Research Article

Separating host and microbiome contributions to drug pharmacokinetics and toxicity

See allHide authors and affiliations

Science  08 Feb 2019:
Vol. 363, Issue 6427, eaat9931
DOI: 10.1126/science.aat9931

Off-target drug metabolism

Anything humans swallow is exposed to the foraging and transforming activities of the gut microbiota. This applies to therapeutic drugs as well as food components and can be a major source of interpersonal variation in drug efficacy and toxicity. Zimmermann et al. found that individual drug responses depend on the genetics of an individual's microbiota. They explored the metabolism of nucleoside drugs (which are used as antivirals and antidepressants) in mice inoculated with a variety of mutant microbiota. They then modeled the pharmacokinetics in different body compartments and identified the host and microbe contributions. In some individuals, up to 70% of drug transformation can be ascribed to microbial metabolism.

Science, this issue p. eaat9931

Structured Abstract

INTRODUCTION

The gut microbiota is implicated in the metabolism of many medical drugs, with consequences for interpersonal variation in drug efficacy and toxicity. However, quantifying microbial contributions to drug metabolism in vivo is challenging, particularly in cases where host and microbiome perform the same metabolic transformation. A quantitative understanding of the physiological, chemical, and microbial factors that determine microbiome contributions to drug metabolism could help explain interpersonal variability in drug response and provide opportunities for personalized medical treatments.

RATIONALE

To experimentally dissect microbiome and host drug metabolism, we combined gut commensal genetics with gnotobiotics to measure metabolism of the nucleoside analog brivudine (BRV) across tissues in mice that vary in a single microbiome-encoded enzyme. Informed by these measurements, we built a pharmacokinetic model to quantitatively predict microbiome contributions to systemic drug and metabolite exposure. Model simulations evaluate the impact of oral bioavailability, host and microbial drug-metabolizing activity, metabolite absorption, and intestinal transit on microbiome contributions to drug metabolism. To test the general applicability of this approach, we performed additional studies with the benzodiazepine clonazepam to quantitatively untangle microbiome contributions to metabolism of a drug subject to multiple metabolic routes and transformations.

RESULTS

We demonstrate BRV conversion to hepatotoxic bromovinyluracil (BVU) by both mammalian and microbial enzymes and reduced systemic BVU exposure in germ-free mice, suggesting a microbiome contribution to serum BVU. Drug conversion assays with axenic cultures and an arrayed transposon library identified BRV-metabolizing gut bacteria and responsible gene products. This enabled us to establish mouse models that are isogenic except for a single bacterial gene responsible for microbial BRV metabolism. Administration of oral BRV and quantification of drug and drug metabolite kinetics in different body compartments provided the data to develop a host-microbiome pharmacokinetic model. This model accurately predicts serum BVU exposure and quantifies host and microbiome contributions to its pharmacokinetics. Model simulations revealed how drug, host, and microbial parameters affect host-microbiome drug metabolism.

To test whether this approach applies to other microbiome-metabolized drugs, we quantified microbiome and host contributions to the metabolism of sorivudine, which is structurally related to BRV but is metabolized to BVU at different rates by both host and microbiome. We also quantified microbiome and host contributions to serum clonazepam metabolites produced through oxidation, nitroreduction, glucuronidation, and enterohepatic cycling.

CONCLUSION

This study provides an experimental and computational strategy to untangle host and microbial contributions to drug metabolism. Quantitative understanding of the interplay between host and microbiome-encoded metabolic activities will clarify how nutritional, environmental, genetic, and galenic factors affect drug metabolism and could enable tailored intervention strategies to improve drug responses. This approach could also be adapted to other xenobiotics, food components, and endogenous metabolites.

Experimental and computational approaches that untangle host and microbial contributions to drug metabolism.

