Gut microbial β-glucuronidases influence endobiotic homeostasis and are modulated by diverse therapeutics.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Matthew Redinbo ( [email protected] ). This study did not generate new unique reagents. Metagenomic sequencing data have been deposited at the NCBI Sequence Read Archive (SRA) and are publicly available as of the date of publication. Accession numbers are listed in the key resources table . Proteomics data have been deposited at the repository listed in the key resources table . All original code has been deposited at Zenodo and GitHub and is publicly available as of the date of publication. DOIs are listed in the key resources table .

To examine how the microbiota impacts levels of endobiotics and other bioactive small molecules in the GI tract ( Figure 1A ), we collected high-coverage metabolomics data on fecal samples from 12 germ-free and 12 conventional C57BL/6 mice ( Figure 1B ; 6M and 6F per group). 89 We chose untargeted metabolomics for this initial experiment because it allowed us to maximize the number of unique endobiotic-glucuronides and aglycones identified when comparing germ-free and conventional mice, even though it misses some low abundance compounds.Twelve glucuronide-conjugated endobiotics were sufficiently abundant to be identified and compared across the samples ( Figure 1C ). The conjugates included dietary compounds ( e.g. , glycitein- and genistein-glucuronide), indoles ( e.g. , indole-3-acetate O- glucuronide, N -acetyl-serotonin-glucuronide), vitamins (D2-3-glucuronide), and hormones ( e.g. , tetrahydroaldosterone-3-glucuronide). To assess whether levels of aglycones from these endobiotic classes also varied between groups, we used a high-coverage combination of extant reference annotation databases and in-house spectral libraries to further annotate relevant metabolites. 89 By doing so, we were able to compare relative fecal levels of low-abundance aglycones between GF and conventional mice, such as trimethylamine N-oxide (TMAO). Other aglycones included hormones (5-androstenediol), hormone precursors (thyronine), and the neurotransmitters serotonin, dopamine, glutamate, acetylcholine, histamine, epinephrine, and gaminobutyrate (GABA) ( Figure 1C ). In these data, we observed compounds either in the aglycone or the glucuronidated state; we did not detect the same metabolite in both states across both groups of mice. However, significant differences were observed in the overall abundances of inactive glucuronides and active aglycones between germ-free (yellow) and conventionally colonized (blue) mice ( Figure 1C ; 1D ). In all cases, conventional mice with microbiomes had more fecal aglycones while germ-free mice exhibited up to 1,000-fold higher levels of inactive glucuronides compared to active aglycones ( Figure 1D ). Metabolic disparities between germ-free and conventional mice were particularly apparent when comparing MS peak intensities for individual metabolites between the groups. For example, levels of serotonin ( Figure 1E ; P < 0.0001) and dopamine ( Figure 1F ; P = 0.0071) are representative of the stark differences in the metabolic profiles of neurotransmitters between the groups. Similarly, peak intensities for 5-androstendiol ( Figure 1G ; P = 0.0018) and L-thyronine ( Figure 1H ; P = 0.0051) demonstrate the dramatic reduction of hormone profiles in the absence of a microbiota. We also observed clear sex-dependent metabolic differences in active compounds within conventionalized mice ( Figure 1C ). Indeed, approximately 50% of the compounds across our panel were differentially abundant between the male and female mice with conventional microbiota ( Figure S2A ). Fecal mice from conventional male mice generally contained more endobiotic-glucuronides and reflected greater amounts of several neurotransmitters including acetylcholine, glutamate, GABA, and serotonin ( Figures S2B – G ). Conversely, fecal samples from conventional female mice reflected elevated levels of the neurotransmitters dopamine, norepinephrine, and taurine ( Figure S2H – J ), and the hormone 5-androstendiaol ( Figure S2K ) relative to male mice. These results collectively demonstrate that the intestinal profiles of hormone and neurotransmitter glucuronides and their bioactive aglycones are distinct between male and female mice with conventional microbiota. Together, our findings show that mice with microbiota primarily harbor active intestinal endobiotics while germ-free mice contain inactive conjugates. Actions by gut microbes towards glucuronidated metabolites increase the bioavailability of active endobiotics, suggesting that the absence of this metabolism may contribute to the alterations in homeostasis commonly reported for germ-free mice. 90 , 91 , 92 Informed by the overall differences observed between germ-free and conventional mice in hormone and neurotransmitter levels, serotonin, dopamine, estradiol, estrone, and thyroxine were selected for in vitro analyses. All are bioactive endobiotic metabolites integral to human physiology, are known to reach the gut as inactive glucuronides, and sample a range of compound sizes and chemical features ( Figure 1I ; Table 1 ). We recombinantly purified fourteen distinct GUS enzymes that included members from the family’s eight functional classes and broad taxonomic diversity representative of the gut microbiome. Each enzyme was then evaluated for their capacity to reactivate the five endobiotics ( Figure 1 I and J ). While all enzymes showed comparable levels of activity with the reporter substrate 4-Methylumbelliferyl-β-D-glucuronide (4-MUG), distinct patterns of enzymatic activity were observed when the five endobiotic glucuronides were used as substrates ( Figures 1J ). Thyroxine-glucuronide was processed by enzymes from six of the eight GUS classes, while the remaining substrates were preferentially processed by Loop 1 and FMN GUS proteins, categories which were previously shown to efficiently process glucuronidated xenobiotics. 22 , 28 , 93 The glucuronides of estradiol and estrone served as efficient substrates by all eight of the Loop 1 and FMN GUS enzymes in our panel. Conversely, the two neurotransmitter-glucuronides were more selectivity reactivated by members of the Loop 1 and FMN GUS classes ( Figure 1J ). These results establish that gut microbial GUS enzymes process diverse endobiotic-glucuronides and that the Loop 1 and FMN structural classes are most efficient with these substrates. We next explored ex vivo differences in endobiotic-glucuronide processing efficiencies between human fecal samples, an approach that has shown utility in evaluating drug and toxin reactivation by the GI microbiota. 25 , 87 , 88 To provide initial insights into the variability of gut microbial β-glucuronidase activities, we collected fecal samples from fourteen healthy donors in Massachusetts and North Carolina. Whole-genome shotgun metagenomic sequencing data demonstrated significant taxonomic diversity at the class level among the samples ( Figure S3A ; Tables S1 ). We then applied structural metagenomics, which assesses sequence identity and the conservation of GUS-essential active site residues, to identify genes encoding bacterial GUS enzymes across the 1.4 million genes present in the cohort metagenome ( Figure S3B ). 22 , 76 , 94 , 95 A total of 157 unique GUS genes were identified, significantly more than the 43 that would have been identified using automated functional annotation approaches. 96 Genes representing all eight structural classes defined for this protein family were present across the collection of 157 GUS genes ( Figure S3C ; Table S2 ). Each sample contained between 4 and 17 unique GUS genes ( Figure S3D ), with at least one gene belonging to the Loop 1 or FMN structural class. 22 We extracted complex protein lysates from each fecal sample ( Figure 2A ), and using a previously optimized activity-based protein profiling (ABPP) pipeline, identified and quantified GUS protein proteomic abundance. 25 , 87 Enriched peptide fragments were used to identify and quantify individual GUS proteins using the cohort metagenome as a peptide reference database. 25 , 87 , 97 In total, 50 unique GUS proteins were identified across the cohort which sampled six of the eight structural classes present in the cohort metagenome ( Figure 2B ; Table S3 ). While the overall abundances of GUS proteins were comparable across the samples, the relative composition of GUS enzymes was unique in terms of both structural class and taxonomic origin ( Table S4 ). All samples contained GUS orthologs from at least three of the eight classes. FMN, Mini-Loop 1, and No Loop proteins were found in every sample, while members of the Loop 1 GUS family were present in eight of the fourteen samples. These results demonstrate that each sample contains a distinct GUS protein compositional “fingerprint”. We next measured rates of endobiotic-glucuronide reactivation for each fecal lysate sample using HPLC-MS/MS ( Figure 2C ; Table S5 ). We found that all substrates were processed by the fourteen samples, and that the rates of reactivation varied from 0.4 to 55 nM/sec. Consistent with our observation that the GUS protein composition and structural class abundances differ between donors, some samples showed relatively low (< 20 nM/sec) activities for all substrates ( e.g. , Donors 5, 12, 6), while others exhibited both low and high (10 – 60 nM/sec) endobiotic reactivation rates ( e.g. , Donors 2, 10, 14). Overall, each human fecal sample showed differential processing capabilities towards the panel of endobiotic substrates, supporting the conclusion that interindividual differences in gut microbial composition results in functionally varying activities which contribute to endobiotic homeostasis. We hypothesized that differential GUS protein compositions ( Figure 2B ) would explain the varying patterns of glucuronide processing observed in the ex vivo fecal lysates ( Figure 2C ), and thus explored associations between reactivation rates and proteomic GUS abundance ( Table S6 ). We found that the average endobiotic processing rate for each donor positively correlated with their total GUS protein abundance (P < 0.01; ρ = 0.596; Figure S4A ). Out of the six clades detected using targeted metaproteomics, only FMN GUS protein abundances were significantly associated with average rate of endobiotic reactivation (P = 0.0001; ρ = 0.852; Figure S4B ). These associations corroborate our in vitro results showing efficient processing of hormone- and neurotransmitter-glucuronide substrates by enzymes belonging to the FMN GUS structural class ( Figure 1J ). We therefore sought to identify the specific GUS molecular features across the 50 unique GUS enzymes detected by proteomics ( Figure 2D ) that favor reactivation of individual hormone and neurotransmitter-glucuronides. We examined