The symbiotic relationship between mammals and the trillions of microbial cells that reside in their gastrointestinal tracts relies on a complex molecular dialogue, with microbial metabolites acting as major mediators of this dialogue. Essential roles for several microbial metabolic pathways in host physiology have been long established, including in the production of vitamin K, the production of water-soluble B vitamins including biotin, folates, nicotinic acid, pyridoxine, riboflavin, cobalamin and panthotenic acid, the degradation of dietary oxalates, and modification of bile salts (reviewed in [1, 2]). However, intense interest in the gut microbiota over the past decade has led to the discovery of many new areas where bacterial transformation of dietary components may play critical roles in host health and disease. This increased understanding of diet–microbiota–host interactions suggests significant opportunities to create new therapeutic approaches, including selectively altering the microbial production of molecules to promote human health and prevent disease [3].

This review will focus on several key metabolites formed by the gut microbiota from dietary components that have been revealed recently to produce remarkable effects on host physiology and that are currently being targeted or have high potential to be targeted as treatments for human disease. We will describe briefly the microbial origin of these metabolites and the biological actions of these metabolites on their host. We will then discuss in more detail current and potential therapeutic approaches to manipulate these metabolite levels and broader areas of research that are needed to understand the potential value of gut microbial metabolites.

TMA and its co-metabolite TMAO were identified by screening metabolites associated with cardiovascular disease (CVD), and TMA was shown to require gut bacteria for its formation [15]. Cleavage of choline to TMA and acetaldehyde by two enzymes originally identified in Desulfovibrio desulfuricans, CutC and CutD, allow choline to be used as an energy source [108]. Recent studies found homologous genes in a variety of Proteobacteria and Firmicutes, and to a much lesser extent Actinobacteria, suggesting spread via horizontal gene transfer [109]. TMA was also recently shown to form from l-carnitine and choline via an analogous reaction catalyzed by the YeaW and YeaX enzymes originally characterized in Escherichia coli [110], and by CntA and CntB, originally characterized in Acinetobacter baumannii [111]. After formation and absorption in the colon, TMA passes into the portal circulation, which directs blood into the liver, where it is oxidized to TMAO by flavin-containing mono-oxygenase 3 (FMO3) [112]. Analysis of genetic variation among inbred strains of mice indicates that plasma TMAO levels significantly correlate with FMO3 activity [112]. Oral antibiotics block the increase in TMAO that normally occurs after dietary challenge with either choline or carnitine, demonstrating that the generation of TMAO requires microbial bacteria [15, 113, 114].

TMAO levels predict risk for atherosclerosis [15, 112, 115], and are elevated in patients with chronic kidney disease (CKD) [116] and obesity [17, 98], and decreased in ulcerative colitis [117]. TMAO directly induces CVD, as administration of TMAO itself or of sufficient choline or l-carnitine to raise TMAO levels can all increase atherosclerosis in Apoe ^−/− mice [15, 114]. The specific molecular mechanisms by which TMAO exerts its pathological effects are currently unknown. Accumulation of TMAO in the kidney may alter osmotic balance and elevated TMAO levels associate in animal models with markers of renal damage such fibrosis and dysfunction [116]. Thus far, no receptor for TMAO has been identified. TMA, but not TMAO, acts as a ligand for trace amine-associated receptor 5 (TAAR5) [118], but TAAR5 appears to be exclusively expressed in the olfactory sensory neurons. Administration of TMAO to Apoe ^−/− mice inhibits reverse cholesterol transport from macrophages in vivo [114], but treating macrophages directly with TMAO in cell culture does not increase their ability to take up cholesterol or inhibit their ability to efflux cholesterol to ApoA1 or HDL [119]. Reduction of FMO3 activity (which increases TMA levels and decreases TMAO levels) decreases intestinal cholesterol absorption, reduces hepatic biliary secretion and LXR signaling, and increases cholesterol disposal via transintestinal cholesterol efflux (active secretion of cholesterol from the small intestine) [120]. Administering antibiotics blocks these effects, while TMAO supplementation does not, suggesting that the effects of reducing FMO3 activity resulted from increased TMA or another microbial substrate of FMO3 [120]. Thus, studies elucidating the molecular targets of TMAO and the potential roles of TMA are greatly needed.