Oral drugs are administered to gnotobiotic mice that differ in a single microbial drug-metabolizing enzyme (GNMUT, GNWT); drug and drug metabolite kinetics are then quantified across tissues. A microbiome-host pharmacokinetic model developed from these measurements accurately predicts serum metabolite exposure and untangles host and microbiome contributions to drug metabolism.

Abstract

The gut microbiota is implicated in the metabolism of many medical drugs, with consequences for interpersonal variation in drug efficacy and toxicity. However, quantifying microbial contributions to drug metabolism is challenging, particularly in cases where host and microbiome perform the same metabolic transformation. We combined gut commensal genetics with gnotobiotics to measure brivudine drug metabolism across tissues in mice that vary in a single microbiome-encoded enzyme. Informed by these measurements, we built a pharmacokinetic model that quantitatively predicts microbiome contributions to systemic drug and metabolite exposure, as a function of bioavailability, host and microbial drug-metabolizing activity, drug and metabolite absorption, and intestinal transit kinetics. Clonazepam studies illustrate how this approach disentangles microbiome contributions to metabolism of drugs subject to multiple metabolic routes and transformations.

Individuals can vary widely in drug response. Most drugs are delivered orally, and more than 70% exhibit low solubility, low permeability, or both (1). These drugs likely encounter commensal microbes at densities exceeding 108 cells/ml in the small intestine and 1011 cells/ml in the large intestine (2). Gut microbes collectively encode 150-fold more genes than the human genome, including a rich repository of enzymes with the potential to metabolize drugs and hence influence their pharmacology. Cryptic microbial contributions to drug metabolism, in which host and microbiota produce the same metabolite, are particularly challenging to quantify (fig. S1A). We used measurements of drug and metabolite levels, collected over time and across tissues from gnotobiotic mice carrying no microbiota, genetically manipulated gut commensals, or a complex microbial community, to build a pharmacokinetic model that quantitatively disentangles host and microbiome contributions to drug metabolism.

Brivudine metabolism by host and microbiota

Brivudine (BRV) is an oral antiviral drug that is metabolized to bromovinyluracil (BVU) (Fig. 1A) by both host and microbiota. Indeed, incubation of human and murine S9 liver fractions and unfractionated fecal microbial communities with BRV leads to stoichiometric conversion to BVU, confirming that both liver and microbiota are capable of this enzymatic transformation (Fig. 1, B and C, and tables S1 and S2). Next, we compared serum kinetics of BRV and BVU in conventional (CV) and germ-free (GF) mice after oral BRV administration. CV mice accumulated five times as much BVU in serum as their genetically identical GF counterparts, without a corresponding decrease in serum BRV concentration, suggesting an intestinal (microbial) contribution to serum BVU (Fig. 2A and tables S3 to S7).

Fig. 1 BRV to BVU conversion in vitro by host and microbiome.

(A) Chemical structure of BRV and BVU. (B) Enzymatic conversion of BRV to BVU by human and murine S9 liver fractions. Shaded areas represent SD (n = 5). (C) In vitro conversion of BRV to BVU by human and murine gut microbial communities. Lines and shading represent mean (n = 4 experimental replicates) and SD (n = 16; 4 biological x 4 experimental replicates), respectively.

Fig. 2 BRV metabolism by GF and CV mice.

(A) BRV and BVU serum kinetics in CV and GF mice. (B) Intestinal BRV and BVU concentrations over time; each field represents the mean of five animals. (C) Cecal BRV and BVU concentrations in individual animals. (D) Total amount of BRV and BVU in cecum and feces. (E) Liver concentrations of BRV and BVU. (F) Liver thymine. For all mouse data, horizontal lines show the mean of five animals and times reflect hours after oral BRV administration. SI: duodenum; SII: jejunum; and SIII: ileum. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 [t test with false discovery rate (FDR) correction for multiple hypotheses testing].

