Integrating the gut microbiome and pharmacology

Abstract

The gut microbiome harbors trillions of organisms that contribute to human health and disease. These bacteria can also affect the properties of medical drugs used to treat these diseases, and drugs, in turn, can reshape the microbiome. Research addressing interdependent microbiome-host-drug interactions thus has broad impact. In this Review, we discuss these interactions from the perspective of drug bioavailability, absorption, metabolism, excretion, toxicity, and drug-mediated microbiome modulation. We survey approaches that aim to uncover the mechanisms underlying these effects and opportunities to translate this knowledge into new strategies to improve the development, administration, and monitoring of medical drugs.

INTRODUCTION

In a recent survey in the United States, 45.7% of respondents report using at least one prescription drug within the past 30 days (1). Although the gut microbiome can metabolize many commonly administered drugs (2, 3), the consequences of these and other microbiome-drug interactions remain unclear. In this Review, we survey how the microbiome affects key drug properties, including bioavailability, absorption, metabolism, excretion, and toxicity (Fig. 1) (4–6). In addition, we discuss experimental and computational methods for exploring microbiome-host-drug interactions, suggest approaches to studying these interactions in humans, and identify challenges ahead for translating this research into clinical treatments.

Fig. 1. Direct and indirect impacts of the gut microbiome on drug pharmacokinetics.

Bacteria may affect drug absorption through bioaccumulation (uptake of the parent drug), bile salt metabolism (changes to emulsification/solubilization of lipophilic drugs), or altered drug uptake through passive (paracellular) diffusion or protein carrier–mediated transport (influx and efflux) (25). Bacteria can also metabolize drugs directly, produce metabolites that alter expression and activity of host DMEs, and modify bile acids that act on host nuclear receptors (e.g., PXR). Bacteria affect drug excretion by removing glucuronides or sulfates from phase 2 host drug metabolites, possibly promoting their entry into the enterohepatic circulation and further metabolism. Bacteria may also convert host drug metabolites back into their active and toxic forms (e.g., via deglucuronidation or sulfatase activity), transform drugs into metabolites with local or systemic toxicity (toxification), or negatively alter gut microbiome composition (dysbiosis).

MICROBIAL EFFECTS ON DRUG ABSORPTION AND BIOAVAILABILITY

Enteral (oral) administration, or per os, is a common route of drug delivery, and drug absorption is a key factor in determining the pharmacokinetics of oral drugs. Absorption can be defined as “the movement of drug across the outer mucosal membranes of the (gastrointestinal) tract” (6). Although difficult to quantify, absorption can be estimated using mathematical predictions and in situ or ex vivo experimental methods (7–9). The major site of absorption is the small intestine (8), where estimates of bacterial densities in humans range from 10¹ to 10³ cells/g of intestinal contents in the duodenum to more than 10⁸ cells/g in the ileum (10–12). Absorption may also occur in distal locations of the gastrointestinal tract: For extended-release formulations and poorly absorbed drugs, absorption in the large intestine may contribute to systemic drug concentrations (13). Drugs that reach the large intestine have increased likelihood of interacting with a higher density and diversity of bacteria (11). In mice, bacterial densities are estimated to reach 10¹¹ colony-forming units/g in the cecum and exceed this amount in the colon (10, 11, 14); bacterial diversity also increases in the cecum and colon (15).

Gut microbes may affect drug absorption through multiple mechanisms, including alteration of drug concentrations in the gut lumen (by sequestering or metabolizing drugs), modulation of passive diffusion, alteration of protein carrier-mediated active drug transport, and changes to drug efflux (Fig. 1). Rodent studies comparing germ-free mice (mice lacking all microorganisms), gnotobiotic mice (with a known gut microbiota), and conventional mice (mice carrying an endogenous mouse microbiota from birth) demonstrate that the bacterial influence on these processes alters efficacy and pharmacokinetics of multiple drugs, including levodopa, amiodarone, capecitabine, acetaminophen, brivudine, and nonoral (e.g., intravenous) drugs that undergo enterohepatic circulation (16–24).

Bacteria have the potential to affect drug concentrations in the gut lumen, and alter drug absorption rates in turn, through bioaccumulation (sequestration of drugs) and metabolism (25). Microbial communities in the small intestine may accumulate drugs such as acetaminophen, as observed in rats (26), possibly through the activity of transporters homologous to human enterocyte drug transporters. A systematic investigation identified 70 cases of drug bioaccumulation in microbes in vitro (27). In one experiment, duloxetine-bioaccumulating or non-bioaccumulating Escherichia coli strains were incubated with duloxetine, and culture supernatants were administered to Caenorhabditis elegans (27). These supernatants had distinct impacts on the treated animals, suggesting that bioaccumulation reduced drug concentrations in the culture supernatants (27). Further studies are needed to understand how bacterial bioaccumulation may affect drug absorption in mammals. Microbial drug metabolism may also reduce lumenal drug concentrations and subsequent absorption rates; levodopa provides one example in mice (16).

Many orally administered drugs are absorbed by transcellular or paracellular passive diffusion (28), the latter favoring molecules <200 kDa (29). Bacterial isolates (e.g., E. coli Nissle 1917) and gut microbial metabolites promote gut barrier integrity (30–32), suggesting that these microbes and their products may affect drug absorption through an impact on paracellular passive diffusion. For example, the pregnane X receptor (PXR) coordinates multiple processes that enhance gut barrier integrity; thus, PXR-deficient mice exhibit increased paracellular transport (31). Clostridium sporogenes produces the PXR agonist indole 3-propionic acid, which promotes gut barrier function through interaction with host PXR in the intestinal epithelium (31). Such interactions may alter intestinal permeability: For example, the nonsteroidal anti-inflammatory drug indomethacin causes increased gut permeability in mice, as measured by absorption of the indicator molecule fluorescein isothiocyanate–dextran; C. sporogenes indole 3-propionic acid reduces indomethacin-dependent absorption of fluorescein isothiocyanate–dextran (31). Nuclear receptors implicated in regulating gut wall integrity and modulating passive diffusion, including farnesoid X receptor (FXR), constitutive androstane receptor, and aryl hydrocarbon receptor, are also activated by microbial metabolites (32). Further studies of bacterial influence on passive transport may identify drugs affected by alterations to gut permeability.