finer sample sets of enzymes identified by proteomics, along with AlphaFold structural models and relative sequence identities. Loop 1 GUS abundance was positively associated with reactivation rates for estradiol-glucuronide for the eight individuals containing Loop 1 GUS proteins (P = 0.011; Figure 2E ). The two Loop 1 enzymes identified in our proteomics data, GUSs from Gemmiger qucibialis and Faecalibacterium prausnitzii , share 89% sequence identity and nearly identical AlphaFold models (0.2 Å root meansquare deviation [RMSD] on Cα positions). Rates of estradiol (P = 0.013; Figure 2F ; Figure S4C ) and estrone (P = 0.013; Figure 2G ) reactivation were also associated with total abundance of FMN GUS enzymes. As FMN GUS were detected in all samples, all donors are represented in, and adhere to, these positive correlations. While the FMN GUS association for estradiol-glucuronide is weaker than that observed for Loop 1 GUS abundance ( Figure 2E ), combining total Loop 1 and FMN GUS proteomic abundance presented the strongest association with estradiol-glucuronide reactivation ( Figure S4D ). Associations with estradiol-glucuronide reactivation could be further generalized to the total proteomic abundance of Firmicutes ( Figure S4E ) but not Bacteroidota GUS ( Figure S4F ; Table S7 ). Thus, the presence of Loop 1 and FMN GUS enzymes preferentially drive the processing of estrogen-glucuronides in human fecal samples. The total abundance of FMN GUS enzymes was also positively associated with reactivation rates for thyroxine-glucuronide (P < 0.0001; Figure 2H ). Thyroxine-glucuronide was the only substrate that showed a correlation between processing rates and total GUS abundance ( Figure S4G ), as well as abundance of GUS enzymes from Firmicutes ( Figure S4H ) but not from Bacteroidota ( Figure S4I ; Table S7 ), which is a significant result given that only Firmicutes produce FMN GUS enzymes. These results corroborate our in vitro results ( Figure 1J ) and support the conclusion that GUS enzymes from Firmicutes, which is the only phyla that produce FMN GUS, preferentially drive the metabolism of glucuronidated endobiotics in the gut. Significant associations were not identified between reactivation rates for neurotransmitter-glucuronides and abundances of GUS enzymes from full structural clades as observed for hormone glucuronides. This finding aligns with our in vitro data which suggests that subsets of Loop 1 and FMN GUS efficiently reactivate the neurotransmitter substrates ( Figure 1J ). Therefore, we grouped the proteomics data into structural subclades to identify the drivers of neurotransmitter-glucuronide processing. We compared AlphaFold models for the fifty GUS proteins identified across our cohort to assess structural similarity by RMSD values on Cα positions, which ranged from 0.06 to 4.50 Å ( Table S8 ). We then clustered proteins based on both structural class and structural similarity (≤ 2.0 Å RMSDs) and explored potential abundancerate relationships for each subclade ( Table S9 ). By doing so, we identified significant associations for both neurotransmitter-glucuronide substrates. Dopamine-glucuronide processing correlated with the abundance of FMN GUS derived from a subclade of Lachnospiraceae and Acutalibacteraceae detected in eight donors (P < 0.05; Figure 2I ). Though this collection of proteins exhibits < 50% sequence identity, the subclade includes Ruminococcus gnavus and Roseburia hominis GUS proteins, both of which efficiently processed dopamine-glucuronide in our in vitro experiments ( Figure 1J ). The subclade generated by structural clustering excluded Roseburia inulinivorans FMN GUS, which did not process dopamine-glucuronide in vitro ( Figure 1J ), even though our in vitro results were not used to generate the cluster. We then examined associations between dopamine reactivation and bacterial abundance from our metagenomic data ( Tables S11 and S12 ). We found that samples with high dopamine-glucuronide processing exhibited a greater metagenomic abundance of Lachnospiraceae overall compared to those with poor processing capabilities (P < 0.05; Figure 2J ). Metagenomic abundance associations were not observed for any other higher-order taxa with dopamine-glucuronide, and associations with Lachnospiraceae abundance did not extend to any other hormone or neurotransmitter-glucuronides. Thus, structural clustering can identify protein subclades that share low sequence identity but still correlate with endobiotic reactivation rates ex vivo , align with in vitro functional data, and present significant associations with the cohort metagenome. A subclade of No Loop GUS enzymes from Clostridia presented the strongest association with the processing of serotonin-glucuronide (P < 0.0001; Figure S4J ) however, the single No Loop GUS examined in vitro ( Fusicatenibacter saccharivorans ; Figure 1J ) did not reactivate any hormone or neurotransmitter-glucuronides. We compared AlphaFold models between the No Loop GUS enzymes associated with serotonin reactivation and F. sacchavorians No Loop GUS to identify a structural basis for this this incongruency ( Figure S4K ). While all enzymes from the serotonin association shared RMSD values of ≤1.5 Å with each other, these proteins exhibited RMSDs ≥3.1 Å when compared to F. sacchavorians No Loop GUS. A subclade of FMN GUS enzymes from Clostridia was also positively associated with serotonin reactivation (P < 0.05; Figure 2K ). The cladogram of the GUS proteins identified by metaproteomics reveals that some of these FMN enzymes cluster among No Loop enzymes ( Figure 2D ), suggesting that they may functionally overlap. Interestingly, combining Clostridia FMN and No Loop GUS protein abundances generated a significant correlation for serotonin-glucuronide processing that encompassed the majority of the cohort (P < 0.005; Figure S4L ). These findings were corroborated by the cohort metagenomics data, which indicated that individuals with higher processing of serotonin-glucuronide contained a greater bacterial abundance of Clostridia (P < 0.05; Figure 2L ; Tables S11 – 12 ). Together, these results show that structural clustering of fecal metaproteomics and metagenomics can identify the specific GUS enzymes responsible for hormone- and neurotransmitter-glucuronide processing. They also support the conclusion that human gut microbial Loop 1 and FMN GUS modulate the bioavailability of hormones and neurotransmitters. Drugs with piperazine and piperidine moieties treat diseases ranging from depression to cancer and are associated with modified hormone and neurotransmitter levels that may influence on- and off-target effects ( Table 2 ; Figure S5A ). 98 , 99 , 100 , 101 , 102 , 103 , 104 , 105 , 106 , 107 , 108 , 109 , 110 , 111 , 112 , 113 , 114 , 115 , 116 , 117 , 118 , 119 , 120 Compounds with these moieties such as UNC10201652 and a small panel of drugs are potent inhibitors of Loop 1 gut microbial GUS enzymes, the extent to which they affect additional GUS structural classes and their roles in hormone and neurotransmitter reactivation remain unknown. 49 Therefore, we examined a panel of ten chemically diverse drugs ( Figure S5A ) that contain a piperazine or piperidine moiety and are expected to reach the human gut in concentrations from 5 to >300 μM, where they are reported to cause GI side effects ( Table 2 ). 82 All drugs, but not free piperazine, inhibited Loop 1 or FMN GUS enzymes in vitro but not gut microbial GUS from the other six structural classes ( Figure S5B ). Loop 1 and FMN GUS efficiently reactivate hormone and neurotransmitter glucuronides, as shown above. The five most abundant and prevalent FMN GUS from our cohort’s metaproteomics data ( Table S3 ) were expressed recombinantly and purified for evaluation: Gemmiger qucibialis , R. hominis 2, R. inulinvorans, F. prausnitzii L2–6, R. gnavus 3 . The Loop 1 GUS from Gemmiger qucibialis was also purified, as well as four additional Loop 1 GUS enzymes ( E. coli, S. agalactiae, C. perfringens, E. eligens ). We then compared the half-maximal inhibitory concentration (IC 50 ) for each drug to the IC 50 for the potent Loop 1 inhibitor, UNC10201652, and its inactive analog, UNC10201651 ( Figure 3A ; Figure S5C ). 76 , 97 As expected, all Loop 1 and FMN GUS enzymes were potently inhibited by UNC10201652, while none were inhibited by UNC10201651 ( Figure 3B ; Figure S5C ). The antipsychotic drugs norquetiapine (the active form of Seroquel ® ) and vortioxetine (Trintillix ® ) inhibited both Loop 1 and FMN GUS, while paroxetine (Paxil ® ) was selective for FMN GUS proteins ( Figure 3B ). The anticancer drugs ceritinib (Zykadia ® ) and crizotinib (Xalkori ® ) were effective across the panel of GUS enzymes but palbociclib (Ibrance ® ) was selective for the Loop 1 proteins. Finally, antimalarial mefloquine (Lariam ® ) exhibited a slight preference for FMN GUS, while the active metabolite of the antihistamine 3-hydroxy-desloratidine was more potent than its prodrug desloratadine (Clarinex ® ). Thus, drugs with piperazine and piperidines in the para - and ortho -position inhibit gut microbial Loop 1 and FMN GUS enzymes in vitro at concentrations below those estimated to reach the human GI tract. 82 We previously showed that piperazine- and piperidine-containing compounds, including UNC10201652 and amoxapine, inhibited Loop 1 GUS enzymes by intercepting the enzyme’s catalytic cycle to generate an inhibitor-glucuronide conjugate at the active site. 49 To determine if this mechanism of inhibition is conserved for FMN GUS, we determined the structures of R. hominis 2 FMN GUS and E. eligens Loop 1 GUS crystallized in the presence of UNC10201652 and a hydrolysable glucuronide substrate (8GES; 8GEN; Figure S5D – F ; Table S11 ). Both structures revealed a UNC10201652-glucuronide ( Figure S5D ) non-covalently bound in the active site ( Figure S5E – F ), confirming that the inhibition mechanism is substrate-dependent and conserved between Loop 1 and FMN GUS enzymes. Further co-crystal structures of E. eligens Loop 1 GUS bound with the drugs 3-OH desloratadine ( Figure 3C ; 8GEO; Table S11 ), ceritinib ( Figure 3D ; 8GEQ; Table S11 ), and norquetiapine ( Figure 3E ; 8GER; Table S11 ) also reveal inhibitor-glucuronide adducts, confirming that the mechanism of GUS inhibition is conserved across sterically and functionally diverse FDA-approved drugs. Additionally, a structure of R. hominis 2 FMN GUS with Norquetiapine (8GET; Table S11 ) was resolved and showed a drugglucuronide complex in the active site ( Figure 3F ). Interestingly, the crystal structure of R. hominis 2 FMN GUS in complex with norquetiapine-glucuronide displays an interaction between the inhibitor and 16 amino acids from its C-terminal domain (CTD) ( Figure 3F ), revealing that the CTDs of FMN GUS stabilizes substrates and inhibitors at the active site ( Figure 3F ). These results show that functionally and chemically diverse FDA-approved drugs inhibit both Loop 1 and FMN gut microbial GUS enzymes in the same substrate-dependent fashion of the extant GUS inhibitor, UNC10201652. We next examined the ability of norquetiapine, ceritinib, 3-OH-desloratidine, and mefloquine, UNC10201652 and UNC10201651 to inhibit GUS activity in complex human fecal lysates. 