To directly investigate microbial BVU generation in vivo, we quantified BRV and BVU concentrations along the intestinal tract over time (Fig. 2B). CV and GF mice exhibit similar BRV kinetics in the duodenum; by contrast, GF mice maintain significantly higher BRV levels further along the gastrointestinal tract and in feces. BVU levels exhibit the opposite pattern, with increased intestinal concentrations in CV mice as compared to GF controls (Fig. 2C). Because GF animals have a larger cecum than their CV counterparts, we compared the absolute amounts (rather than concentrations) of BRV and BVU in the large intestine. The quantity of BVU in the feces of CV mice is insufficient to account for the amount of intestinal BRV metabolized, consistent with absorption of microbiota-derived BVU from the intestine into circulation (Fig. 2D and fig. S1, B and C).

The increased concentration of serum BVU in CV as compared to GF mice is paralleled by increased BVU concentrations in the liver (Fig. 2E). BVU interferes with human pyrimidine metabolism by covalently binding to dihydropyrimidine dehydrogenase (DPD) in the liver, with lethal consequences for patients administered chemotherapeutic pyrimidine analogs such as 5-fluorouracil (5-FU) (3, 4). BRV-treated CV mice also accumulate more endogenous DPD substrates [e.g., thymine (4)] in the liver compared to their GF counterparts, illustrating the contribution of the microbiota to toxicity without 5-FU coadministration (Fig. 2F) (5).

Identification of brivudine-metabolizing gut bacteria and gene products

We next sought to directly quantify the contribution of microbial drug metabolism to serum drug and metabolite exposure by specifically modulating this activity in otherwise identical mice. To this end, we first determined the capacity of eight individual bacterial species, representing five major phyla that dominate the mammalian gut microbiota (6), to convert BRV to BVU (Fig. 3A and tables S8 and S9). Of these species, Bacteroides thetaiotaomicron and B. ovatus possess the highest metabolic activity, consistent with previous reports that members of this genus can metabolize the structurally similar drug sorivudine (7). To identify BRV-metabolizing enzymes in Bacteroides, we condensed a mapped, arrayed library of B. thetaiotaomicron transposon mutants (8) to eliminate redundancy, resulting in 1290 strains that collectively disrupt expression of 2350 genes [~75% of predicted nonessential genes (8)] (fig. S2 and table S10). We tested each of these strains for the ability to convert BRV to BVU and identified a single mutant, carrying a transposon insertion in bt4554, that exhibits a loss-of-function phenotype (Fig. 3B). Targeted gene deletion, complementation at different expression levels, and enzyme assays with purified protein confirmed that bt4554, encoding a predicted purine nucleoside phosphorylase (9) that is also present in B. ovatus and conserved among Bacteroidetes (tables S11 and S12), is necessary and sufficient for BRV metabolism and that its expression is rate-limiting (Fig. 3C and fig. S3).

Fig. 3 Identification of a microbiome-encoded enzyme responsible for BRV metabolism.

(A) BRV conversion to BVU by representative human gut isolates. (B) Log2 fold change of BRV and BVU concentrations of B. thetaiotaomicron transposon insertion mutants (blue, n = 1290) compared to media controls (gray, n = 83) after 24 hours of incubation. Each point represents one strain, sorted along the x axis in the same order in top (BRV) and bottom (BVU) panels. Mean fold changes and 95% prediction intervals for controls and strains are indicated by solid lines and shaded areas, respectively. (C) BRV conversion by B. thetaiotaomicron WT (n = 4), bt4554 mutant (n = 4), and complemented strains expressing bt4554 at different levels (n = 8). In (A) and (C), lines and shaded areas depict the mean and SD of independent cultures (n = 4 to 8).