Gut microbes may also affect passive diffusion by altering the bile salt pool, thereby affecting drug solubility and nuclear receptor activation. Primary bile salts, the ionized form of primary bile acids associated with Na⁺ or K⁺ ions, emulsify dietary lipids, vitamins, and lipophilic xenobiotic compounds, including medical drugs (33). Gut bacteria encode multiple classes of bile salt hydrolases and other enzymes that alter the variety and abundance of bile acids (34). The primary bile acid conjugate taurocholate and the secondary bile acid conjugate taurodeoxycholate (a product of microbial metabolism) have distinct effects on the solubility of multiple drugs (35). Furthermore, host-derived primary and bacteria-derived secondary bile acids vary in their ability to activate or antagonize host nuclear receptors such as FXR, further linking microbial modulation of the bile acid pool with possible alterations to intestinal permeability (35–37). Although these connections highlight potential microbial effects on drug absorption by the host, further research in this area is necessary.

Some drugs are absorbed through carrier-mediated active transport (facilitated diffusion) rather than passive diffusion (28), including cefadroxil (transported through peptide transporter 1) (38), methotrexate [through human proton-coupled folate transporter (hPCFT)] (39), and fexofenadine [through organic ion transporting peptide 2B1 (OATP2B1)] (40) (see Table 1 for examples of transporters affected by bacteria). Studies in mice suggest that gut microbes can affect expression of many of these transporters, including the mouse homolog of peptide transporter 1 (41). Bacteria may also affect absorption indirectly: In one example, azo dyes have been shown to inhibit the transporter OATP2B1 and thus absorption of OATP2B1-dependent drugs, including fexofenadine (42). Gut microbes metabolize azo dyes into metabolites that do not affect OATP2B1-dependent transport (42). Indeed, conventional mice administered a low dose (2.5 mg/kg) of the azo dye FD&C Red 40 displayed similar plasma concentrations of fexofenadine compared to untreated controls that did not receive the dye, most likely because of complete metabolism of the dye by gut microbes before it was able to inhibit transport of fexofenadine (42). However, conventional mice administered a higher dose (25 mg/kg) of the dye exhibited reduced plasma concentrations of fexofenadine, potentially because of incomplete microbial metabolism of this higher dose of azo dye (42). How these indirect effects on absorption affect drug pharmacokinetics in humans is unknown.

Table 1:

Enzymes and Transporters That Affect Drug Metabolism

Source	Class	Name	Microbial-drug interaction	References
Bacteria	enzyme	cgr2	Produced by E. lenta; reduces digoxin to inactive form	(73, 75)
Bacteria	enzyme	TyrDC	Produced by E. faecalis; decarboxylates levodopa to dopamine, decreasing bioavailability	(16, 77)
Bacteria	enzyme	BT4554	Produced by B. thetaiotaomicron; transforms brivudine to toxic metabolite bromovinyluracil	(18)
Human	enzyme	CYP3A4	Homologous to murine Cyp3a11; down-regulated in germ-free and antibiotic-treated mice	(60)
Human	enzyme	CYP2B6	Homologous to murine Cyp2b10; down-regulated in germ-free and antibiotic-treated mice	(61)
Human	enzyme	SULT1A	Bacterial p-cresol competes with acetaminophen for active site, altering bioavailability and toxicity	(20, 21, 70)
Human	transporter	OATP2B1	Bacterial metabolism of azo dyes influences inhibition of OATP2B1 transport of fexofenadine	(42)
Human	transporter	ABCG2/BCRP	Bacterial urolithins inhibit ABCG2/BCRP-dependent efflux of mitoxantrone in cell culture	(45)
Mouse	enzyme	Cyp3a11	Down-regulated in germ-free and antibiotic-treated mice	(41, 49, 51, 52, 54, 56)
Mouse	enzyme	Cyp2b10		(50, 54, 62)
Mouse	enzyme	Ugt1a1	Up-regulated in germ-free and antibiotic-treated mice	(41)
Mouse	enzyme	Ugt1a3	Down-regulated in antibiotic-treated rats	(64)
Mouse	enzyme	Gsta1	Up-regulated by Lactobacillus rhamnosus GG and its metabolite 5-methoxyindoleacetic acid	(55)
Mouse	enzyme	Nqo1
Mouse	enzyme	Trx1

Drugs can be retained within intestinal epithelial cells for seconds to hours (6), with a portion then excreted by host efflux pumps back into the intestinal lumen (43). Transporters responsible for gut epithelial efflux include P-glycoprotein, multidrug resistance protein 2, and breast cancer resistance protein (ABCG2/BCRP) (44). Gut microbes may also alter gut epithelial cell efflux. Bacterial urolithins act as substrates for the ABCG2/BCRP efflux pump in intestinal epithelial cells; these molecules and their metabolites inhibit the efflux of mitoxantrone, an antineoplastic drug and a substrate of ABCG2/BCRP, in a dose-dependent manner in cell culture (45).

Absorption is one of multiple factors that determine drug bioavailability. Bioavailability is “the extent and the rate at which a substance or its active moiety is delivered from a pharmaceutical form and becomes available in the general circulation” (5–6). Bioavailability of an oral drug is influenced by absorption, first-pass metabolism (which is any metabolism occurring before systemic circulation of a xenobiotic, including metabolic activities of the gut epithelium), and excretion. Bioavailability of a drug is determined by comparing serum exposure (the amount of drug found circulating in the serum of the host) between a drug administered extravascularly (e.g., orally) and the same drug administered intravenously. Well-absorbed drugs can still exhibit limited bioavailability because of first-pass metabolism and other factors.