4-MUG provides robust activity and therefore this reporter substrate was used to evaluate the ability of these compounds to inhibit GUS activity in these complex mixtures. UNC10201651, our inactive analog ( Figure 3A ), failed to inhibit any of the fourteen donor samples, while UNC10201652 exhibited the highest degree of inhibition, up to 84% ( Figure 4A ). Norquetiapine, ceritinib and mefloquine displayed more moderate inhibition ranging from 0–70% ( Figure 4A ). On a per sample basis, average inhibition ranged from lows of 2–8% for Donors 2, 3, 5 and 7, to highs of 22–61% for Donors 1, 6 and 11 ( Figure 4A ). Donors were classified as either “High Inhibition” or “Low Inhibition” based upon average inhibition Z score (High = Z score > 0, Low = Z score < 0), and a clear separation between these groups was found in a sparse partial least squares discriminant analysis (sPLS-DA) sample plot of our targeted metaproteomics data ( Figure 4B ). Total GUS abundance was negatively associated with average inhibition ( Figure 4C ), an expected result given that all gut microbial GUS enzyme classes process 4-MUG but only Loop 1 and FMN proteins are inhibited by these compounds ( Figure 1J ; Figure S5B ). No associations could be made between inhibition and proteomic abundance of complete structural classes of GUS enzymes ( Table S6 ). However, structural clustering revealed significant associations for all drugs and UNC10201652 between average donor inhibition and abundance of FMN GUS enzymes from the Roseburia genus ( Figure 4D ). An analysis of genus-level bacterial abundance derived from the cohort metagenome showed clear separation on an sPLS-DA plot between High and Low Inhibition donors ( Figure 4E ). Further metagenomic analyses revealed that Roseburia was the only differentially abundant taxon between the two groups ( Figure 4F ; Table S10 ). Roseburia was absent altogether in five of seven individuals in the Low Inhibition group, and individuals with Roseburia were 3.49-fold more likely to demonstrate high inhibition ( Figure S6A ). AlphaFold models for all six unique Roseburia GUS proteins across the cohort metagenome exhibited remarkable similarity (average RMSD of 1.3 Å; Figure 4G ), despite sharing < 50% sequence identities ( Figure S6B ). These findings show that an elevated abundance of Roseburia FMN GUS enzymes, which are uniform in structure, results in greater inhibition by drugs and compounds with aromatic cores and a basic piperazine or piperidine. They also suggest that gut microbial FMN GUS inhibition specifically may affect host hormone and neurotransmitter levels. FMN GUS proteins drive the reactivation of serotonin in human fecal lysates ( Figure 2 ) and are inhibited by extant targeted GUS inhibitors and select piperazine- and piperidine-containing FDA-approved drugs ( Figure 3 ). We hypothesized that GUS inhibition in a biological model would alter serotonin levels in the gut and serum. We first tested this hypothesis by administering to mice either the most potent GUS inhibitor UNC10201652 or its inactive analog UNC10201651, which differs by a single methylene ( Figure 4H ). Healthy BALB/c mice were orally administered 2 mg/kg of either UNC10201652, UNC10201651, or saline daily for three weeks ( Figure 4H ). Subsequently, the absolute abundances of serotonin and serotonin-glucuronide in serum and cecal content were determined by HPLC-MS/MS. 19 , 50 UNC10201652 significantly decreased the relative levels of free serotonin and increased the relative levels of serotonin-glucuronide in both the mouse cecum (P = 0.0004; Figure 4I ) and serum (P = 0.0051; Figure 4J ). UNC10201651, the inactive control ( Figure 4B ), had no effect on serotonin-glucuronide levels and mirrored the control animals in vivo ( Figure 4I – J ). 70 , 89 We next examined the effects that the anticancer drug ceritinib has on serum and cecal serotonin-glucuronide levels in mice. Ceritinib was chosen because it is not used clinically to address neurotransmitter modulation, and because of its in vitro ( Figure 3B ) and ex vivo ( Figure 4A ) efficacies ( Table 2 ). We administered ceritinib orally to BALB/c mice at the human-equivalent dose of 50 mg/kg and examined serum at days 7 and 12 and cecal contents at day 12, the study’s termination. Though all groups reflected equivalent levels of serotonin-glucuronide at the start of the study, ceritinib significantly increased the levels of serotonin-glucuronide in serum of treated mice at days 7 and 12 ( Figure S6C ). Ceritinib also significantly increased levels of serotonin-glucuronide in the cecal contents of treated mice compared to vehicle ( Figure S6D ). We were not able to measure changes on other endobiotic levels in these mouse studies presumably due to their low overall concentrations. These results show that gut microbial GUS inhibition significantly alters serotonin homeostasis in the mouse gut and serum. Thus, drugs and drug-like molecules have the capacity to influence mouse hormone and neurotransmitter homeostasis in a gut microbiome-dependent manner. If these results translate to humans, such effects may explain some of the commonly observed inter-individual variabilities in drug efficacy and toxicity. Finally, to explore the translatability of our findings across an extant multi-omics dataset, we examined glucuronide profiles in the fecal metabolomes of individuals across the Inflammatory Bowel Disease (IBD) Multi-omics Database (IBDMDB) as a function of fecal metagenomic profiles. 121 Notably, the only glucuronide identified in these data was ethyl-glucuronide. We first examined the abundance of ethyl-glucuronide as a function of GUS-producing species abundance for the 105 donors without IBD. We found that the abundance of ethyl-glucuronide decreased with increased abundance of GUS-producing species (P = 0.005; Figure S7A ). We next added the data from the 340 individuals diagnosed with IBD and found that the association remained significant (P = 0.004; Figure S7B ). These results support the conclusion that glucuronide levels are impacted by gut microbial taxa encoding GUS enzymes in humans in both health and disease.

We show that germ-free mice harbor largely inactive glucuronide-conjugates in their GI tracts while conventional mice with intact gut microbiota primarily contain active aglycones ( Figure 1A – G ). We demonstrate differential reactivation of hormones and neurotransmitters by testing fourteen gut microbial GUS proteins in vitro belonging to all eight structural classes of GUS ( Figure 1H – J ). Using metagenomics and probe-enabled metaproteomics, we pinpoint the enzymes driving these differences in complex human fecal lysates ex vivo ( Figure 2 ), and we reveal that a range of FDA-approved drugs containing piperazine and piperidine moieties inhibit both Loop 1 and FMN GUS enzymes through catalytic cycle interception ( Figure 3 ). Finally, we show that inhibitory effects extend to complex fecal lysates ex vivo ( Figure 4A – 4F ) and modulate active serotonin levels in vivo ( Figures 4H – 4J ; Figure S6C – D ). These results support the overall conclusion that differential gut microbial GUS compositions impact hormone and neurotransmitter levels and can be influenced by diverse FDA-approved drugs not targeting the gut microbiome. Furthermore, they suggest that microbial GUS enzymes contribute to the disparities observed between germ-free mice and conventionally raised animals by regulating hormone and neurotransmitter bioavailability. 90 , 91 , 92 We outline specific yet varied examples of how GUS enzymes reactivate hormone- and neurotransmitter-glucuronides, suggesting that microbial GUS may play a role in a range of disease etiologies. Estrogens are key regulators of development and reproduction, and the transition to menopause, which is triggered by a decreased abundance of estradiol, has been associated with altered GI microbial profiles. 122 , 123 Indeed, members of the Firmicutes phyla become less abundant post-menopause, and both the genus Gemmiger and Faecalibacterium are particularly diminished in several studies examining pre and post-menopausal cohorts. 122 , 123 , 124 , 125 Here, we show that purified Loop 1 and FMN GUS enzymes from these taxa are the most efficient reactivators of estrogens within the gut; both structural classes are predominantly produced by Firmicutes across reference metagenomes. 22 , 76 Across our cohort, Firmicutes were the sole producers of Loop 1 and FMN GUS enzymes, and these enzymes presented the strongest associations with estradiol reactivation both in vitro and within human fecal lysates. Collectively, these results suggest that GUS enzymes from these taxa may play a substantial role in estradiol regulation. 22 , 25 Moreover, decreases in Firmicutes abundance, and Faecalibacterium in particular, have also been associated with reduced estradiol and premature ovarian insufficiency. 122 Conversely, an increased overall abundance of Firmicutes have been associated with both increased estradiol and ER-positive breast cancer. 126 Estrone levels are also increased in ER-positive breast cancer, and our data show FMN GUS enzymes, which are only produced by Firmicutes, are positively associated with estrone reactivation. Therefore, gut microbial GUS enzymes may contribute to dysregulated levels of estrogenic hormones that impact systemic homeostasis and transitions to disease. FMN GUS enzymes as well as overall Firmicutes GUS proteomic abundance were positively associated with ex vivo reactivation rates for thyroxine. These findings corroborate work by us and others demonstrating that Firmicutes GUS process small molecule-glucuronides while Bacteroidota GUS are selective for larger polysaccharides, 22 , 69 , 127 , 128 , 129 a distinction that may have the potential to contribute to disease. While the ratio of Firmicutes to Bacteroidota is generally considered to be representative of positive health and gastrointestinal eubiosis, highly elevated gut Firmicutes and decreased Bacteroidota have been associated with Hashimoto’s Thyroiditis (HT). 130 , 131 , 132 , 133 , 134 Circulating thyroxine triggers negative feedback pathways in the thyroid gland and the hypothalamus that down-regulate overall thyroid activity. 