Brivudine metabolism in mice that vary in a single microbiome-encoded enzyme

B. thetaiotaomicron wild-type (WT) and bt4554 mutant strains exhibit comparable growth rates in vitro and colonize GF mice at similar levels (fig. S4, A and B). Administration of BRV to gnotobiotic (GN) mice monocolonized with WT (GNWT) or bt4554 mutant bacteria (GNMUT) results in indistinguishable BRV serum kinetics, consistent with the physiological similarity between these animals and further suggesting that microbial BRV metabolizing activity in the intestine does not influence BRV bioavailability or systemic elimination. By contrast, serum BVU exposure is significantly higher in GNWT as compared to GNMUT animals [area under the curve (AUC) ratio = 2.4, p < 0.001; Fig. 4A, fig. S5A, and table S7]. GNWT mice also exhibit increased BVU levels and thymine accumulation in the liver after BRV administration (Fig. 4, B and C). As observed in comparisons between CV and GF animals, increased systemic BVU exposure in GNWT mice is paralleled by significant intestinal BRV metabolism (Fig. 4D and fig. S5B). Because other aspects of host physiology, such as cecum size and intestinal transit time, are matched between GNWT and GNMUT animals, intestinal drug and metabolite concentrations can be directly compared and balanced. This reveals that WT B. thetaiotaomicron completely metabolizes cecal BRV, and the resulting BVU is almost entirely absorbed from both cecum and colon. By contrast, BRV is poorly absorbed from the lower intestine, and GNMUT mice excrete the drug in feces (Fig. 4E).

Fig. 4 Gnotobiotic mouse model to quantify the microbial contribution to BRV pharmacokinetics and toxicity.

(A) Serum and (B) liver BRV and BVU kinetics in GNWT and GNMUT mice. (C) Liver thymine. (D) Intestinal BRV and BVU concentrations over time; each field represents the mean of five animals. (E) Cecal and fecal BRV and BVU concentrations in individual animals. For all mouse data, horizontal lines show mean of five animals, and times reflect hours after oral BRV administration. SI: duodenum; SII: jejunum; and SIII: ileum. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 (t test with FDR correction for multiple hypotheses testing).

A physiologically based pharmacokinetic model of host-microbiome drug metabolism

We next used these quantitative drug and metabolite measurements, collected in various compartments over time and in the presence and absence of microbial drug metabolism, to build a pharmacokinetic model (10) that quantifies the contribution of host and microbiota to systemic drug and metabolite exposure. To parameterize processes independent from microbial BRV metabolism (gray compartments in Fig. 5A), we used a global optimization procedure to fit measured BRV and BVU kinetics in serum and intestinal compartments of GNMUT mice. These processes include rates for (i) BRV absorption from small and large intestine to blood (kaSIP and kaLIP); (ii) BRV elimination (keP); (iii) host BRV to BVU conversion (kcH); (iv) BVU elimination (keM); and (v) intestinal propagation (transit) (kp1 to kp5) (Fig. 5B). To parameterize processes dependent on microbial BRV metabolism (green compartments in Fig. 5A), including bacterial BRV to BVU conversion (kcB) and BVU absorption rates from cecum and colon (kaLI1M and kaLI2M), we used measured BRV and BVU kinetics in cecum and colon (but not serum) of GNWT mice (Fig. 5C; fig. S6, A and B; and table S13). The resulting model is robust (fig. S6C) and accurately predicts BRV kinetics in the serum of GNWT mice (Pearson’s correlation coefficient, PCC = 0.98). Furthermore, the model predicts host and microbial contributions to serum BVU; the sum of these predicted contributions accurately matches total serum BVU measured in GNWT animals (PCC = 0.76) (Fig. 5C and table S14). Comparison of the AUC for estimates of host and microbial contributions to serum BVU reveals that microbial activity accounts for nearly all of the serum BVU measured at later time points and 71% of total BVU exposure in the serum of GNWT mice (Fig. 5C). To predict the microbial contribution to serum BVU in the context of a complex microbiota, we next parameterized the model using drug and metabolite kinetics from the gut and serum of GF mice (fig. S7A), and the gut of CV mice, to predict BVU serum exposure (Fig. 5D and fig. S7, B and C). Despite increased microbial complexity, the model accurately predicts serum BRV kinetics in these animals (PCC = 0.99); the sum of predicted host and microbial contributions to serum BVU matches the measured total serum BVU in CV mice (PCC = 0.78) (Fig. 5D and table S14).