Without identified TMAO molecular targets, interventions to reduce CVD must focus on reducing TMAO levels. Reducing dietary choline or l-carnitine would lower TMAO levels, but may have undesirable effects. In particular, supplementation with lower levels of l-carnitine than needed for TMAO formation may improve cardiovascular function [121]. A meta-analysis of 13 controlled trials (n = 3629) showed that l-carnitine supplementation reduces all-cause mortality by 27 % [122]. While potentially beneficial for cardiovascular health, choline deficiency markedly increases risk for non-alcoholic liver disease. Inhibiting FMO3 to reduce TMAO levels is also undesirable, as accumulation of TMA results in fish malodor disorder. Because of these limitations, current pharmaceutical development is focusing on a revolutionary approach: non-lethal targeting of microbes by selectively inhibiting pathways detrimental to their host, such as microbial CutC/D, CntA/B and YeaW/X. A structural analog of choline, 3,3-dimethyl-1-butanol (DMB), non-lethally inhibits microbial CutC/D and reduces TMAO levels in mice fed a high-choline or l-carnitine diet [123]. Importantly, DMB inhibits macrophage foam cell formation and atherosclerotic lesion development in Apoe ^−/− mice [123]. Future clinical trials are needed to determine the safety and efficacy of CutC/D inhibitors in reducing TMAO levels and disease in humans, as well as whether resistance to their effects will occur with long-term treatment strategies. Nevertheless, this revolutionary strategy of selective, non-lethal inhibition of microbial function likely represents an important new front in the pharmacological treatment of human diseases.

Tryptophan is an essential amino acid found in a variety of foods such as red meat, fish and eggs. Commensal bacteria expressing tryptophanase catabolize tryptophan to indole, a quorum-sensing compound for bacteria [124] (Fig. 1). Lactobacillus spp. convert tryptophan to indole-3-aldehyde (I3A) through unidentified enzymes [125]. Clostridium sporogenes convert tryptophan to IPA [6], likely via a tryptophan deaminase. After absorption from the intestinal tract into portal circulation, the liver converts indole to IndS.

Indole and its metabolites affect host physiology via a number of molecular mechanisms (Fig. 1). Indole and I3A are agonists for the aryl hydrocarbon receptor (AhR), a transcription factor that regulates interleukin (IL)-22 expression, increases T_H17-cell activity, and helps maintain intraepithelial lymphocytes [125]. Indole upregulates the expression of tight junction proteins and modulates the expressions of pro- and anti-inflammatory genes in intestinal epithelial cells [126, 127]. These activities of AhR help ensure that commensal bacteria outcompete pathogenic bacteria in the gut microbiota [128], and the absence of AhR increases the severity of DSS-induced colitis [129] and response to Citrobacter rodentium infection [130] (a model of human enteropathogenic E. coli infections). In addition to these effects, recent studies show that indole also modulates GLP-1 release from L cells [131], so that indole formation may contribute to satiety and inhibition of obesity. Other recent studies demonstrate that IPA is a pregnane X receptor (PXR) agonist, particularly in the presence of indole [132]. A wide-range of PXR agonists inhibit NF-κB [133], and downregulation of intestinal tumor necrosis factor (TNF)-α and upregulation of junction proteins by IPA requires PXR [132]. IPA also potently scavenges hydroxyl radicals [134], thereby protecting against oxidative injury in various animal models [134–137]. Thus, future studies are needed to determine if enhancing IPA formation by bacteria or directly administering IPA is beneficial in inflammatory conditions such as inflammatory bowel disease and colorectal cancer.

While indole appears to be primarily beneficial, its metabolite IndS is a uremic toxin that accumulates in patients with CKD [138]. IndS is also associated with accelerated glomerular sclerosis [139], enhanced endothelial dysfunction [140], enhanced monocyte adhesion to the vascular endothelium [141], and increased oxidative stress [141, 142]. The oral charcoal adsorbent AST-120 binds indoles in the gut lumen and reduces plasma IndS levels, thereby reducing kidney damage and atherosclerosis associated with kidney injury [143]. Future studies are needed to determine if diverting tryptophan metabolism away from IndS towards IPA will be beneficial in renal disease or other conditions.