134 , 135 By increasing thyroxine reactivation in the gut, Firmicutes GUS enzymes may contribute to hypothyroidism symptoms characteristic of HT by promoting negative pathways that decrease thyroid function. Structural clustering revealed significant proteomic associations between dopamine-glucuronide processing and FMN GUS abundance from Lachnospiraceae families, and we further found that donors with a greater metagenomic abundance of Lachnospiraceae were more efficient reactivators of dopamine. Several reports have shown that fecal Lachnospiraceae is diminished by an average of 43% in PD patients. 136 , 137 , 138 , 139 , 140 While the biological mechanisms connecting Lachnospiraceae to PD have not been established, the absence of microbes capable of GUS-mediated dopamine reactivation may contribute to the diminished dopamine levels central to this disorder, something that would be the subject of future studies. 141 , 142 Alterations in GI motility and dysfunctional gut-brain interactions in irritable bowel syndrome (IBS) are thought to be caused by high serotonin levels, which has been hypothesized to be driven by an elevated presence of Clostridia characteristic to the disease. 143 Indeed, as Clostridia are known to promote the biosynthesis and release of serotonin from enterochromaffin cells, serotonin reactivation by Clostridia GUS may also contribute to increased serotonin in IBS. 144 , 145 By contrast, both decreased Clostridia abundance and reduced systemic serotonin levels have been observed in Alzheimer’s Disease (AD) cohorts, 146 , 147 , 148 factors that suggest a possible role for Clostridia GUS in disrupted serotonin homeostasis in AD. In our study, the reactivation of serotonin was linked to the proteomic abundance Clostridia GUS as well as the metagenomic abundance of Clostridia ( Figure 3N ). Our findings in the context of prior work connecting Clostridia and serotonin to a range of disease states suggests that microbial GUS enzymes from Clostridia may contribute to dysregulated serotonin levels impacting systemic homeostasis. A panel of ten functionally diverse piperazine- and piperidine-containing FDA-approved drugs known to both cause GI side effects and reach the human gut at high-micromolar concentrations were shown here to inhibit Loop 1 and FMN gut microbial GUS using catalytic cycle interception, like extant GUS inhibitors. 49 Inhibition was observed with IC 50 values as low as 200 nM for GUS enzymes that were also shown to efficiently reactivate endobiotics in vitro , suggesting the potential for GUS inhibition in a physiological context. 70 , 89 , 28 , 70 , 15 , 77 Indeed, we found that drugs targeting a range of biological systems ranging from histamine receptors to tyrosine kinases also inhibit GUS activities in human fecal lysates that reactivate several hormones and neurotransmitters critical to homeostasis. Targeted metaproteomics revealed clear separation between donors with high and low inhibition capacities, and structural clustering connected FMN GUS proteins from Roseburia with donor inhibition for all compounds in our inhibitor panel. In donors exhibiting high inhibition, a significant portion of their total proteomic GUSome was comprised of these Roseburia FMN GUS enzymes. Metagenomics also revealed differences in Roseburia abundance between groups, with individuals with more Roseburia being 3.5-fold more likely to exhibit high inhibition compared to individuals without the genus. Purified Roseburia GUS enzymes derived from the cohort metaproteome were potently inhibited by these FDA-approved drugs in vitro , and the predicted structural similarities for the six Roseburia FMN GUS proteins in the metagenomes suggest they share similar functions. These results provide a molecular rationale for both the comparable potencies observed between Roseburia FMN GUS in vitro and the generalizability of Roseburia associations with inhibition in human fecal lysates. 149 These data support the conclusion that Roseburia FMN GUS enzymes and their inhibition may modulate hormone and neurotransmitter bioavailability in vivo . Mice colonized with bacteria lacking a GUS gene exhibit reduced bioavailability of neurotransmitters. 19 , 50 Here, we establish that active serum and intestinal serotonin levels in mice can be altered by the targeted inhibition of gut microbial GUS enzymes. It is possible that the clinical gut toxicities associated with the drugs examined here may be explained by intestinal alterations in hormone and neurotransmitter levels. We note that quetiapine (Seroquel ® ), vortioxetine (Trintillix ® ) and paroxetine (Paxil ® ) treat diseases associated with serotonin, and it is possible that their on-target efficacies may in part arise from effects on gut microbial GUS enzymes capable of shifting the balance of active and inactive forms of serotonin. In addition, acute psychological toxicity has been associated with anti-malarial prophylaxis using mefloquine (Lariam ® ), including hallucinations and psychosis that can persist after discontinuation and arise in 29–77% of patients via an unknown mechanism. 98 , 110 , 150 Mefloquine’s inhibition of gut microbial GUS activities may provide a partial explanation for these clinical observations. Here, we demonstrate that a basal gut microbiome energy scavenging mechanism also regulates hormones and neurotransmitter homeostasis, and that this metabolism is disrupted by chemically and functionally diverse clinical therapeutics. Together, these findings highlight the role of the gut microbiota in mediating endobiotic bioavailability and demonstrate the extent to which drugs may interact with the microbiota to alter host homeostasis.

Animals were purchased, study was conducted, and fecal samples were analyzed exactly as described in Lai et al. 2021 wherein these experiments were initially conducted. 89 All animal studies were approved by the University of North Carolina Institutional Animal Care and Use Committee under protocol 19–235.0, in accordance with the Care and Use of Laboratory Animals guidelines set by the National Institutes of Health. As described in Lai et al., Conventionally raised (CONV-R) wild-type C57BL/6 mice were purchased from the Jackson Laboratory (Bar harbor, ME, USA) and housed under specific-pathogen-free (SPF) conditions at the UNC animal facility for multiple generations; germ-free (GF) mice were generated and housed in germ-free conditions at the National Gnotobiotic Rodent Resource Center of UNC in Association for Assessment and Accreditation of Laboratory Animal Care International accredited facilities. At age ~7 weeks, C57BL/6 littermates raised under CONV-R and GF conditions were age-matched upon selection, resulting in a final total of 24 mice (12 GF, 12 SPF CONV-R, each with N= 6 males and N = 6 females). All animals were raised under uniform conditions: 22 °C, 40–70% humidity and a 12:12 h light-dark cycle. All mice were consistently administered the same sterile purified Prolab RHM 3000 pelleted rodent diets (St. Louis, MO, USA) with tap water ad libitum. Mice were observed under their typical housing conditions during the week before sample collection; animals exhibiting signs of serious injury or morbidity (e.g., malocclusion or fight wounds) were not included in the study. Feces were harvested, snap-frozen, and stored in −80 °C freezer before analysis. 89 Animal studies were approved by the University of North Carolina Institutional Animal Care and Use Committee, Protocol 21–175, in accordance with the Care and Use of Laboratory Animals guidelines set by the National Institutes of Health. Solid inhibitor (UNC10201651 or UNC10201652) was resuspended to a concentration of 20 mg/mL in 100% Hybri-max DMSO (Sigma) then frozen at −20 °C. Each inhibitor was prepared fresh weekly. Before administration, inhibitor was thawed to ambient temperature and diluted in sterile saline to 20 mg/mL. For each inhibitor, BALB/c mice were administered the compound at a dosage of 2 mg/kg daily via oral gavage over 21 days. Total dosage volume did not exceed 150 μL for any animals in the study. Body mass of the mice was measured daily to ensure the inhibitors were not negatively influencing the health of the animals. Moreover, upon administering compounds, mice were monitored for adverse effects including scruffy coat, hunching, lethargy, inflammation, or diarrhea. Treatment was to be halted if any highly concerning health conditions began to present including ulceration, anal bleeding, prolapse rectum, or 20% loss of BWT and animal will be euthanized accordingly. At the end of 21 days, the animals were sacrificed by carbon dioxide euthanasia, then cecal and serum contents were harvested. Blood was harvested via cardiac puncture then incubated at ambient temperature for 20 mins. Afterwards, the blood was centrifuged at 2,000 × g for 10 mins, then the serum supernatant was isolated. Serum and cecal contents were both snap-frozen in dry ice then stored at −80 °C. Animal studies were approved by the University of North Carolina Institutional Animal Care and Use Committee, Protocol 19–292, in accordance with the Care and Use of Laboratory Animals guidelines set by the National Institutes of Health. Six-week old specific pathogen-free female wild-type C57BL/6J mice were purchased from Jackson Labs, and acclimated for two weeks prior to study start at UNC vivarium maintained at 22C with a 12h light/dark cycle; mice had ad libitum access to chow (irradiated Purina PicoLab ® Select Rodent 50 IF/6F 5V5R*) and drinking water. Ceritinib (50 mg/kg) dissolved in sterile 0.5% w/v methylcellulose/0.5% v/v Tween-80 was administered for 12 consecutive days by oral gavage using flexible feeding tubes (Instech Laboratories). Mice were individually placed in sterilized empty pipet tip boxes for daily monitoring. At study end, mice were deeply anesthetized using CO2, blood collected by cardiac puncture, and euthanized with cervical dislocation. Whole blood was incubated undisturbed at room temperature for 30 minutes, then centrifuged at 1000 × g at 4C to separate serum. Collected serum samples were snap frozen in liquid nitrogen and stored at −80C. From January 2018 to January 2020, fecal samples were collected from informed, consenting healthy volunteers by the Merck Exploratory Science Center and at the University of North Carolina at Chapel Hill (IRB#17–1528); exclusion criteria were antibiotic usage within the preceding three months. Volunteers self-collected freshly voided fecal samples using an at-home commode specimen collection system (Fisher Scientific). Within three hours of voiding, samples were transported to the laboratory in insulated coolers maintained at 4–10 C using gel refrigerant, transferred into a Whitley MG500 anaerobic workstation wherein they were aseptically aliquoted, then stored at −80 C until further use. Cohort demographics are shown in Table S12 . E. coli BL21 DE3 Gold cells were cultured in LB with vigorous shaking at 37°C. Endobiotic glucuronides were annotated using external reference libraries, while aglycones were annotated with external reference libraries as well as an in-house reference library of authentic chemical standards. Data was re-analyzed within the context of glucuronidated endobiotics and active aglycones that were observed in these data to be predominantly glucuronidated, as shown in Figure 1B – 1G and Figure S2 . All GUS genes were assessed for signal peptides using SignalP6.0 and signal peptides were removed if present, then genes were codon-optimized for E. coli expression. 