Fig. 5 Physiologically based model of host and microbial contribution to BRV and BVU pharmacokinetics.

(A) Schematic representation of compartments and subprocesses included in the model. (B) Parameterization of microbiota-independent processes using measurements from GNMUT mice. (C and D) Parameterization of microbiota-dependent intestinal drug metabolism and prediction of microbial and host contributions to serum BVU in (C) GNWT and (D) CV mice.

Global sensitivity analysis, which estimates the effect of varying each of the 13 rates included in the model on serum BVU exposure, reveals that the parameters that most affect host and microbial contributions to serum BVU are distinct and that overall serum exposure is dependent on both host and bacterial drug-metabolizing activity (fig. S8). For example, simulating interpersonal differences in gut community composition or antibiotic exposure by changing bacterial drug conversion rate (kcB) reveals that the predicted microbiome contribution to serum BVU can vary from 0 to 71%, and that total systemic BVU exposure can vary more than threefold, in response to this parameter (Fig. 6A). Multiple parameters can also be altered simultaneously to predict the pharmacokinetics of other drugs that are subject to different bioavailability, host- and microbiome-mediated drug metabolism, and drug and metabolite absorption. For example, simultaneous alteration of parameters for both host- and microbiome-mediated drug metabolism produces a three-dimensional surface that estimates total serum metabolite exposure and relative microbiome contribution as a function of both parameters; the model further reveals how bioavailability affects these estimates (Fig. 6B, table S15, and movies S1 and S2).

Fig. 6 Simulation of the impact of chemical, microbial, and physiological parameters on pharmacokinetics and expansion of the approach to other drugs.

(A) Predicting the impact of microbial drug metabolism rate on microbial contribution to serum BVU. (B) Absolute metabolite exposure and relative bacterial contribution to serum BVU as a function of host and microbial drug metabolism rate at a given bioavailability (tables S24 and S25). (C) Prediction of host and microbial contribution to serum BVU after oral sorivudine (SRV) administration to CV mice. (D) Prediction of microbial and host contributions to serum clonazepam (CLZ; P) and aminoclonazepam (NH2-CLZ; M) in CV mice. (E) Schematic representation of an extended model that includes enterohepatic circulation and three drug metabolites (M1 to M3). (F) Prediction of microbial contribution to serum exposure of CLZ (P) and NH2-CLZ (M2) and (G) OH-CLZ (M1) and NH2OH-CLZ (M3) in CV mice. Horizontal lines show mean of five animals and times reflect hours after oral drug administration. For detailed description of parameters, see tables S13 and S21.

Generalization of the approach

We used two additional examples to test whether our approach can model the pharmacokinetics of other microbiome-metabolized drugs. In one example, we focused on sorivudine (SRV), which is structurally similar to BRV but is metabolized to BVU at different rates by both the host and the microbiome (fig. S9). We orally administered SRV to CV and GF mice, measured drug and metabolite levels across tissues and over time as described above, and provided serum drug and intestinal drug and metabolite measurements as inputs to the model for parametrization and prediction (fig. S10). Predicted serum metabolite kinetics for BVU derived from SRV are an order of magnitude lower than for BVU derived from BRV, which matches experimental measurements of serum BVU levels in SRV-treated mice (PCC = 0.86). The model also reveals the relative contribution of host and microbial SRV–metabolizing activity to this exposure (Fig. 6C, fig. S11, and table S14). These results demonstrate that the pharmacokinetic model can predict both levels and sources of metabolite exposure for a drug subject to different host and microbiome drug-metabolizing rates than BRV.