MICROBIAL EFFECTS ON DRUG METABOLISM

Host metabolism

The chemical structure of a drug can be modified by an array of host-encoded drug-metabolizing enzymes (DMEs). Phase 1 enzymes, including cytochrome P450 reductases (CYPs), carboxylesterases, and epoxide hydrolases (46), often expose or introduce polar functional groups. Phase 2 enzymes commonly introduce larger, polar moieties onto phase 1 metabolites, forming conjugated drug molecules with higher water solubility that allows for more efficient excretion. Enzyme families associated with phase 2 transformations include uridine 5′-diphosphate–glucuronyl transferases (UGTs), sulfotransferases (SULTs), glutathione S-transferases, and catechol-O-methyltransferases (46, 47). Phase 1 and phase 2 enzymes can act on a drug before systemic circulation because of activity in the small intestinal epithelium (first-pass or presystemic metabolism) or after transport to the liver via the portal vein. These enzymes also affect the concentrations of drugs and their metabolites upon systemic circulation, generally because of their activity in the liver (46). Host regulation of these DMEs occurs mainly through PXR and constitutive androstane receptor, with additional contributions from FXR and aryl hydrocarbon receptor (48). Hence, many DMEs are induced by drugs and xenobiotic compounds via these nuclear receptors (48), suggesting that small molecules from the gut microbiota could also affect enzyme expression (Fig. 1). Multiple studies comparing germ-free or antibiotic-treated mice to conventional animals highlight the broad impact of the gut microbiome on host DME expression and drug metabolism (41, 49–55). Whereas the specific microbial factors responsible for these effects are in most cases unknown, there are several examples of specific bacterial molecules that modulate host DME expression, including 5-methoxyindoleacetic acid (55), lipopolysaccharide/lipoteichoic acid (56), and bacteria-modified bile acids (57–59).

Expression of multiple CYPs and other phase 1 enzymes is affected by the gut microbiome (Table 1) (41, 49, 50). For example, human CYP3A4 homolog Cyp3a11 shows reduced mRNA expression and reduced Cyp3a11 protein in germ-free and antibiotic-treated mice compared with conventional animals (41, 49, 54). Midazolam (a postoperative sedative and substrate of CYP3A enzymes) shows reduced bioavailability when administered to conventional mice compared with their germ-free counterparts (51). Increased hydroxylase activity in conventional mice may also contribute to this effect. CYP3A4 acts on ~50% of clinical drugs (60), broadly implicating the gut microbiome in modulating host drug metabolism by influencing the activity of important host DMEs.

Another phase 1 human DME, CYP2B6, is implicated in the metabolism of 10 to 12% of clinical drugs, including bupropion and cyclophosphamide (61). Germ-free and antibiotic-treated mice exhibit lower mRNA expression and protein of the murine CYP2B6 homolog Cyp2b10 in liver tissue (50, 54, 62). Microsomal liver fractions from Germ-free mice exhibit lower activity against selective CYP2B substrates benzyloxyresorufin and pentoxyresorufin compared with liver microsomes from conventional animals (54). Natural variation in gut microbiome composition is also linked to changes in CYP enzyme activity in mouse liver tissue (51, 52). For example, variation in Cyp3a activity in conventional mice has been connected to variation in their gut microbiomes (52), possibly due in part to bacterial cell wall components such as lipopolysaccharide or lipoteichoic acid (56). Direct administration of these bacterial compounds to conventional mice reduces Cyp3a11 mRNA expression, reduces Cyp3a11 protein, and increases midazolam bioavailability (56). However, it is important to note that nuclear receptor homologs such as PXR are activated differently in mice and humans, with consequences for downstream CYP3A/Cyp3a expression regulated by PXR (63). For this reason, caution should be used when extrapolating data from murine studies to humans.

Multiple phase 2 enzymes are also affected by the gut microbiome. For example, Ugt1a1 is expressed at higher levels in germ-free than conventional mice, specifically in the large intestine (41). By contrast, rats treated with antibiotics exhibit reduced duodenal expression of Ugt1a3, a gene encoding a host UDP-glucuronyltransferase targeting the antipsychotic olanzapine. Consistent with this observation, antibiotic-treated rats display altered olanzapine pharmacokinetics (64). Expression of other UGT phase 2 enzymes is also dependent on microbiome colonization of the gut (41, 49, 64). Introduction of Lactobacillus rhamnosis GG or its metabolite 5-methoxyindoleacetic acid induces expression of nuclear factor erythroid 2–related factor 2 (Nrf-2), which up-regulates multiple phase 2 enzymes (65), including glutathione transferase Gsta1, reduced form of nicotinamide adenine dinucleotide (phosphate) hydrogen:quinone oxidoreductase Nqo1, and thioredoxin reductase Trx1 (55). This Nrf-2 pathway acts to mitigate the damaging effects of oxidative stress (65), and administration of L. rhamonsis GG or 5-methoxyindoleacetic acid from L. rhamnosis GG is protective during acetaminophen or ethanol overdose in mice (55).

Primary and secondary bile acids can also act as agonists/antagonists of nuclear receptors that regulate DME gene expression (66). Thus, changes in concentrations or structures of primary and secondary bile acids through microbial deconjugation (via bile salt hydrolases), dehydroxylation, dehydrogenation, and epimerization (57) can subsequently affect expression of phase 1 and phase 2 enzymes. The secondary bile acids lithocholic acid (58) and ursodeoxycholic acid (67) affect PXR activity, with consequences for absorption and CYP3A activity. Consistent with this potential link, germ-free mice colonized with wild-type Bacteroides thetaiotaomicron exhibit large differences in DME expression compared with germ-free mice colonized with an isogenic bile salt hydrolase mutant strain of B. thetaiotaomicron, potentially due to changes in PXR activity (59). PXR also mediates drug-drug interactions; for example, coadministration of rifampicin and isoniazid results in liver injury because of PXR-related metabolic changes (68). FXR also regulates expression of CYP3A4 in hepatocytes (69), suggesting that microbial modulation of FXR activation may also affect CYP3A4 expression. Intriguingly, bacteria-drug interactions with host DMEs via PXR, FXR, or other nuclear receptors may then alter the pharmacokinetic/pharmacodynamic (PK/PD) properties of drugs.

Beyond altering expression of host DMEs, gut microbial metabolites also interfere with the activity of these enzymes (Fig. 1). Acetaminophen and the bacterial metabolite p-cresol are substrates of host SULT1A (70). Individuals with high p-cresol levels exhibit reduced acetaminophen sulfation through substrate competition for SULT1A, thus leading to hepatotoxicity due to increased liver exposure to acetaminophen (70). Interestingly, antibiotic treatment of rodents may lead to increases (21) or decreases (20) in acetaminophen bioavailability compared with untreated controls, possibly due to differences in gut microbiome composition between studies.