151 Genes were synthesized and ligated into a pLIC-His vector, then purchased from Bio Basic. The vectors were each transformed into chemically competent BL21-Gold (DE3) E.coli cells then grown on LB agar with ampicillin (100 μg/mL) at 37 °C overnight. A single colony was selected and grown overnight in 100 mL of LB broth with ampicillin (100 μg/mL) at 37 °C and shaking at 215 RPM. After reaching saturation, 50 mL of the culture was added to 1 L of LB broth with ampicillin (100 μg/mL), ~3 μL Antifoam 204, and 500 μM FMN. The culture was incubated at 37 °C and 215 RPM until it reached an OD of 0.6 at 600 nm. After reaching the target OD, 1-thio-β-D-galactopyranoside (IPTG; 100 μM) was added to induce protein expression, the temperature was lowered to 18 °C, and the culture was incubated overnight. The cells were collected by centrifugation at 4,500 × g at 4°C in a Sorvall (model RC-3B). Cell pellets were resuspended in 35 mL Purification Buffer A (20 mM Potassium Phosphate, 50 mM imidazole, 500 mM NaCl, 50 μM FMN, pH 8.0) with DNase, lysozyme, and a complete-EDTA free protease inhibitor tablet (Roche). Resuspended cells were sonicated and clarified via centrifugation at 17,000 × g for 60 min in a Sorvall (model RC-5B). The lysate was flowed over a Ni-NTA HP column (GE Healthcare) then loaded onto the Aktaxpress FPLC system (Amersham Bioscience) and washed with Purification Buffer A. Protein was eluted with Buffer B (20 mM Potassium Phosphate, 250 mM Imidazole, 500mM NaCl, 50 μM FMN, pH 8.0). Fractions containing the protein of interest were concatenated then passed through a HiLoad 16/60 Superdex 200 gel-filtration column (GE Life Sciences). Protein was eluted in S200 buffer (20 mM HEPES, 50 mM NaCl, 50 μM FMN, pH 8.0). Fractions containing the protein of interest were analyzed via SDS-PAGE, then those with >95% purity were combined and concentrated to ~10 mg/mL using 50 kDa cutoff molecular weight centrifuge concentrators (EMD Millipore). Samples were snap-frozen using liquid nitrogen and stored at −80 °C. E1–3G and E2–17G were purchased as solids (Toronto Research Chemicals), resuspended in DMSO, then quantified using HPLC exactly as described previously. 10 5-HTG and DA-G were purchased as solids (Toronto Research Chemicals), then resuspended in water to a concentration of 10 mM. Assays were conducted in 96-well, clear bottom assay plates (Costar) at 37 °C in 50 μL total volume. Reactions consisted of 10 μL of assay buffer (50 mM HEPES, 50 mM NaCl, various pH), 10 μL of enzyme (various concentrations), and 30 μL of 5-HTG or DAG (various concentrations) diluted in assay buffer. The pH of each reaction was chosen based on the optimal pH determined for each GUS with p NPG, and control reactions were performed in which enzyme was substituted with buffer. 93 Reactions were quenched at six intervals with 50 μL of 50% acetonitrile (ACN) / 50% assay buffer. After centrifugation at 13,000 × g for 10 min, the supernatants were analyzed with HPLC to quantify production of active neurotransmitter. The concentration of 5-HTG or DA-G remaining at each time point was quantified on an Agilent 1260 Infinity II liquid chromatograph system. Samples were separated on an Agilent InfinityLab Poroshell 120 C18 column (4.6 × 100 mm, 2.7-μm particle size) at 38 °C with a flowrate of 0.9 mL/min and an injection volume was 40 μL. LC conditions were set at 98% water with 0.1% formic acid (A) for 2 mins and then ramped linearly over 10 min to 98% acetonitrile with 0.1% formic acid (B) and held until 14 min. At 15 min, the gradient was switched back to 100% A and allowed to re-equilibrate until 18 min. 5-HTG, 5-HT, DA-G, and DA were monitored at 280 nm. The concentrations of 5-HT and DA produced were determined from a standard curve (0–200 μM 5-HTG/DA-G in assay buffer). Rate of product formation was determined then plotted against substrate concentration and fit with linear regression in Microsoft Excel to determine catalytic efficiency (k cat /K M ). Final values represent averages of three biological replicates. Thyroxine-glucuronide was purchased as a solid (Toronto Research Chemicals). Assay mixtures contained 10μL GUS (various final concentration), 30μL Thyroxine-G (various concentrations), and 10μL assay buffer (50 mM HEPES, 50 mM NaCl, various pH). Control reactions replaced GUS with buffer. Reactions were quenched at five time points with 50μL 25%TCA. Samples were centrifuged at 16,000 × g in a tabletop centrifuge for 10 minutes. The supernatant was removed and analyzed by HPLC using an Agilent 1260 Infinity II system equipped with an Agilent InfinityLab Poroshell 120 C18 column (4.6 × 100 mm, 2.7 μM particle size). 40 μL sample was injected onto the column at 38°C with a flow rate of 0.9 mL/min. 98% A (water with 0.1% formic acid) and 2% B (acetonitrile with 0.1% formic acid) was flowed for 2 minutes. Conditions were then changed over 8 minutes using a linear gradient from 2% B to 98% B and held at 98% B for 4 minutes. Conditions were ramped back down to 2% B for 1 minute and then held at 2% B for 2 minutes to re-equilibrate the column. An Agilent DAD detector was used to detect analytes at a wavelength of 280 nm. Absorbance was converted to concentration of thyroxine-G using a standard curve of 0–250 μM thyroxine-G. Rate of product formation was determined then plotted against substrate concentration and fit with linear regression in Microsoft Excel to determine catalytic efficiency (k cat /K M ). Final values represent averages of three biological replicates. 5 ng of genomic DNA was processed using the Nextera XT DNA Sample Preparation Kit (Illumina). Target DNA was simultaneously fragmented and tagged using the Nextera Enzyme Mix containing transposome that fragments the input DNA and adds the bridge PCR (bPCR)-compatible adaptors required for binding and clustering in the flow cell. Next, fragmented and tagged DNA was amplified using a limited-cycle PCR program. In this step index 1(i7) and index 2(i5) was added between the downstream bPCR adaptor and the core sequencing library adaptor, as well primer sequences required for cluster formation. The thermal profile for the amplification had an initial extension step at 72°C for 3 min and initial denaturing step at 95°C for 30 sec, followed by 15 cycles of denaturing of 95°C for 10 seconds, annealing at 55°C for 30 seconds, a 30 second extension at 72°C, and final extension for 5 minutes at 72°C. The DNA library was then be purified using Agencourt ® AMPure ® XP Reagent. Each sample was quantified and normalized prior to pooling. For validation of the DNA isolation process, a known bacterial community, ZymoBIOMICS Microbial Community Standard (Cat# D6300), and blanks composed of only DNA isolation reagents were included in the DNA extraction process and again in the library preparation. In addition to the isolation controls, the library preparation also included library blanks composed of library preparation reagents alone. The DNA library pool was loaded on the Illumina platform reagent cartridge (Illumina) and on the Illumina instrument. 152 Sequencing output from the Illumina MiSeq nano 2×150 was converted to FASTQ format and demultiplexed using Illumina Bcl2Fastq 2.18.0.12. Subsequently, the pool ran on NovaSeq S2 PE/ 2×150 was converted to FASTQ format and demultiplexed using Illumina Bcl2Fastq 2.18.0.12. Quality control of the demultiplexed sequencing reads was verified by FastQC. Fecal samples were collected then immediately stored at −80 °C until being processed as previously described and as depicted in Figure 2A . 5 – 10 g of thawed fecal material collected from each donor was resuspended in 25 mL cold extraction buffer (25 mM HEPES pH 6.5, 25 mM NaCl, one Roche Complete EDTA-free protease inhibitor tablet in 50 mL buffer) and 500 mg autoclaved garnet beads then vortexed. Samples were centrifuged at 300 × g for 5 mins at 4 °C and supernatant was collected. 25 mL cold extraction buffer was added to the centrifuged pellet, which was again vortexed and centrifuged. Both supernatants were combined and centrifuged at 300 × g for 5 mins at 4 °C two additional times to further remove insoluble fiber. The supernatant was then sonicated twice on a Fischer Scientific Sonic Dismembrator Model 500 with 0.5 second pulses for 1.5 mins and the lysate was mixed by inversion between each sonication. Lysate was then centrifuged at 17,000 × g for 20 mins at 4 °C to remove insoluble debris then decanted. The lysate was then concentrated with Amicon Ultra 15 mL 30 kDa centrifugal filters and exchanged with fresh extraction buffer three times to remove metabolites. After buffer exchanging, the total protein concentration of the final fecal lysate for each sample was measured with a Bradford assay using purified Escherichia coli β-Glucuronidase as a reference standard. Complex protein lysates were aliquoted at 500 μL then flash frozen in liquid nitrogen and stored at −80 °C until later use in proteomics and fecal lysate assays. Cyclophellitol-based probe JJB397, a biotin-linked covalent inhibitor of GUS enzymes, was provided to us by the laboratory of Hermen Overkleeft from a preparation as described previously. 153 General Proteomics workflow is shown in Figure 2A and was adapted from our previously reported GUS-targeted Activity Based Proteomic Profiling pipeline. 25 , 154 Briefly, 3.5 mg purified fecal extract was incubated with 10 μM biotin-activity-based probe complex in 500 μL extraction buffer with 1% DMSO (final) for 1 hr at 37 °C. 125 μl 10% sodium dodecyl sulfate (SDS) was added to quench the reaction, then samples were heated to 95 °C for 5 min. Samples were then cooled on ice and washed with extraction buffer containing 0.05% SDS three times by centrifugation for 5 min at 13,000 ×g in 1.5 mL 10 K cutoff spin concentrators (Amicon). After centrifugation, the total volume was normalized to 1 mL using extraction buffer + 0.05% SDS. 15 μL streptavidin sepharose beads (GE) were added to the protein mixture, and samples were then incubated at room temperature for 1 h. Afterwards, beads were washed 3 times with 300 μL extraction buffer with 0.1% SDS, three times with 300 μL extraction buffer alone, and finally three times with 300 μL 50 mM NH 4 HCO 3 . Samples were centrifuged at 400 ×g for 2 min at 4 °C between washes, and the supernatant decanted. Beads were then resuspended in 100 μL 50 mM NH 4 HCO 3 and stored at −20 °C, then subjected to subsequent LC-MS/MS analysis exactly as described previously. 