As a further example, we studied clonazepam (CLZ), an oral anticonvulsant and anti-anxiety drug that undergoes a complex metabolic pattern of oxidation, nitroreduction, glucuronidation, and enterohepatic cycling (fig. S12A) (11). Intestinal microbes have been shown to contribute to the reductive metabolism of CLZ and related drugs, which is associated with toxicity (12). Quantification of CLZ and CLZ metabolite serum kinetics after oral administration to GF and CV mice demonstrated that aminoclonazepam (NH2-CLZ) and aminohydroxyclonazepam (NH2OH-CLZ) are the major systemic metabolites in CV animals (figs. S13 and S14 and tables S16 to S20). First, we used gut and serum CLZ and NH2-CLZ measurements from GF mice, and gut measurements from CV mice, to parameterize the host-microbiome pharmacokinetic model, which predicted a substantial microbial contribution to serum NH2-CLZ (Fig. 6D and fig. S15). Second, we expanded the model topology to allow enterohepatic circulation of CLZ and reparameterized the model, which predicted that the microbiome contributes 78% to systemic NH2-CLZ (PCC: 0.85 versus 0.56 without enterohepatic circulation) (Fig. 6, E and F, and fig. S16). Third, we focused on NH2OH-CLZ, which is the product of both hydroxylation (solely performed by host enzymes; fig. S12) and nitroreduction (performed by both microbial and host enzymes (11)). Hydroxyclonazepam (OH-CLZ) and glucuronidated OH-CLZ are also subject to biliary excretion into the intestinal tract, where microbes can further modify (deglucuronidate and reduce) them. Indeed, quantification of these two metabolites and NH2OH-CLZ in bile and intestinal compartments over time demonstrates biliary excretion and microbial deglucuronidation and reduction of OH-CLZ to NH2OH-CLZ in the distal gut (figs. S13 and S14). To quantify the microbial contribution to systemic NH2OH-CLZ, we further expanded the model to include these additional metabolites, used GF and CV metabolite kinetics for parametrization as described, and predicted a microbial contribution of 66% (PCC: 0.93) to serum NH2OH-CLZ (Fig. 6, E and G; figs. S16 to S18; and tables S21 and S22). The sum of predicted host and microbiome contributions match observed total metabolite levels in serum and urine (Fig. 6, F and G, and figs. S16 to S18). Although comparisons between GF and CV animals cannot account for physiological effects of bacterial colonization, as is possible in comparisons between gnotobiotic mice that vary in a single microbiome-encoded enzyme (e.g., Fig. 4 and fig. 5, B and C), these results illustrate the applicability of host-microbiome pharmacokinetic models to disentangling microbial and host contributions to the metabolism of drugs that undergo complex and multistep in vivo disposition.

Together, the results of this study provide an experimental and computational strategy to disentangle host and microbial contributions to drug metabolism, even in cases when host and microbial activities are chemically indistinguishable. Quantitative understanding of these host and microbiome-encoded metabolic activities will further clarify how nutritional, environmental, genetic, and galenic factors affect drug metabolism and could enable tailored intervention strategies to improve drug responses. This approach could be adapted for drugs converted to chemically distinct metabolites by the host and microbiome, and to other xenobiotics, food components, and endogenous metabolites.

Methods summary

Detailed materials and methods are provided in the supplementary materials.

Chemicals

Brivudine, sorivudine, and 5,6-dihydrouracil were purchased from Santa Cruz Biotechnology, liquid chromatography–mass spectrometry (LC-MS)–grade solvents from Fisher Scientific, and all other chemicals from Sigma Aldrich, if not specified otherwise.

