and DesG were done twice. The SEM for P. lymphophilum strains were done once. Only the representative images were taken for SEM and unprocessed gels. All figures were created or modified in BioRender (Wang, T. (n.d.) https://BioRender.com/c04p300).

Extended Data

Extended Data Figure 1. Proton and Carbon NMR analysis of purified reaction product of androstenedione in cultures of Clostridium scindens VPI12708.

¹H and ¹³C-NMR spectroscopic data were obtained on a JNM-ECA800 (JEOL, Ltd., Tokyo, Japan) instrument operated at 800 and 200MHz, respectively, with CDCh as the NMR solvent. Chemical shifts were expressed in d (ppm), and coupling constants J_H,H are given in Hz. ¹H-¹H nuclear overhauser effect spectroscopy (NOESY), ¹H-¹H correlation spectroscopy (COSY), ¹H-¹³C heteronuclear single-quantum correlation spectroscopy (HSQC), and ¹H-¹³C heteronuclear multiple-bond correlation spectroscopy (HMBC) spectra were obtained using gradient-selected pulse sequences. The ¹³C distortionless enhancement by polarization transfer (135°, 90°, and 45°) spectra were measured between CH₃, CH₂, CH, and coherence based on their proton environments.

Extended Data Figure 2. Comparative genome analysis of Csci35704 and Csci12708 revealed reductase candidates for 17a-HSDH.

Highlighted in yellow is the candidate significantly upregulated in RNA-Seq dataset (See main text Figure 1).

Extended Data Figure 3. RNA-seq analysis to identify the gene encoding 17a- hydroxysteroid dehydrogenase involved in epitestosterone formation by Clostridium scindens.

a. In vitro induction of the gene encoding 17a-hydroxysteroid dehydrogenase by addition of 50 pM 11OHAD. DMSO was used as the vehicle control (Experiment one, n=4 biological replicates). b. Percentages of the reads mapped to the reference genome of Clostridium scindens VPI 12708 (Experiment one). c. MDS plot of RNA-Seq samples for the first experiment. d. In vitro induction of the gene encoding 17a-hydroxysteroid dehydrogenase by addition of 50 μM 11OHAD (Experiment two, validation experiment, n=3 biological replicates). e. Percentages of the reads (Experiment two) mapped to the reference genome of Clostridium scindens VPI 12708. f. MDS plot of RNA-Seq samples for the second experiment. g. Differential gene expression scatter plot of Experiment two. Significantly upregulated genes in red, downregulated genes in blue, and not differentially regulated in black.

Extended Data Figure 4. Maximum likelihood phylogeny of DesF.

Tree is colored by taxonomic affiliation, according to the included key. Radial tree shows the complete one-thousand sequence tree, and the region near C. scindens is shown as a phylogram. Numbers near nodes are branch support values (SH-like approximate likelihood-ration test and ultrafast bootstrap), with only values greater than 50 shown (an * denotes that one of the values is under 50 but the other is not).

Extended Data Figure 5. Phylogenomic and diversity of des gene presence in strains of C. scindens.

a. The formation of two clades is shown, Clade 1 (green) includes 15 strains and Clade 2 (blue) 19 strains. Bootstrap support values above 50% are shown in yellow stars at nodes. Dots show des gene presence in strains of each clade. b. LC-MS analysis confirmed that C. scindens S076 is capable of converting 11-deoxycortisol to androstenedione and epitestosterone as predicted by desAB and desF presence.

Extended Data Figure 6. Live cell imaging analysis of LNCaP cell proliferation over time in the presence (2 μM) or absence of AR-antagonist, enzalutamide.

Proliferation was measured at 24 h (a.), 48 h (b.), 72 h (c.) or 96 h (d.) in the presence of 10 nM steroids or VC (n = independent 3 biological replicates, each with 6 technical replicates). Cell proliferation data for each group at each time point are normalized to time 0 h. P values were calculated by unpaired t-test and Benjamini-Hochberg correction, * P < 0.05, ** P < 0.01, *** P < 0.001. VC: Vehicle control (0.5% methanol v/v); AD: Androstenedione; T: Testosterone; epiT: Epi-testosterone; 11OHT: 11β-hydroxy-testosterone.

Extended Data Figure 7. Steroid-induced proliferation of VCaP cell line.

Data are shown with mean ± standard deviation (3 independent biological replicates, each with 6 technical replicates). Cell proliferation data for each group at each time point are normalized to VC. P values were calculated by unpaired t-test and Benjamini-Hochberg correction, * P < 0.05, ** P < 0.01,*** P < 0.001. VC: Vehicle control (0.5% methanol v/v); AD: Androstenedione; T: Testosterone; epiT: Epi- testosterone; 110HT: 11β-hydroxy-testosterone.