PCS and 4-ethylphenyl sulfate (EPS) are structurally similar uremic toxins formed by hepatic sulfation of the microbial metabolites para-cresol and 4-ethylphenol, respectively. The lack of PCS or EPS in the plasma and urine of germ-free mice demonstrates their microbial origins. Inactivating mutants of the hydroxyphenylacetate decarboxylase operon genes (hpdB/C/A) from Clostridium difficile prevent fermentation of tyrosine or its metabolite hydroxyphenylacetate to para-cresol [144]. Few other gut bacteria encode HpdB/C/A [144]. Bacterial pathways for formation of 4-ethylphenol have not yet been characterized, but the wine spoilage yeast Brettanomyces generates 4-ethylphenol from the tyrosine metabolite para-coumaric acid that is present in many foods via cinnamate decarboxylase and vinyl phenol reductase. 4-Ethylphenol also forms from orally administered genistein, a phytoestrogen found in soy, by uncharacterized but presumably microbial pathways [145].

Both PCS and EPS accumulate in patients with severe CKD undergoing hemodialysis [146]. PCS levels predict clinical outcomes in patients with CKD [147] and correlate with cardiovascular mortality in CKD patients [148, 149]. While conventional dialysis fails to remove PCS, treatment with the oral adsorbent AST-120 [150] or with the prebiotic arabino-xylo-oligosaccharide [151] lowers plasma PCS levels. Vegetarians have lower levels of PCS than omnivores [152]. There are very few studies of EPS. EPS levels are elevated in a rat model of chronic renal failure and AST-120 treatment lowers these levels [153]. EPS levels increase 46-fold in a mouse model of autism and treatment with Bacteroides fragilis blocks this increase [28]. Administration of EPS to mice results in anxiety-like behaviors [28].

Molecular mechanisms of action ascribed to PCS include direct damage of cell membranes [154], induction of apoptotic pathways [155], activation of NADPH oxidase 4 (NOX4) resulting in reactive oxygen species (ROS) formation [156], activation of JNK and p38-MAPK [157], activation of Rho-kinase (ROCK) leading to endothelial damage [158], activation of epidermal growth factor (EGF) receptor leading to expression of matrix metalloproteinases 2 and 9 [159], and inhibition of a variety of drug-metabolizing enzymes including CYP2E1, CYP3A4, UGT1A1, UGT1A9, and UGT2B7 [160]. Given its chemical similarity to PCS, EPS is expected to exert similar effects, but no specific molecular targets have been demonstrated to date. Future studies are needed to identify pharmaceutical inhibitors of the PCS and EPS biosynthetic pathways and whether such inhibitors have beneficial effects in disease.

The microbiota of ruminants have long been known to transform the essential fatty acids linoleic acid (LA) and linolenic acid to CLAs such as cis-9 and trans-11 CLA, and conjugate linolenic acids (CLnAs) such as cis-9, trans-11 and cis-15 CLnA, respectively [161–163], via the action of isomerases. However, recent studies found that the microbiota of mice and humans, particularly Lachnospiraceae, Lactobacillus spp. and Bifidobacteria, possess the capacity to generate both CLAs and CLnAs [164–166]. In Lactobacillus, intermediates for the formation of conjugated fatty acids include the oxygenated metabolites HYA and 10-hydroxyoctadecanoate (HYB) [167, 168]. The enzymes involved in the transformation of LA to CLAs by Lactobacillus were recently characterized and include myosin-cross-reactive antigen, short-chain dehydrogenase/oxidoreductase, and acetoacetate decarboxylase [169].

Conjugated fatty acids exert many highly beneficial effects, including reduction of adiposity, improved insulin sensitivity, reduced carcinogenesis, and reduced atherosclerosis (reviewed in [170]). CLAs and CLnAs act via PPAR-γ (reviewed in [171]), PPAR-α [172], and inhibition of cyclooxygenases and lipoxygenases [173, 174]. Whether typical intestinal microbiota generate sufficient CLA/CLnA to exert the extraintestinal effects seen with CLA/CLnA supplementation is unclear, as feeding essential fatty acids increases gut but not circulating levels of CLAs and CLnAs [164]. Like CLAs and CLnAs, HYA also exerts anti-inflammatory activities, including downregulating lipopolysaccharide (LPS)-induced maturation of dendritic cells, blocking TNF-induced barrier impairment, and protecting against DSS-induced intestinal injury [175, 176]. HYA acts via the GPR40–MEK–ERK pathway [176]. Future studies are needed to determine if increasing microbial HYA production can be used therapeutically.