Bacterial metabolism

In addition to altering host DME expression and activity, gut microbes metabolize drugs directly (Fig. 1). The gut microbiome encodes at least 100-fold more genes than the human genome, including a large repertoire of enzymes with the capacity to metabolize drugs (2, 3). These drugs are likely not primary targets for bacterial DMEs (71), and the consequences of host- and microbiome-mediated drug metabolism are often distinct. For example, host DMEs typically increase excretion of drugs through hydrophilic conjugations (46), whereas bacterial DMEs can reverse these modifications (72) and metabolize drugs further (2, 3, 73), possibly using these compounds as carbon sources or substrates for anaerobic respiration (71). Bacterial drug-metabolizing activities can activate prodrugs (74), decrease serum drug exposure (3, 16, 75–78), or increase exposure to microbial drug metabolites (3, 18) or may have no effect on drug pharmacokinetics depending on the sites of absorption, microbial activity, and other factors. Prodrugs are delivered as an inactive compound and activated in the body. The ulcerative colitis prodrug sulfasalazine is enzymatically cleaved by gut microbes to form sulfapyridine and 5-amino salicylic acid (74). Likewise, the prodrug antibiotic prontosil is enzymatically cleaved by gut microbes to its active form, sulfanilamide (79). Consequently, activation of both sulfasalazine and prontosil by the gut microbiota contributes to their effects in vivo.

Bacterial metabolism can decrease drug concentrations, lowering systemic drug exposure. For example, strains of Eggerthella lenta that encode the cgr operon reduce the cardiac glycoside digoxin to an inactive form (75, 78), which decreases systemic digoxin exposure in mice (75). This effect is evident in comparisons between germ-free mice colonized with cgr-positive compared with cgr-negative E. lenta strains (75), although other genetic differences between these strains should also be considered. Arginine inhibits expression of the cgr operon and inhibits cardiac glycoside reductase 2 (Cgr2) activity, linking this microbial enzyme to dietary protein consumption (75, 78). Treatment with dietary protein containing arginine partially reverses the impact of E. lenta on digoxin exposure (75).

Tacrolimus, a common immunosuppressant drug used to combat organ transplant rejection with a narrow therapeutic range, provides a second example. Tacrolimus is converted into a less active metabolite by multiple Clostridiales species (76). Abundance of one of these metabolizing species, Faecalibacterium praustnitzii (76), positively correlates with a higher dose requirement in certain patients (80). Furthermore, kidney transplant patients treated with this drug exhibit interindividual variation in the amount of inactivated metabolite in their stool, which is recapitulated in ex vivo incubation of patient fecal bacterial communities with the drug (76). It is thus possible that the gut microbiome contributes to the low and variable bioavailability of tacrolimus in patient populations (81).

In a third example, the dopamine precursor levodopa, a mainstay of treatment for Parkinson’s disease, is converted to dopamine and then m-tyramine in the gut via a multistrain bacterial enzymatic pathway (Fig. 2) (16, 77). The responsible enzyme for bacterial decarboxylation of levodopa to dopamine is tyrosine decarboxylase (TyrDC), which is widely found in a common gut bacterial species, Enterococcus faecalis. Rats colonized with E. faecalis tyrDC⁺ display lower systemic levodopa concentrations when compared with rats colonized with an isogenic E. faecalis tyrDC mutant strain (77). In addition, tyrDC gene abundance in small intestinal microbiomes of these rats negatively correlates with circulating levodopa concentrations (77). In another study, gnotobiotic mice colonized with E. faecalis exhibit reduced levodopa bioavailability in comparison with E. faecalis–monoassociated mice treated with a chemical inhibitor of bacterial TyrDC (16). Last, E. faecalis and tyrDC gene abundance positively correlate with reduced response to levodopa, measured in human patients using a standardized levodopa challenge test and genomic analysis of patient fecal samples (82). Although the upper gastrointestinal tract readily absorbs levodopa (83), these studies suggest that certain microbes reach sufficient abundance in this location to reduce gut lumen drug concentration before absorption, possibly affecting bioavailability of the drug. These studies add to the growing evidence (84–86) that targeting bacterial DMEs may be a valid therapeutic route to improve patient drug responses (Fig. 3). A wider examination of microbiome-mediated xenobiotic and drug metabolism can be found in several recent publications (2, 3, 71).

Fig. 2. Effects of blocking bacterial drug metabolizing enzymes.

Inhibition of the metabolism of the drug levodopa to dopamine by the bacterial enzyme TyrDC increases the bioavailability of levodopa. (Left) Bacteria in the small intestine can decarboxylate levodopa to dopamine through the action of TyrDC. Because levodopa, but not dopamine, can cross the blood-brain barrier into the central nervous system, bacterial metabolism of levodopa in the gut reduces its bioavailability such that less levodopa enters the central nervous system. (Right) Microbial decarboxylation of levodopa in the gut of gnotobiotic mice can be blocked using an inhibitor of TyrDC, (S)-α-fluoromethyltyrosine (S-AFMT) (16). This results in increased serum concentrations of levodopa enabling more levodopa to enter the central nervous system, thus improving drug efficacy in the mice (16).

Fig. 3. Host-microbiome-drug interactions in humans and mice.

(A) Parallel studies in humans and in mice with a humanized gut microbiota can provide mechanistic insights into the role of the gut microbiome in drug PK/PD and toxicity. For example, complete or fractionated fecal microbiomes from patients (e.g., drug responders and nonresponders) can be transplanted into wild-type or transgenic Germ-free mice to determine how gut microbiome differences contribute to drug responses. (B) Shown are opportunities to leverage the human gut microbiome to improve development, administration, and monitoring of clinical drugs. Understanding the role of the gut microbiome in drug responses could enable prescreening of individuals and selection of those most likely to respond to a drug successfully and safely. This knowledge also could be used to optimize drug dosing and could inform microbiome interventions such as (1) prebiotics, (2) probiotics, (3) postbiotics, (4) targeted antibiotics or bacterial enzymatic inhibitors, (5) bacteriophages, and (6) fecal microbiota transplants.