25 , 87 Endobiotics were detected by MS/MS using the acquisition parameters and m/z quantification values shown in Table S13 . E1–3G and E2–17G (Toronto Research chemicals) were purchased as solids and suspended in 100% DMSO to a concentration of 10 mM. Assays were conducted in 96-well, clear bottom assay plates (Costar) at a total volume of 50 μL. The reaction consisted of 15 μL assay buffer (50 mM HEPES, 50 mM NaCl, pH 6.5), 5 μL fecal lysate (0.1 mg/mL final), and 30 μL of either E1–3G or E2–17G (200 μM final). After the addition of substrate, reactions were incubated at 37 °C. Five wells were prepared in total, which were individually quenched with an equivalent volume of 25% trichloroacetic acid every 15 mins over the course of 1 hr, beginning at 0 mins. After all wells were quenched, reactions were transferred to epi-tubes then centrifuged at 13,000 × g for 20 minutes. 80 μL of the supernatant was subjected to analysis via liquid chromatography-mass spectrometry (LC-MS/MS). Analysis of the reaction mixture was conducted on an Agilent 1260 Infinity II liquid chromatography system, and separated with an Agilent InfinityLab Poroshell 120 C18 column (4.6 × 100mm, 2.7-μm particle size) at 38 °C. The flow rate was 0.6 ml/min with an injection volume of 5 μL. Reverse-phase gradient elution was conducted using the following solvents: 98% water with 0.012% formic acid and 5 mM ammonium acetate (Solvent A), and methanol (HPLC-grade) with 0.012% formic acid and 5 mM ammonium acetate (Solvent B). Elution gradient was as follows: 90% A/10% B, 0–5 mins: linearly ramp to 10% A/90% B; 5–8 mins linearly ramp to 90% A/10% B, 8–9 mins: 90% A/10% B. After separation, samples were sent to an Agilent 6460 Triple Quad Mass Spectrometer. E1–3G or E2–17G were detected using MS/MS in multiple reaction monitoring mode using negative polarity. The source gas temperature was 325°C at 10 L/min flow rate. The sheath gas temperature was 400°C at 12 L/min. The capillary voltage was −3500V. Nebulizer pressure was 45 psi. The AUC for E1–3-G or E2–17-G was converted to concentration of analyte from a standard curve collected before the start of the reaction (0–200 μM of estrogen-glucuronide in assay buffer). Rate of loss of E1–3-G or E2–17-G over the course of an hour was determined in nM/s. The final value represents the average of three biological replicates. 5-HTG and DA-G were purchased as solids (Toronto Research Chemicals), then resuspended in water to a concentration of 10 mM. Assays were conducted in 96-well, clear bottom assay plates (Costar) at a total volume of 50 μL. Reactions consisted of 10 μL of assay buffer (50 mM HEPES, 50 mM NaCl, pH 6.5), 10 μL of fecal lysate (0.1 mg/mL final), and 30 μL of 5-HTG or DAG (200 μM final) diluted in assay buffer. Reactions were quenched at six intervals over an hour with 50 μL of 50% methanol / 50% assay buffer. After centrifugation at 13,000 × g for 20 min, 80 μL of the supernatant was subjected to analysis via liquid chromatography-mass spectrometry (LC-MS/MS). Analysis of the reaction mixture was conducted on an Agilent 1260 Infinity II liquid chromatography system, and separated with an Agilent InfinityLab Poroshell 120 C18 column (4.6 × 100mm, 2.7-μm particle size) at 38 °C. The flow rate was 0.2 ml/min with an injection volume of 5 μL. Reverse-phase gradient elution was conducted using the following solvents: 98% water with 0.012% formic acid and 5 mM ammonium acetate (Solvent A), and methanol (HPLC-grade) with 0.012% formic acid and 5 mM ammonium acetate (Solvent B). Elution gradient was as follows: 0 mins: 90% A/10% B, 0–3 mins: linearly ramp to 80% A/20% B 3–5 mins linearly ramp to 90% A/10% B, 5–6 mins: 90% A/10% B. After separation, samples were sent to an Agilent 6460 Triple Quad Mass Spectrometer. The source gas temperature was 325°C at 10 L/min flow rate. The sheath gas temperature was 400°C at 12 L/min. The capillary voltage was −3500V. Nebulizer pressure was 45 psi. Production of active neurotransmitter was detected using MS/MS in multiple reaction monitoring mode using positive polarity. The AUC of DA or 5-HT was converted to concentration of analyte using a standard curve collected before the start of the reaction (0–200 μM of neurotransmitter-glucuronide in reaction conditions). Rate of production of 5-HT or DA over the course of an hour was determined in nM/s. The final value represents the average of three biological replicates. Thyroxine-G was purchased as a solid (Toronto Research Chemicals). Assay mixtures contained 10μL fecal protein lysate (0.1 mg/mL final), 30μL Thyroxine-G (200 μM final), and 10μL assay buffer (50 mM HEPES, 50 mM NaCl, pH 6.5). Control reactions replaced GUS with buffer. Reactions were quenched at five time points with 50μL 25%TCA. Samples were centrifuged at 16,000 × g in a tabletop centrifuge for 10 minutes. The supernatant was removed and analyzed by HPLC using an Agilent 1260 Infinity II system equipped with an Agilent InfinityLab Poroshell 120 C18 column (4.6 × 100 mm, 2.7 μM particle size). 40 μL sample was injected onto the column at 38°C with a flow rate of 0.9 mL/min. 98% A (water with 0.1% formic acid) and 2% B (acetonitrile with 0.1% formic acid) was flowed for 2 minutes. Conditions were then changed over 8 minutes using a linear gradient from 2% B to 98% B and held at 98% B for 4 minutes. Conditions were ramped back down to 2% B for 1 minute and then held at 2% B for 2 minutes to re-equilibrate the column. An Agilent DAD detector was used to detect analytes at a wavelength of 280 nm. Absorbance was converted to concentration of thyroxine-G using a standard curve of 0–250 μM thyroxine-G, and final values represent averages of three biological replicates. 4-Methylumbelliferyl glucuronide (4MU-G) was purchased as a solid (Sigma Aldrich) and resuspended in water to a concentration of 50mM. UNC10201651, UNC10201652, norquetiapine, vortioxetine, ceritinib, mefloquine, 3-OH Desloratadine, desloratadine, paroxetine, crizotinib, palbociclib, and piperazine were obtained as solids and suspended in 100% DMSO at various concentrations above 20 mM. Each compound was then diluted in ddH 2 O and an equivalent % DMSO for all final concentrations. Assays were conducted in Costar halfarea 96-well assay plates at a total volume of 50 μL. The reaction consisted of 10 μL assay buffer (125 mM HEPES, 125 mM NaCl, pH 6.5), 5 μL of purified GUS enzyme (15 nM of Ec , Sa , Cp , Ee , Gemmiger L1 , Rh2 , Rg3 , FpL2–6 , and Gemmiger FMN , and 60 nM final of Ri ), 5 μL inhibitor (at various concentrations), and 30 μL 4MUG (100 μM final). After the addition of inhibitor, the reaction was incubated for 5 mins at 37 °C, then initiated by the addition of 4MUG and incubated at 37 °C for 1 hr. The reaction was then quenched with 50 μL sodium carbonate (0.2M). Analysis of the reaction mixture was performed using a CLARIOstar Plus Microplate Reader (BMG Lab Tech), measuring the fluorescence of 4MU at excitation 350nm / emission 450 nm. 155 The fluorescence data collected was converted to percent inhibition as described previously. 49 Results are shown in Figure 3B and Figure S5 . Reaction mixtures contained 5 μL fecal extract (0.1 mg/mLfinal), 10 μL 4MU-G (100 μM final), 5 μL inhibitor (10 μM final), and 30 μL assay buffer (25 mM HEPES, 25 mM NaCl, pH 6.5). Reactions were incubated with inhibitor for five minutes, initiated by the addition of 4MU-G, then quenched after 1 hr with 50 μL sodium carbonate (0.2M). Analysis of the reaction mixture was performed using a CLARIOstar Plus Microplate Reader (BMG Lab Tech), measuring the fluorescence of 4MU at excitation 350nm / emission 450 nm. The absorbance data collected was converted to percent inhibition as described previously. 49 Results are shown in Figure 4B . Crystals of Ee GUS bound to UNC10201652 glucuronide were produced via the sitting drop vapor diffusion method using an Oryx Nano. Ee GUS at 11.5 mg/mL was preincubated with 10-fold molar excess UNC10201652 for one hour, then p NPG was added and the mixture was cooled on ice before adding into the crystalline solution. Crystals were formed by incubating ligand bound Ee GUS in 0.07 M BICINE, pH 9.0, 1.4 % (w/v) 1,4-Dioxane, 7 % (w/v) PEG 20,000, and 30 % (v/v) glycerol. The mother liquor acted as cryoprotectant for the crystals. Protein crystals were removed directly from the crystallization drop then quickly flash cooled (< 1 minute) in liquid nitrogen. Structure is publicly accessible at PDB 8GEN. Crystals of Rh2 GUS bound to UNC10201652 glucuronide were produced via the hanging-drop vapor diffusion method. Rh2 GUS at 15.8 mg/mL was preincubated with 10-fold molar excess UNC10201652 for one hour, then p NPG was added and the mixture was cooled on ice before adding into the crystalline solution. Crystals were formed by incubating ligand bound Rh2 GUS in 0.17 M Ammonium Acetate, 0.085M Sodium Citrate:HCl, pH 5.6 and 25.5 % (w/v) PEG 4000, 15 % (v/v) glycerol. The crystals were cryoprotected using mother liquor + 20% (v/v) glycerol. Protein crystals were transferred directly from the crystallization drop to the cryoprotectant solution and then quickly flash cooled (< 1 minute) in liquid nitrogen. Structure is publicly accessible at PDB 8GES. Crystals of Ee GUS bound to Norquetiapine glucuronide were produced via the hanging-drop vapor diffusion method. Ee GUS at 11.5 mg/mL was preincubated with mefloquine (6-fold molar excess) and 4-MU-G (4-fold molar excess) for 1 hour at 37°C prior to addition into the crystalline solution. Crystals were formed by incubating ligand bound Ee GUS in 10% PEG400 and 0.1 M sodium acetate pH = 5.0. The crystals were cryoprotected using 35% (w/v) PEG400 and 0.1 M sodium acetate pH = 5.0. Protein crystals were transferred directly from the crystallization drop to the cryoprotectant solution and then quickly flash cooled (< 1 minute) in liquid nitrogen. Structure is publicly accessible at PDB 8GER. Crystals of Ee GUS bound to 3OH-Desloratidine glucuronide were produced via the sitting drop vapor diffusion method using an Oryx Nano. Ee GUS at 11.5 mg/mL was preincubated with 10-fold molar excess 3OH-Desloratidine for one hour, then p NPG was added and the mixture was cooled on ice before adding into the crystalline solution. Crystals were formed by incubating ligand-bound Ee GUS in 0.07 M Sodium cacodylate trihydrate, pH 6.5, 0.98 M sodium acetate trihydrate, and 30 % (v/v) glycerol. The mother liquor acted as cryoprotectant for the crystals. Protein crystals were removed directly from the crystallization drop then quickly flash cooled (< 1 minute) in liquid nitrogen. Structure is publicly accessible at PDB 8GEO. Crystals of Ee GUS bound to Ceritinib glucuronide were produced via the sitting drop vapor diffusion method using an Orxy Nano. Ee GUS at 11.5 mg/mL was preincubated with 10-fold molar excess Ceritinib for one hour, then p NPG was added and the mixture was cooled on ice before adding into the crystalline solution. Crystals were formed by incubating ligand bound Ee GUS in 0.07 M BICINE, pH 9.0, 1.4 % (w/v) 1,4-Dioxane, 7 % (w/v) PEG 20,000, and 30 % (v/v) glycerol. The mother liquor acted as cryoprotectant for the crystals. Protein crystals were removed directly from the crystallization drop then quickly flash cooled (< 1 minute) in liquid nitrogen. Structure is publicly accessible at PDB 8GEQ. Crystals of Rh2 GUS bound to Norquetiapine glucuronide were produced via the hanging-drop vapor diffusion method. Rh2 GUS at 15.8 mg/mL was preincubated with 10-fold molar excess Norquetiapine for one hour, then p NPG was added and the mixture was cooled on ice before adding into the crystalline solution. Crystals were formed by incubating ligand bound Rh2 GUS in 0.17 M Ammonium Acetate, 0.085M Sodium Citrate:HCl, pH 5.6 and 25.5 % (w/v) PEG 4000, 15 % (v/v) glycerol. The crystals were cryoprotected using mother liquor + 20% (v/v) glycerol. Protein crystals were transferred directly from the crystallization drop to the cryoprotectant solution and then quickly flash cooled (< 1 minute) in liquid nitrogen. Structure is publicly accessible at PDB 8GET. Diffraction data for all crystals was collected at 100 K at GM-Ca-CAT 23-ID-B and 23-IDD (Advanced Photon Source, Argonne National Laboratory). CCP4i (v7.1.015) was used to scale the raw data, then Coot (v0.9.8.7) and Phenix (v.1.17) and were used to visualize and refine the model, respectively, to the statistics shown in Table S11 . 