Extended Data Figure 8. Proliferation (MTS assay) of LNCaP cells in the presence (100 nM) or absence of dutasteride.

Proliferation was measured at 96 h. Data are shown with mean ± standard deviation (3 independent biological replicates, each with 6 technical replicates). Cell proliferation data for each group are normalized to VC. P values were calculated by unpaired t- test and Benjamini-Hochberg correction, * P < 0.05, ** P < 0.01, *** P < 0.001. VC: Vehicle control (0.5% methanol v/v); T: Testosterone; DHT: Dihydrotestosterone; epiT: Epitestosterone; epiDHT: Dihydroepitestosterone.

Extended Data Figure 9. Androgenic derivatives from 1,4-androstadiene-3,11,17-trione (AT) by C. scindens and P. lymphophilum.

a. A proposed biochemical pathway by which Csc/12708 converts AT to epiAT and the change of the under-curve area of AT and epiAT over time (n=3). b. A proposed biochemical pathway by which P. lymphophilum API-2 converts AT to ATT and the change of the under-curve area of AT and ATT over time (n=3). c. In vitro incubation of Clostridium scindens VPI 12708/P. lymphophilum API-2 with 50 μM AT as substrate. The culture medium after removing the bacterial cells was diluted and used for the LNCaP cell line promotion effect study. d-e. Proliferation (MTS assay) of LNCaP cells in the presence (2 μM) or absence of AR-antagonist, enzalutamide at 96 h with medium culture control (CL), medium culture control with AT (AT), and metabolite of AT in the medium culture (epiAT/ATT). Data are shown with mean ± standard deviation (3 independent biological replicates, each with 6 technical replicates). Cell proliferation data for each group are normalized to VC. P values were calculated by unpaired t-test and Benjamini-Hochberg correction, * P < 0.05, ** P < 0.01, *** P < 0.001.

Extended Data Figure 10.

Bacterial colony morphology, SEM images and the circular genome maps of the androgen-producing Propionimicrobium lymphophilum strains from males diagnosed with prostate cancer and age-matched control males.

Extended Data Figure 11. Maximum likelihood phylogenetic analysis of DesG from Propionimicrobium lymphophilum strain API-1.

Tree is colored by taxonomic affiliation, according to the included key. Radial tree shows the complete five-hundred sequence tree, and the region near P. lymphophilum is shown as a phylogram. Numbers near nodes are branch support values (SH-like approximate likelihood-ration test and ultrafast bootstrap), with only values greater than 50 shown (an * denotes that one of the values is under 50 but the other is not).

Extended Data Figure 12. Functional sampling of desG homologs for 17β-HSDH activity.

a. Zoom in of DesG branch with proteins chosen for functional characterization (asterisks). b. Genomic context of desG candidates from taxa represented in DesG branch. c. SDS-PAGE of affinity (Streptactin) purified recombinant DesG candidates. d. LC-MS traces after enzyme assay containing either DesG candidate (10 nM) + NADPH (200 pM) and androstenedione (50 μM), or DesG candidate (10 nM) + NADP+ (200 μM) and testosterone (50 μM) incubated for 6 h before steroid extraction. Authentic androstenedione and testosterone standards were also included.

Extended Data Figure 13.

Genome quality metrics (a.), average nucleotide identity, (b.) and synteny between Propionimicrobium lymphophilum strains (c.) isolated from human male urine.

Extended Data Figure 14. Abiraterone (A) does not inhibit bacterial steroid-17,20- desmolase (desAB).

a. LC-MS chromatograms of C. scindens ATCC 35704 or P. lymphophilium API-1 treated with 50 pM A and then incubated with 50 pM cortisol for 72 hours. Quantification analysis showed that A did not inhibit the depletion of cortisol by C. scindens ATCC 35704 (b.) or P. lymphophilium API-1 (c.). There are no differences in the metabolite 11OHAD or 11OHT between the groups in the presence/absence of A. Data are shown with mean ± standard deviation (n=3 independent biological replicates).

Extended Data Figure 15.

Schematic representation of potential host-microbiome interactions relating to the conversion of glucocorticoids to androgens.

Data availability

The raw RNA-Seq reads are available at the National Center for Biotechnology Information (NCBI) with accession numbers PRJNA1108660 (Experiment one, desF), PRJNA1108687 (Experiment two, desF), and PRJNA1115127 (desG) respectively. The raw genome sequencing data are available at the NCBI with accession numbers PRJNA1108738 (P. lymphophilum strains) and PRJNA1026650 (C. scindens strains) respectively. C. scindens VPI 12708 genome is available with accession number PRJNA902789.

Codes availability

All scripts and HMMs used in the metagenomic analyses are available at https://github.com/AnantharamanLab/Clostridium_scindens_mining. All the codes and related data are available at https://github.com/wangtaojun11/Bacterial_enzymatic_pathways_DesF_and_DesG.