In previous sections, we have touched briefly on potential future studies for individual metabolites, but there are additional developments needed in broad areas of research and understanding to fully realize the potential of gut microbial metabolites for disease treatment. We will conclude by highlighting four of these needed developments.

First, the development of minimal sets of biomarker microbial metabolites that identify particular disease states or that distinguish between closely related disease conditions. The analysis carried out by de Preter and colleagues for inflammatory bowel disease is proof of principal for this strategy [22], and similar approaches for highly heterogeneous conditions such as autism spectrum disorder, in which the microbiota has also been implicated [177], might be even more valuable. This also applies to the identification of individuals who might be at risk for disease, such as was found for individuals who carried high levels of bacterial strains that converted cholesterol to coprostanol that made them more vulnerable to C. difficile infections. For translation to actual treatment, measurements will need to be carried out in clinical laboratories in which immunoassay arrays, rather than the more sophisticated MS or NMR methods available in research settings, will likely continue to be the primary methods available. Thus, identifying the minimal number of biomarker metabolites needed to selectively assess a condition is critical. Similar strategies can be used to determine the efficacy and safety of interventions.

Second, the development of algorithms to predict personalized responses to dietary and pharmaceutical interventions based on microbial metabolites. An exciting example of this approach was recently reported by Zeevi and colleagues, who demonstrated that the highly variable glycemic response of different individuals to the same foods could be predicted using their gut microbiota and other data [32]. Similarly, being able to predict the responses of specific metabolites such as SCFAs to individual foods using tools such as CASINO [31] may be critical for allowing individuals with intolerance for particular dietary components to successfully use functional foods to increase colonic levels of SCFAs. Algorithm-based personalization seems essential for any nutrition-based approaches, given the variability of microbial composition among individuals.

Third, the development of readily generalizable methods to increase gut microbial production of beneficial metabolites, either by selectively increasing the abundance of native species that produce that metabolite or by engineering endogenous gut microbiota to produce it in high levels. An example of this latter approach is our study using heterologous expression of the satiety factor N-acylphosphatidylethanolamine in commensal E. coli (strain Nissle 1917), leading to inhibition of obesity in mice fed a HFD [178]. Such strategies might be helpful to produce sufficient IPA, CLA or HYA to block inflammatory diseases, but could also be utilized to test novel metabolites as they are identified. One advantage of engineered bacteria may be the ability to produce beneficial metabolites in bacterial strains that colonize well in the gut of a diseased individual in the place of native bacteria that produce these same beneficial metabolites but poorly colonize in the diseased gut.

Fourth, the development of non-lethal specific inhibitors for various microbial pathways that produce harmful metabolites, similar to work done with CutC/D. In particular, inhibition of the formation of para-creysl and 4-ethylphenol appear amendable to this strategy. This revolutionary approach to controlling harmful bacterial metabolites seems unlikely to result in the rapid evolution of resistance that occurs with standard antibiotics, since there is a much more limited fitness advantage of carrying resistance. If this is the case, then long-term use of such metabolic pathway inhibitors will have great potential benefit in chronic diseases.

The past decade has seen remarkable progress in our understanding of the significant role that gut microbial metabolites play in modulating the health of their hosts. MS and NMR studies have identified a significant number of microbial metabolites that differ in disease conditions, and these same methods are now being exploited to better identify subtle differences in closely related diseases. Some of these identified metabolites, such as TMAO, IndS and PCS, appear to directly increase susceptibility to disease, while others, such as SCFA, IPA, CLA and HYA, appear to exert protective effects. Much work remains to fully characterize the physiological effects of these and the many other microbial metabolites that may be important in human health. It seems highly likely that future studies will identify many other disease states in which gut microbial metabolites are significantly enriched or depleted. It is important to keep in mind that by themselves such studies do not demonstrate causality. Thus, it seems there is a considerable need for carefully controlled studies to determine the physiological effects of each identified microbial metabolite and its specific mechanisms of action. Furthermore, in order to fully exploit the potential of the gut microbiota for disease prevention, we need a much greater understanding of how dietary components and host genetics affect the production of various metabolites. Finally, translation of these findings to clinical practice will require the development of widely available clinical chemistry methods to detect changes in an individual’s key metabolites. Despite these tremendous challenges to fully exploiting the gut microbiota for human health, the remarkable progress of the last decade suggests that such approaches have significant potential to revolutionize therapeutic approaches to human disease.