156 , 157 , 158 Structures shown in manuscript were generated using The PyMOL Molecular Graphics System, Version 2.1 Schrödinger, LLC (PyMOL). Frozen samples of serum and cecal contents were fully thawed on ice, then 500 nM (final) 5-HT-D 4 (Cayman) was added to the samples as internal standards for 5-HT and 5-HT-G. The samples were vortexed to uniformly distribute deuterated standards, then ~ 100 μL garnet beads (Omni International) were added and the tubes were vortexed, then chilled on ice for 5 mins. The samples were next homogenized with 0.5 second pulses for a total of 2 min at 30 Hz using a Qiagen Tissuelyzer II, using pre-chilled metal blocks. Afterwards, the samples were incubated on ice for 5 mins, then sonicated in a water bath at 4 °C for 2 mins, inverted, then sonicated again in the same cycle. The samples were then spun in a pre-chilled (4 °C) centrifuge at 13,000 × g for 20 mins. The supernatant was decanted, and small-molecule metabolites were purified using a pre-equilibrated Amicon Ultra 0.5 mL centrifugal filter with a 3 kDa molecular weight cutoff pore size (Millpore Sigma). The filter was pre-equilibrated with water, and the samples were centrifuged at 13,000 × g for 20 mins. The flow-through was collected, and the filters containing samples were again centrifuged at 13,000 × g for 20 mins. The flow-through was collected and combined with the previous flowthrough in a 1.7-mL microcentrifuge tube, which was then centrifuged at 13,000 × g for an additional 20 mins. Afterwards, 150 uL was decanted from the top-center of the supernatant then flash-frozen in liquid nitrogen for future analysis by HPLC-MS/MS. Analysis of the reaction mixture was conducted on an Agilent 1260 Infinity II liquid chromatography system, and separated with an Agilent InfinityLab Poroshell 120 C18 column (4.6 × 100mm, 2.7-μm particle size) at 38 °C. The flow rate was 0.2 ml/min with an injection volume of 5 μL. Reverse-phase gradient elution was conducted using the following solvents: 98% water with 0.012% formic acid and 5 mM ammonium acetate (Solvent A), and methanol (HPLC-grade) with 0.012% formic acid and 5 mM ammonium acetate (Solvent B). Elution gradient was as follows: 0 mins: 90% A/10% B, 0–3 mins: linearly ramp to 80% A/20% B 3–5 mins linearly ramp to 90% A/10% B, 5–6 mins: 90% A/10% B. After separation, samples were sent to an Agilent 6460 Triple Quad Mass Spectrometer. MS/MS was operated in dynamic multiple reaction monitoring mode using positive polarity. The source gas temperature was 325°C at 10 L/min flow rate. The sheath gas temperature was 400°C at 12 L/min. The capillary voltage was −3500V. Nebulizer pressure was 45 psi. AUC values were converted to fold-change as represented in Figure 4I and 4J . Levels of 5-HT, 5-HT-G, and the deuterated internal standard were determined using the acquisition parameters and m/z quantification values shown in Table 15. Raw metagenomics files were trimmed, filtered, and annotated, then assembled into gene and protein sequences using Metagenomics Analysis Toolkit (MOCAT2 v2.0.1). 159 To determine the relative abundance of bacterial taxa for each sample, paired-end reads were analyzed using Metaphlan (v3.0.13) and results for Class relative abundance were graphed using ggplot2 (v3.3.5) in R (v4.1.2). 160 , 161 Metagenome-derived amino acid sequences were each aligned pairwise to 17 representative GUS enzymes with reported crystal structures using Protein-Protein BLAST (BLASTP v2.5.0+). 162 Candidate sequences with ≥ 25 % identity to any representative GUS enzyme were then assessed for the presence of 7 conserved residues. 22 Sequences that both met the identity threshold and contained all 7 conserved residues were accepted as putative GUS enzymes. Accepted sequences were filtered for redundancies at a sequence identity threshold of 100% using CD-HIT (v4.8.1), and the output was used to form a representative set of unique GUS sequences for downstream analysis. 163 Accepted sequences were aligned to representative sequences from each loop class in a Multiple Sequence Alignment (MSA) using Clustal Omega (v1.2.4), and GUS class was assigned according to parameters reported previously and shown in Figures S1 and S4 . 21 , 22 , 164 The resulting GUS classes were further screened for “No Loop” class enzymes which conserve both a c-terminal domain with previously reported FMN-binding GUS (PDB: 6MVF, 6MVG, 6MVH). 76 , 164 Resulting sequences were assigned the class “FMN”. Taxonomy was assigned to representative GUS sequences by mapping queries to the Unified Human Gastrointestinal Protein (UHGP) catalog using Diamond (v2.0.15.153) as reported previously, and resulting taxonomic identifiers were used to rename these sequences. 22 , 162 , 165 , 166 Unique GUS genes found in each sample were counted and graphed using ggplot2 (v3.3.5) in R (v4.1.2). Phylogenetic inference was performed on representative sequences using Muscle5 and IQ-TREE 2. 167 , 168 The resulting Newick tree file was combined with the annotations for GUS class and taxonomy to create cladograms using ggtree (v3.2.1) and ggplot2 in R ( Figure 2D ). 161 , 169 , 170 We compared the results of our structural metagenomics functional annotations to those generated by the eggnog-mapper web server ( http://eggnog-mapper.embl.de/ ), which uses traditional homology-based annotation. 96 Data were processed using Metalab (v1.1.148) with MaxQuant (v1.6.2) to identify peptides and protein groups. 171 , 172 A sample-specific database was derived from the cohort metagenome then combined with the UniProtKB/Swiss-Prot human sequence database to be used as the database search. 97 Search parameters were static carbamidomethyl cysteine modification, specific trypsin digestion with up to two missed cleavages, variable protein N-terminal acetylation and methionine oxidation, and match between runs. A false discovery rate (FDR) of 1% was used for filtering protein identifications, and potential contaminants and decoys were removed. Best-match protein headers were mapped back to their corresponding amino acid sequences from the sample metagenomes, and GUS enzymes were identified as described in “ Identification and Characterization of GUS Sequences ”. To correct for possible misclassification of GUS gene fragments, structural classes of GUS sequences detected in metaproteomics were reassigned based on the structural class of the most similar UniProt GUS sequence. Raw intensities in Table S3 were log 2 -transformed to reach the normalized abundances shown in Figure 2 . Abundances of total GUS proteins were summed by structural class to predict sample reactivation or inhibition. Correlational analysis based on several RMSD thresholds sought to identify the largest number of unique, highly similar proteins of the same GUS class whose combined abundance can predict rates of reactivation or inhibition. Groupings which maximized the number of individual donors included in regression (≥ 6) and the number of substrates with significant regressions for the grouping (≥ 2) were selected. P values for all reflect confidence in a slope that is significantly non-zero as determined by the Wald Test, and For all proteins with an UniProt Knowledgebase sequence match meeting a percent identity ≥ 95%, models were obtained from the Alphafold Protein Structure Database ( https://alphafold.ebi.ac.uk ). 173 , 174 , 175 If a protein model was not available meeting this percent identity threshold, the AlphaFold model was predicted using ColabFold v1.5.2. 176 Protein models were then loaded into PyMOL and residues with predicted local distance difference test (pLDDT) scores < 50% were removed. For each pairwise protein comparison, root mean square deviation (RMSD) values were determined using the cealign function in PyMOL. To identify whether meta-omic features may be associated with GUS inhibition in fecal lysates described in “ Complex Lysate GUS Inhibition Assays ”, Sparse Partial Least Squares discriminant analysis (sPLS-DA) was performed on metagenomic and metaproteomic sample quantitation using mixOmics (v6.18.1). 177 Samples were assigned “High or Low Inhibition” status based on whether their inhibition percentages were greater than or equal to the median Z score. The metadata for all samples was joined to their metagenomic or metaproteomic quantitation, using the Genus-level relative abundance as described in “ Preliminary Metagenomics Analysis ” and the individual GUS protein intensity described in “ Metaproteomics Data Analysis ” respectively. The sPLS-DA method was applied to yield the sample plots presented in Figure 4B and 4E . All data was accessed and downloaded from the Inflammatory Bowel Disease Multi’Omics Database portal ( https://ibdmdb.org/ ). 121 Processed fecal metabolomics data was downloaded and presence of glucuronidated metabolites was assessed, only ethyl-glucuronide was identified. Fecal whole-genome sequencing files for donors corresponding to metabolomics sample donors were downloaded and bacterial abundances were determined using Metaphlan4. 178 Abundances for each donor were assessed alongside known GUS-producing species 76 to determine the percentage of total bacterial composition that could produce GUS enzymes. These numbers were considered as a function of ethyl-glucuronide intensity to generate the regressions shown in Figure S7 . P values for each reflect significance that the slope is non-zero as determined by the Wald Test.