Microbes may also metabolize drugs without affecting systemic drug exposure, including in cases where absorption occurs before microbial drug metabolism. For example, omeprazole is readily converted to its sulfide metabolite by rat cecal and colonic intestinal contents in vitro (87), as well as many individual bacterial gut isolates in vitro (3). Despite this robust in vitro activity, this drug is rapidly absorbed in vivo, and systemic drug exposure after oral omeprazole administration is equivalent in antibiotic-treated and untreated rats (87). Additional studies are needed to assess potential indirect effects of antibiotics on omeprazole pharmacokinetics and to determine whether omeprazole pharmacokinetics are influenced by the gut microbiota. In another example, the antiviral drug brivudine is converted to the toxic metabolite bromovinyluracil by gut microbes (18). Brivudine is rapidly absorbed, and serum brivudine concentrations are similar in gnotobiotic mice colonized with either a bacterial strain capable of metabolizing brivudine or an isogenic mutant lacking the responsible enzyme (18). Brivudine concentrations are reduced in the cecum and colon of mice colonized with the wild-type strain, however, and gut and serum concentrations of the toxic metabolite bromovinyluracil are elevated in mice carrying brivudine-metabolizing gut microbes.

MICROBIAL EFFECTS ON DRUG EXCRETION

Many drugs are modified by phase 2 enzymes to facilitate excretion in feces or urine. Glucuronidated drug metabolites are generally inactive and secreted via the biliary route into the small intestine (88). Glucuronidated compounds returned to the gastrointestinal tract encounter microbes that can remove the glucuronyl moiety (Fig. 1) (72), possibly using these liberated glucuronides as a carbon source (53). Substantial evidence from experimental studies and in silico modeling indicates that this activity plays an important role in slowing excretion and increasing systemic exposure to certain drugs and drug metabolites (53, 89, 90). Drugs affected by bacterial deglucuronidation include irinotecan, regorafenib, diclofenac, indomethacin, ketoprofen, and mycophenolate (84, 85, 91–94).

The relationship between microbial drug deglucuronidation and systemic exposure can be complex. Digoxin provides one example: This drug is glucuronidated in the liver and returns to the gut via biliary secretion, where gut microbial deglucuronidation facilitates its return to the systemic circulation. However, microbes such as E. lenta strains carrying cgr2 also inactivate the drug by reduction (addition of two hydrogen atoms to a double bond) (73). In a human case study, oral administration of digoxin coupled with erythromycin (antibiotic) treatment resulted in elevated serum digoxin exposure compared with oral or intravenous digoxin administration without erythromycin treatment (73). This counterintuitive finding may be the result of lowered Cgr2 activity after erythromycin treatment, which increases serum drug exposure despite lowered enterohepatic recycling. In another example, gut microbial deglucuronidation of diclofenac glucuronide results in intestinal enteropathy in mice. Inhibition of the responsible gut microbial β-glucuronidases reduces the severity of this gastrointestinal side effect, whereas systemic exposure to diclofenac remains unaltered (95).

Drug sulfation by host SULTs is another important phase 2 host transformation. Although endobiotic compounds are thought to be the primary substrates of host SULT enzymes (47), many drugs are also subject to host sulfation, including acetaminophen, nonsteroidal anti-inflammatory drugs, and opioids (96). Importantly, the gut microbiome produces sulfatases that can remove these sulfate moieties (Fig. 1) and reactivate drugs after biliary secretion. One study identified 728 bacterial sulfatases in the Human Microbiome Project protein database and 1766 sulfatases in the Integrated Gene Catalogue (97), highlighting the widespread distribution of these enzymes. In addition, several of these enzymes display sulfatase activity toward endobiotic sulfates found in the gut (97). More work in this area, including studies in animal models and in humans, will be required to determine the effects of these interactions on drug excretion.

MICROBIAL EFFECTS ON DRUG TOXICITY

Drug toxicity refers to adverse events occurring from clinically relevant dosing of a compound (98). Many mechanisms cause drug toxicity, including on-target toxicity, hypersensitivity, and off-target effects (98). Glucuronidated drugs reactivated (returned to the parent drug structure and activity) because of bacterial glucuronidase activity can lead to intestinal toxicity, causing diarrhea, inflammation, and ulcers (Fig. 1) (86, 92); specific inhibitors that target the responsible bacterial enzymes can also inhibit toxic reactivation of drugs (84, 85, 91). The antineoplastic drug irinotecan provides an example: The active form, SN-38, is released through bacterial deglucuronidation, causing intestinal damage (86). Inhibition of gut microbiome–related irinotecan reactivation leads to reduced side effects, reduced drug-related gut epithelial damage, and longer periods of drug tolerance in mice (85). These alterations to drug tolerance are reflected in increased antitumor efficacy through improved tumor regression (85). Similarly, experimental inhibitors for bacterial β-glucuronidases targeting regorafenib and nonsteroidal anti-inflammatory drugs such as diclofenac, ketoprofen, and indomethacin aim to reduce drug reactivation and consequent adverse effects (91). These studies provide a second example of therapeutic treatments that target bacterial enzymes to improve drug responses. It is possible that microbiome-encoded sulfatase activity also produces toxic side effects because these enzymes may be responsible for reactivation of sulfate-conjugated drugs in the gut. Further studies are needed to identify drugs subject to these processes and the responsible bacterial sulfatases.

Bacteria may also convert a drug to a toxic metabolite. For example, bacteria contribute to metabolism of the antiviral drug brivudine to bromovinyluracil, which can be hepatotoxic (18). The prodrug mycophenolate mofetil is converted to its active form mycophenolic acid by bacterial species and communities in vitro (2, 3). This activity, along with bacterial deglucuronidation of the phase 2 metabolite mycophenolic acid–glucuronide, together may contribute to local or systemic adverse drug events and systemic mycophenolic acid exposure (92, 94). Vancomycin treatment in mice treated with mycophenolate mofetil decreases bacterial deglucuronidation and reactivation of mycophenolic acid in the gut, preventing downstream side effects such as colonic inflammation (92), possibly providing a crude but useful therapeutic tool to prevent toxic bacterial reactivation of drugs in the gut.

Last, enterohepatic circulation may also increase systemic exposure to drug metabolites produced by microbial transformations. For example, reversal of chloramphenicol glucuronidation by bacteria in the intestinal tract (99) may lead to bacterial production of a hematotoxic chloramphenicol metabolite, dehydrochloramphenicol (100).

NON-ANTIBIOTIC DRUGS AFFECT GUT MICROBES

Antibiotic (101) and host-directed drugs (102) can alter the composition of the human gut microbiome, which may, in turn, contribute to gastrointestinal adverse events and potentially select for antibiotic resistance (Fig. 1). This impact likely arises from a combination of direct growth inhibition (102), altered bacterial gene expression (103), and indirect (host-mediated) effects. Direct growth inhibition of individual bacterial species is pervasive: 203 of 835 human-directed drugs that had not been classified as antibiotics inhibit the growth of at least one commensal species in vitro at concentrations predicted to occur in the human gastrointestinal tract (102). Bacterial community interactions and microbiome-mediated drug metabolism likely further shape these interactions in vivo.