Bacterial culture conditions

General culture conditions: Escherichia coli S-17 λ pir strains (13) were grown at 37°C in LB medium supplemented with carbenicillin (50 μg/ml). B. thetaiotaomicron VPI-5482 [American Type Culture Collection (ATCC) 29148]–derived strains were grown anaerobically at 37°C in liquid TYG medium (14). All anaerobic culturing was performed on brain-heart-infusion (BHI; Becton Dickinson) agar supplemented with 10% horse blood (Quad Five Co.). Cultures of bacterial gut communities and isolates for drug degradation assays were grown in gut microbiota medium (GMM) (15). For selection, gentamicin (200 μg/ml), erythromycin (25 μg/ml), and/or 5-fluoro-2-deoxy-uridine (FUdR) (200 μg/ml) were added as indicated. A flexible anaerobic chamber (Coy Laboratory Products) containing 20% CO2, 10% H2, and 70% N2 was used for all anaerobic microbiology steps. Growth curves were performed as described in the supplementary materials and methods.

Construction of B. thetaiotaomicron targeted mutants

Strains and plasmids are listed in table S8 and primers are listed in table S23. Gene deletion and complementation: B. thetaiotaomicron tdk is indistinguishable from its parent strain with respect to BRV to BVU conversion (fig. S3A). A counterselectable allelic exchange procedure (16) was used to generate in-frame, unmarked deletions in a B. thetaiotaomicron VPI-5482 tdk background (WT). Experimental details of gene deletion and complementation are provided in the supplementary materials and methods.

Construction of condensed transposon mutant library

B. thetaiotaomicron mariner transposon insertion strains were selected from a previously reported library of 7155 B. thetaiotaomicron mutants, which had been clonally arrayed and mapped by insertion sequencing (INSeq) (8). Strain selection rationale and assay conditions are described in the supplementary materials and methods.

Enzyme assays

Liver assays of conversion of BRV and SRV to BVU: Human and murine S9 liver fractions were purchased from Thermo Fisher Scientific (HMS9L and MSMCPL, respectively). Enzyme assays were performed as previously described for the deglycosylation of arabinosyluracil derivatives (17) as described in the supplementary materials and methods.

Cloning, purification, and enzymatic assay of BT4554 : BT4554 was purified as an epitope-tagged protein fusion, and enzyme assays were performed using the conditions described for liver across a range of BRV concentrations as described in the supplementary materials and methods.

Bacterial BRV conversion assays

Bacterial community and axenic culture assays: All handling of human materials was conducted with the permission of the Yale Human Investigation Committee. Samples were collected and stored as previously described (15). Assay conditions for bacterial communities, individual species, and transposon mutants are described in the supplementary materials and methods.

Animal experiments

All experiments with mice were performed using protocols approved by the Yale University Institutional Animal Care and Use Committee. Methods for conventional and gnotobiotic husbandry, colonization of gnotobiotic animals, drug administration, serum and tissue collection, and bioavailability studies are provided in the supplementary materials and methods.

Sample preparations for drug and metabolite analysis

Sample extraction: Sample preparation was performed as described previously (18). Experimental procedures for extraction of liquid and solid samples are provided in the supplementary materials and methods.

LC-MS quantification of drugs and metabolites

LC-MS analysis: Chromatographic separation was performed on a C18 column, and the qTOF (Agilent 6550) was operated in positive scanning mode as described in the supplementary materials and methods. Compounds were identified based on the retention time of chemical standards and their accurate mass (tolerance 20 ppm).

Data analysis: The MassHunter Quantitative Analysis Software (Agilent, version 7.0) was used for peak integration. Statistical analysis and plotting were performed in Matlab 2017b (MathWorks). LOQ determination, pharmacokinetic parameter estimation, and additional methods are provided in the supplementary materials and methods.