The human gut microbiota processes a range of xenobiotics including dietary compounds, industrial chemicals, and small-molecule drugs; however, few connections have linked gut microbial enzymes and endobiotic compounds in the gastrointestinal (GI) tract. 1 , 2 , 3 , 4 , 5 Endobiotics including neurotransmitters and hormones with significant gut residence time coordinate an array of homeostatic functions and their dysregulation contributes to the onset of cancer, heart, liver and endocrine disorders, as well as psychiatric conditions like depression and anxiety ( Table 1 ). 6 , 7 Endobiotics, like xenobiotics, are inactivated by glucuronidation and trafficked to the gut for excretion. 8 , 9 , 10 Endobiotic-glucuronide conjugates in the intestinal lumen are subject to reactivation by β-glucuronidase (GUS) enzymes expressed by commensal microbiota that release the active parent compound in the GI tract and facilitating enterohepatic recirculation. 7 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 GUS enzymes are one of the best characterized protein families in the human microbiome; metagenomic analyses paired with structure, function, and inhibition studies have defined hundreds of functionally diverse GUS orthologues categorized into eight unique structural classes ( Figure S1 ). 10 , 21 , 22 , 23 Specific structural classes of GUS enzymes efficiently reactivate small-molecule xenobiotics including cancer drugs, NSAIDs, and consumer product toxins ( i.e. , Loop 1, Mini-Loop 1, FMN), while others act only on larger polysaccharides like heparan- and chondroitin-sulfates ( i.e. , Loop 2, No Loop). 10 , 24 , 25 , 26 , 27 , 28 To date, gut microbial GUS enzymes have been studied primarily in the context of xenobiotic substrates that induce GI toxicity upon reactivation. It is comparatively less clear how these enzymes process chemically diverse endobiotics such as neurotransmitters, 7 , 29 , 30 , 31 , 32 , 33 , 34 thyroid hormones, 35 , 36 , 37 , 38 and sex hormones, 10 , 39 , 40 , 41 all of which are inactivated by host UDP-glucuronosyltransferases (UGTs) and sent to the gut as inactive glucuronides. We recently examined gut microbial GUS enzymes in vitro for their abilities to process the glucuronides of estrone (E1) and estradiol (E2). 10 , 42 While these hormones are fundamental to development and reproduction, they also drive disorders including hormone-dependent cancers and endometriosis. 10 , 42 , 43 , 44 , 45 , 46 , 47 , 48 We found that only a subset of gut microbial GUS enzymes efficiently reactivate E1 and E2, indicating that these bacterial proteins exhibit substrate-level preferences for both endobiotic and xenobiotic conjugates. 49 The differential reactivation of hormones and neurotransmitters by unique compositions of gut microbial GUS enzymes has the potential to explain aspects of endobiotic dysregulation associated with disease. 50 , 51 Like estrogens, the reactivation of neurotransmitters and thyroid hormones by gut microbial GUS enzymes has the potential to impact homeostasis. For example, serotonin (5-hydroxytryptamine; 5-HT) and dopamine (3,4-dihydroxyphenethylamine; DA) 44 , 66 , 67 , 68 are neurotransmitters that are primarily stored in the GI tract 29 , 52 where they enable crosstalk between the enteric and central nervous systems (ENS, CNS). 7 , 29 , 30 , 31 , 32 , 33 , 34 , 53 Neurotransmitters and their metabolites produced by the gut microbiota are absorbed from the GI tract and into systemic circulation. 14 , 18 , 54 , 55 , 56 , 57 This network, termed the gut-brain axis, enables gut microbial metabolites to modulate physiology and behavior, in turn potentially contributing to disease states. Indeed, aberrant serotonin and dopamine serum levels are implicated in disorders ranging from Alzheimer’s Disease (AD) and Parkinson’s Disease (PD) to depression and anxiety, 17 , 54 , 58 , 59 , 60 , 61 , 62 diseases that are also linked to differences in gut microbial composition. 63 , 64 , 65 , 66 , 67 Similarly, the thyroid hormone thyroxine (T4), which regulates growth, development, and metabolism, is subject to reactivation in the gut by microbial GUS enzymes. 68 , 69 More than 5% of Americans (~20 million) suffer from either hyperthyroidism or hypothyroidism, both of which contribute to diseases characterized by weight fluctuations, immune dysregulation, and changes in mood. 70 , 71 Serum abundance of thyroxine has been shown to be regulated in a gut microbiota-dependent manner, leading to the proposal of a gut-thyroid axis and indicating that uncharacterized actions by the gut may contribute to thyroid-linked disease etiology. 72 , 73 , 74 Targeted gut microbial GUS inhibitors have been shown to alleviate GI damage caused by anticancer and therapeutic drugs, and to prevent consumer toxin-induced colitis. 75 , 76 , 77 , 78 , 79 , 80 , 81 In a key example, GUS inhibitors dramatically improved the antitumor efficacy of irinotecan in mice by preventing the dose-limiting gut damage of this clinically prevalent chemotherapeutic agent. The most potent inhibitors to date contain piperazine moieties that form a covalent inhibitor-N-glucuronide that remains at the GUS active site. 25 , 49 Numerous FDA-approved drugs for a range of conditions also contain piperazine or piperidine moieties ( Table 2 ), several of which also contain pharmacophores and core scaffolds similar to that of mechanism-based GUS inhibitors. 82 These drugs reach the lower GI at concentrations exceeding 10 μM and cause adverse GI effects associated with hormone and neurotransmitter dysregulation through mechanisms that remain poorly understood. 1 , 10 , 27 , 49 , 82 , 83 , 84 , 85 , 86 It has been hypothesized that some of the clinical effects and off-target effects of these medications may arise from their inhibition of gut microbial GUS enzymes. 10 , 76 , 97 , 98 Recent studies have identified human fecal GUS proteins that are responsible for drug and toxin reactivation. 25 , 87 , 88 Here, we use in vitro, ex vivo, in vivo , and multi-omic approaches to investigate the role of human fecal microbial GUS enzymes in endobiotic reactivation. We pinpoint structurally and functionally unique GUS proteins that process chemically diverse endobiotic-glucuronide substrates and show that FDA-approved drugs disrupt these reactions in vitro and ex vivo . We also establish that gut microbial GUS inhibition alters serum and intestinal serotonin levels in mice. These results significantly further our understanding of the roles gut microbial enzymes play in mediating endobiotic levels and demonstrate the extent to which drugs may interact with the microbiota to alter host homeostasis.

Table S1. Metagenomic relative taxonomic abundance (Metaphlan3), Related to Figures 2 and 4 . Table S2. Detailed GUS gene annotation, Related to Figures 2 and 4 . Table S3. Detailed GUS proteomic intensities, Related to Figures 2 and 4 . Table S4. GUS protein intensity summed by total GUS and structural class, Related to Figures 2 and 4 . Table S5. Quantitation of GUS activity and inhibition, Related to Figures 2 and 4 . Table S6. Activity and inhibition as a function of GUS protein intensity, by GUS class and total GUS, Related to Figures 2 and 4 . Table S7. T-test results comparing GUS activity or inhibition with metaproteomic GUS abundance at each taxonomic level, Related to Figures 2 and 4 . Table S8. Root-mean square deviation (RMSD) matrix comparing all GUS proteins detected in proteomics, Related to Figures 2 - 4 . Table S9. Activity and inhibition as a function of GUS protein intensity, by GUS class, all groups with RMSD ≤ 2Å, Related to Figures 2 and 4 . Table S10. T-test results comparing GUS activity or inhibition with bacterial abundance at each taxonomic level, Related to Figures 2 and 4 . Table S11. Summary of resolved GUS crystal structures and their associated statistics, , Related to Figure 3 , Figure S5D – F , and STAR Methods . Table S12. Cohort demographics, Related to STAR Methods . Table S13. Acquisition parameters for endobiotic detection using HPLC-MS/MS, Related to STAR Methods .