As a specific example, different studies report that administration of the type 2 diabetes drug metformin correlates with increased abundance of Enterobactericeae (104, 105), decreased abundance of Bacteroides spp. (106), or decreased abundance of Intestinibacter spp. (104). These varied results may be due to study design: One study used samples from a large cohort of patients with type 2 diabetes who were or were not treated with metformin (104). Another study documented alterations in a population of three Dutch patient cohorts with only a small subset administered metformin (105), and the third interventional study collected samples from the same patients with type 2 diabetes before and after metformin treatment (106). Whereas interventional studies provide valuable information relating each patient to their pretreatment microbiome composition as the control, changes in disease state, diet, and other medications remain important variables to consider.

In another example, the antidepressant drug duloxetine is associated with decreased microbial diversity and decreased abundance of Ruminococcus spp. (107), whereas administration of another antidepressant drug, fluoxetine, correlates with reduced Lactobacillus johnsonii and Bacteroidales S24–7 (108). Proton pump inhibitors also alter the gut microbiome (105), with one study correlating proton pump inhibitors with changes in taxa associated with Clostridium difficile infections (109). Last, gut microbiome compositional responses to the immunomodulator methotrexate correspond with a positive therapeutic response (110). These effects on gut bacterial communities may be especially enhanced in drugs that are administered over extended time periods, as shown with proton pump inhibitors (111).

Host-directed drugs may also alter bacterial physiology through altered microbial gene expression. Certain host-directed drugs, such as digoxin, induce gene expression changes in complex human fecal communities in vitro after only four hours of exposure (103). RNA sequencing profiling of bacterial communities treated with individual drugs indicated that tetramisole, digitoxin, nizatidine, phenacetin, digoxin, and sulfasalazine collectively altered expression of 328 bacterial gene clusters. These differentially expressed pathways affect membrane transport, metabolism via the pentose phosphate pathway, and xenobiotic metabolism/biodegradation (103), likely allowing microbes to respond to the increased burden of drug exposure. Methotrexate acts similarly on both bacterial and human cells by inhibiting folate synthesis, thereby affecting pyrimidine and purine synthesis pathways and inhibiting growth in bacteria (110). Similarly, the anticancer fluoropyrimidine drug 5-fluorouracil inhibits mammalian and bacterial pyrimidine metabolism and affects expression of bacterial pyrimidine metabolizing enzymes; its prodrug, capecitabine, alters expression of bacterial genes associated with flagellar assembly machinery (17). Last, metformin elicits gene expression changes in E. coli that include increased expression of genes involved in butanoate production, quinone biosynthesis, sugar derivative degradation, and polymyxin resistance (105), which may provide a growth advantage during drug exposure.

Changes to bacterial gut composition and physiology may have consequences for human health. For example, host-directed drugs that disrupt growth of commensal microbes are more likely to be associated with antibiotic-like side effects in humans (102), potentially because these drugs and antibiotics both affect the gut microbiome. These interactions may have positive outcomes as well, as illustrated by methotrexate (110). This finding suggests that for some drugs, a patient’s response may result from the combination of drug-related alterations to the gut microbiome and the on-target effects of the drug. Transplantation of gut microbiomes from metformin-treated humans into germ-free recipient mice resulted in improved glucose tolerance relative to mice that received gut microbiome samples from non–metformin-treated donors (112). Beneficial effects from microbiome-metformin interactions may be partially explained by increases to genera such as Akkermansia (113), although it should be noted that the role of Akkermansia in health remains unclear (114). In addition, metformin-dependent microbiome changes may alter FXR-modulating members of the microbiome (106), which has been documented in the context of berberine-dependent changes in the gut microbiome that influence FXR signaling (115). Whereas the impact of metformin on some microbes may be beneficial, this drug is also associated with increases in Enterobacteriaceae that are associated with negative side effects in the gut (105, 116–119). Other studies highlight a potential role for host-directed drugs on antimicrobial resistance (102, 120). For example, overexpression of common antibiotic resistance mechanisms (e.g., drug efflux pumps) confers resistance to human-targeted drugs that inhibit commensal growth, suggesting that these compounds could promote the spread of antibiotic resistance in the microbiome (102). A metagenomic analysis of the impact of 759 drugs on the gut microbiomes of 4198 Japanese patients correlates patient medical drug use (particularly polypharmacy or the administration of multiple drugs) with increased antimicrobial resistance potential (120). These findings are further supported by in vitro studies demonstrating that exposure of E. coli to physiologically relevant concentrations of antidepressant drugs can promote development of resistance to multiple antibiotics (121). Thus, drug-induced changes to gut microbiome composition may represent both a signature of disease or treatment and a potential mechanism contributing to the therapeutic or toxic effects of a drug.

STRATEGIES TO DISSECT DRUG-MICROBIOME INTERACTIONS

Computational approaches

Computational approaches provide useful and scalable tools for identifying potential drug-microbiome interactions. Genomic and metagenomic datasets allow sequence homology searches to evaluate the distribution of known microbiome-encoded DMEs and identify enzymes with possible drug metabolizing activity. For example, pairwise alignment of known bacterial β-glucuronidases with human and mouse gut metagenomic datasets revealed new β-glucuronidase orthologs with different substrate specificities, and analysis of predicted active sites of these enzymes highlighted differences in substrate binding affinity (122, 123). Other studies use structure-guided approaches to characterizing potential DMEs, which can highlight critical active site residues (124).

In a broader genome-centered approach, AGORA2 (Assembly of Gut Organisms through Reconstruction and Analysis, version 2) uses metabolic reconstructions from more than 7000 bacterial genomes to predict drug metabolism by individual human gut microbiomes (125). This approach identifies microbiome-mediated drug transformations that are likely to be broadly shared across individuals and others that exhibit extensive interpersonal variation. Future experimental validation will provide valuable training sets to further refine these approaches.