Pharmacokinetic multicompartment modeling

Model overview: The multicompartment pharmacokinetic model of drug metabolism in the mouse contained seven main compartments (small intestine I to III, cecum, colon, feces, and serum; Fig. 5A). Two additional compartments (small_intestine_gi and small_intestine_serum) were used as reservoirs for the initial drug dose. The serum compartment incorporated processes occurring in the liver, kidneys, and all other body parts apart from the gastrointestinal (GI) tract. Exposure to the drug was modeled as an input to the small_intestine_serum compartment of the initial amount of drug equal to D × F, where D is the provided dose and F is the bioavailability coefficient, and input to the small_intestine_gi compartment of the initial amount of D × (1 − F). Drug propagation through the body was driven by the flow of GI material in different GI tract sections and tissue:serum diffusion coefficients. Model parameters and equations are provided in table S13. All equations were defined for drug and metabolite amounts. For the parameter fitting, metabolite concentrations were converted into amounts using estimated compartment volumes provided in table S13. For drug metabolite levels in serum, the metabolite levels contributed by the host (due to host drug metabolism, MH) were distinguished from the metabolite levels contributed by the microbiota (due to microbial metabolism in the cecum and metabolite absorption, MBAC). The model was created by using the MatLab 2017b SimBiology Toolbox (MathWorks).

Extended model overview: The multicompartment pharmacokinetic model of drug metabolism described above was extended to incorporate enterohepatic circulation and two additional drug metabolism products. Enterohepatic circulation was modeled by adding a bile compartment, enterohepatic cycling coefficient keh (which determines the rate of compound diffusion from serum to bile and from bile to small intestine I), and compound-specific absorption coefficients from the small intestine (table S21).

Detailed methods for parameter fitting, serum metabolite exposure prediction, and sensitivity analysis are provided in the supplementary materials and methods.

Model simulation: To investigate the influence of bioavailability (F), host drug to metabolite conversion coefficient, and microbial drug to metabolite conversion coefficient on the total BVU serum exposure and relative microbial contribution to serum BVU, the sbiosimulate function was used to determine BRV model behavior across all combinations of the parameter values ranging from 0.01 to 0.99 for F with step 0.01, and 0.001 to 1000 in logarithmic scale with step 1 (power of 10) for the conversion coefficients. All other parameters were set to values estimated with GNMUT and GNWT data. For each model run, the AUC of BVU serum concentrations was calculated. The bacterial contribution was calculated as the ratio between microbial BVU absorbed from cecum to serum, and total BVU in the serum (movies S1 and S2 and table S15).

Supplementary Materials

www.sciencemag.org/content/363/6427/eaat9931/suppl/DC1

Materials and Methods

Figs. S1 to S18

Tables S1 to S26

Movies S1 to S2

References (2027)

References and Notes

Acknowledgments: We thank the Goodman lab for helpful discussions and N. A. Barry, L. Valle, and D. Lazo for technical assistance. Funding: This work was supported by NIH grants GM118159, GM105456, AI124275, the Center for Microbiome Informatics and Therapeutics, the Burroughs Wellcome Fund, the Yale Cancer Center, and the HHMI Faculty and Pew Scholars Programs to A.L.G. M.Z. received an Early and Advanced Postdoc Mobility Fellowships from the Swiss National Science Foundation (P2EZP3_162256 and P300PA_177915, respectively) and a Long-Term Fellowship (ALTF 670-2016) from the European Molecular Biology Organization. M.Z.-K. received an Early Postdoc Mobility Fellowship from the Swiss National Science Foundation (P2EZP3_178482). Author contributions: M.Z. and A.L.G. conceived and initiated the project; M.Z. performed the experiments; R.W. performed and analyzed the loss-of-function screen; M.Z. and M.Z.-K. analyzed the data. M.Z.-K. performed statistical analyses, developed in silico models, and prepared graphical illustrations; M.Z., M.Z.-K., and A.L.G. wrote the manuscript. Competing interests: M.Z., M.Z.-K., and A.L.G. have filed a patent application based on these studies with the U.S. Patent and Trademark Office (62/693,741). The authors declare no other competing interests. Data and materials availability: All data are available in the manuscript or the supplementary materials. The MatLab modeling framework is available at https://github.com/mszimmermann/PBPK_host-microbiome_model under GPLv3 license and archived at Zenodo (19).
View Abstract

Navigate This Article