Complementing these sequence-focused approaches, candidate bacterial DMEs can also be identified on the basis of their substrates: If an enzyme metabolizes a substrate with chemical similarity to a medical drug, then it may similarly transform the drug. For example, quantitative structure-activity relationship modeling can be used to predict new bacterial drug transformations (126). This approach uses bacterial chemical reactions in the MetaCyc database to identify similar transformations that could occur in medical drugs, as predicted by chemical relatedness (Tanimoto similarity) between the natural substrates of these enzymes and drugs (126, 127). This approach can also link known bacterial DMEs to additional drug substrates, as shown with E. lenta Cgr2 (127); the MicrobeFDT database aggregates these data for public access. However, these methods rely on previously identified bacterial DME substrates.

Other databases such as Kyoto Encyclopedia of Genes and Genomes and Clusters of Orthologous Groups can be used to identify new enzymatic pathways within the human microbiome, including those related to drug metabolism (128). Machine learning models also enable prediction of drug metabolism or accumulation by the gut microbiome. For example, researchers developed a machine learning model predicting the likelihood of microbiome-related drug metabolism or accumulation through physiochemical relatedness with known depleted (metabolized or bioaccumulated) or nondepleted (neither metabolized nor bioaccumulated) drugs (129). In addition, databases such as Side Effect Resource (SIDER) (130) and DrugBank (131) help to identify potential consequences of microbiome-drug interactions, although it should be noted that SIDER does not include information for all approved drugs.

Prediction of drug PK/PD through physiology-based pharmacokinetic models has generally focused on host factors and drug properties (132). This type of modeling uses previous knowledge regarding host physiology and anatomy, along with drug physiochemical properties, to parameterize algorithms to predict drug properties and drug exposure within the host (132, 133). Although a useful tool for understanding drug PK/PD, these approaches do not traditionally include the gut microbiome as a variable. One study used measurements of brivudine (drug) and bromovinyluracil (metabolite) exposure in the gut of gnotobiotic mice that exhibit or lack microbiome-mediated drug metabolism to parameterize a physiology-based pharmacokinetic model that includes gut microbiome activity as a component (18). The model estimates the gut microbiome contribution to serum drug and drug metabolite exposure and further predicts how other factors (absorption, transit time, enterohepatic circulation, drug metabolism by the host, and excretion) modulate this effect (134). Together, these studies provide new approaches to identifying and quantifying gut microbiome contributions to drug PK/PD and absorption, distribution, metabolism, excretion, and toxicity profiles.

In vitro and ex vivo approaches

In vitro and ex vivo studies provide an important complement to computational strategies. A high-throughput screen of the drug metabolizing capacity of human gut microbes mapped the activity of 76 individual bacterial gut isolates on the metabolism of 271 chemically diverse drugs and determined that 176 of the tested drugs were metabolized by at least one gut isolate (3). In ex vivo approaches, gut microbial communities from humans or animals are cultured in the laboratory to recapitulate the donor’s gut microbiome in a more controlled setting (2, 16). This approach, supported by culturing methods for improved replication of a donor community’s composition and functional diversity, identifies new transformations in complex bacterial communities (2). Whereas some microbes implicated in drug metabolism (e.g., E. lenta) have traditionally been resistant to standard genetic approaches, new methods are emerging that address these limitations (135).

Proteomics provides another powerful tool to link enzyme presence with observed metabolic transformations during ex vivo assays. In one example, an activity-based probe-enabled proteomics pipeline was used to characterize and quantify the distribution of bacterial β-glucuronidases in multiple human fecal communities (136). The authors identified the specific bacterial β-glucuronidases responsible for toxic reactivation of SN-38, as well as the differential distribution of these enzymes in a human cohort (136). Such studies provide a foundation for targeted inhibition of key β-glucuronidase enzymes to prevent reactivation of drugs responsible for adverse side effects in humans, similar to studies already completed in mice (84–86).

Although in vitro and ex vivo studies have key advantages (scalability and reproducibility), important limitations exist. First, in vitro culture can alter community composition. Cofactors, nutrients, and host factors can affect bacterial community function by modulating microbial gene expression and altering the amounts of alternate substrates or enzyme inhibitors (2, 137). In addition, these approaches generally do not incorporate host absorption, metabolism, excretion, or gut physiology, yet each of these factors can be critical in understanding the impact of the gut microbiome on drug metabolism and systemic drug exposure. For example, if a drug is absorbed before interaction with the active microbes (without enterohepatic circulation), then the impact of microbial metabolism on systemic parent drug exposure may be limited. Even in this case, however, microbial metabolism can play a key role in determining drug and drug metabolite concentrations in the gut and systemic exposure to microbially produced drug metabolites or metabolites resulting from combined host and microbial metabolism.

In vivo approaches

Many animal species, including dogs, pigs, nonhuman primates, and rodents, have been used for preclinical drug metabolism studies (138). Mice are widely used as a mammalian model for studying the effects of the gut microbiome on drugs. However, mouse and human gut microbiomes are distinct, and mice used in different studies of the same drug may differ in lineage, vendor, housing, and diet, all potentially contributing to variations in gut microbiome composition (139). These variations may result in inconsistent results regarding drug PK/PD or adverse effects. Using littermates and cohousing animals can help to reduce gut microbiome variability within studies. Mouse vendor-specific or vivarium-specific differences in drug metabolism may also provide important clues in understanding the effects of the gut microbiome. For example, mouse vendor-specific differences in immune profiles helped to uncover the role of specific gut microbes in the development of T helper 17 cells (140).

Antibiotic-treated, germ-free, or gnotobiotic mice can be used to directly measure the contribution of the gut microbiome to drug responses (3, 18, 41, 49, 134). Gnotobiotic mouse models (including those carrying a defined microbial community or transplanted with a human gut microbiota) allow gut microbiome manipulation in genetically identical hosts, which can help to distinguish the impact of host and microbiome variation on drug metabolism (Fig. 3A). Mice can be stably colonized with simple or complex human microbial communities (141), potentially providing a useful and replicable model for studying the role of human microbes in drug metabolism in vivo. However, it should be noted that human microbial community composition changes upon transplantation into the mouse gut and that humanized gnotobiotic mice do not necessarily mirror the gut microbiome composition of their human donors (142). Further, some physiological effects of Germ-free status (for example, delayed immune development) may be retained even after microbial colonization of the gut. Some strategies to minimize these effects include colonization of mice through fecal microbiota transplants soon after birth or at the time of weaning (143) or use of offspring of ex–Germ-free mice that were colonized with the desired bacterial community before giving birth.

Last, mice expressing human DMEs or populated with human hepatocytes may be useful in understanding how drugs interact with the host via the gut microbiome (Fig. 3A) (144). Such animals could potentially be colonized with human microbiomes, including patient microbiomes or samples representing known microbiome variation in a patient population. This could open the door to studies of microbiome-drug interactions in a replicated, controlled environment that would allow invasive sample collection (e.g., collection of tissue or lumen contents from small intestine, cecum, and proximal colon) and interventions that are not possible in humans (Fig. 3A).

Human studies

To understand how drugs affect the human gut microbiome, studies generally use correlative analysis of large populations (on the order of hundreds to thousands of study participants) (82, 104, 105, 145) or small-scale patient studies that incorporate metabolomic and metagenomic data (77, 146, 147). As a result of the efforts of the Human Microbiome Project and MetaHIT Consortium, large human gut microbiome metagenomic datasets are now accessible and numerous (148–150). For example, microbiome data and associated patient metadata from the MetaHIT Consortium have been used to identify factors within the human gut microbiota influencing metformin efficacy (104). In addition, two separate studies investigated Dutch (LifeLines-DEEP) and UK (TwinsUK) cohorts for taxonomic and functional changes within the gut microbiota in response to commonly prescribed drugs and therapeutic agents (105, 145).

Smaller-scale human studies have addressed mechanistic determinants of drug bioavailability, the presence of drug metabolites, or microbiome effects on drug efficacy (77, 146, 147). For example, a cohort of 10 individuals helped to reveal a positive correlation between the abundance of the tyrDC gene in patient microbiomes and the required levodopa dose (patients who required higher levodopa dosage for symptom relief had higher tyrDC abundance in their gut microbiomes) (77). In another example, patient responsiveness to methotrexate treatment was correlated with pretreatment gut microbiome composition and the degree of methotrexate metabolism by patient gut microbiomes ex vivo (146), suggesting the involvement of the gut microbiome in methotrexate treatment efficacy. In a third example, chemotherapeutic efficacy of drug cocktails targeting pancreatic ductal adenocarcinoma was correlated with neutropenia (lower numbers of neutrophils) and higher amounts of the microbial metabolite indole-3-acetic acid (147). It is important to also note that other studies have found a lack of correlation between metagenomic abundance of predicted DMEs and observed drug metabolites (94, 151).

CONCLUSION

Recent progress in understanding microbiome-host-drug interactions benefits from the wide variety of research approaches used, including systematic high-throughput studies (2, 3, 27), large human microbiome metagenomic datasets (148–150), integration of experimental approaches using animal models (3, 18, 41, 49, 134), and incorporation of the gut microbiome into physiology-based pharmacokinetic modeling (18, 134). Opportunities exist to further understand these important relationships between the microbiome and drug responses (Fig. 3A). For example, known bacterial drug transformations and drug accumulations identified in high-throughput screens (2, 3, 27) provide a starting point to address two key questions: Given the composition and gene content of a patient’s gut microbiome, can we predict whether and how different drugs will be metabolized? In addition, given the chemical structure of a candidate drug, can we predict whether and how gut microbiomes from different individuals will affect its efficacy and toxicity?

Pharmacokinetic and gut microbiome data collected in parallel from longitudinal studies of large cohorts of patients will be critical to dissect drug-gut microbiome interactions in humans. In vitro–in vivo extrapolation approaches (which provide strategies to predict in vivo drug PK/PD from in vitro measurements) could also be extended to include microbiome activity. In addition, collection of human gut microbiome samples from drug responders and nonresponders will provide a resource for obtaining new microbial isolates with unique drug-metabolizing properties that will expand our understanding of the underlying mechanisms of drug metabolism. Parallel PK/PD studies in patients and in humanized mice carrying patient microbiomes will provide further opportunity to functionally connect gut microbiome activities to patient drug responses (Fig. 3).

Increased understanding of drug-microbiome-patient interactions could be used to ensure that clinical trials measuring drug safety and responsiveness include relevant microbiome variation and to select the choice of drug, route of delivery, and dose for individual patients based on their gut microbiome composition (Fig. 3B). Looking further ahead, interventions that target problematic drug-microbe interactions (or promote beneficial interactions) could be achieved by manipulation of a patient’s microbiome through multiple routes, including prebiotics (152, 153), probiotics (152, 153), postbiotics (153, 154), targeted antibiotics or bacterial enzymatic inhibitors (16, 84–86, 95, 152), bacteriophages (152, 155), or fecal microbiota transplants (152) (Fig. 3B). Further, pharmaceutical compounds may help to protect commensal bacteria from antibiotic-mediated gut microbiome disruption (156).

Several examples highlight the potential benefits of progress in this field. In one case, vaginal administration of tenofovir gel significantly reduced HIV incidence in women who carried a Lactobacillus-dominant vaginal microbiota (61% reduction, P = 0.013). By contrast, women who harbored Gardnerella-dominated vaginal microbiomes showed no significant reduction in HIV risk when treated with the same drug (157). Gardnerella species directly metabolize tenofovir in vitro, suggesting that these relationships can be predicted. However, microbiome analysis is not typically included in clinical trials. In a second set of promising examples, responsiveness to immune checkpoint blockade inhibitors is associated with gut microbiome composition in mice and humans (158–162). Distinct microbial species correlate with treatment responsiveness in different cohorts (158), suggesting multiple mechanisms or convergent effects. Enterococcus may provide one mechanistic link: Patients with melanoma who responded well to PD-1 blockade therapy displayed higher abundance of gut Enterococcus faecium (159). The underlying mechanism of this increased efficacy may include prophage expression, which correlates with therapeutic efficacy in humans and improved responses in mice (160). It may also include the activity of an Enterococcus enzyme, SagA, which leads to production of immunostimulatory muropeptide fragments that enhance immune checkpoint inhibitor efficacy (161). In addition, production of inosine by Bifidobacterium pseudolongum leads to increased antitumor immunity upon immune checkpoint inhibitor treatment in mice (162). Together, these studies may provide a glimpse into the future of microbiome-informed drug development, delivery, and monitoring that expands beyond the gut microbiome and outside of small-molecule drugs. Strategies that integrate the impact of the microbiome on pharmacotherapy thus represent new opportunities to improve health across disease indications.