Gut Microbiota and Cardiovascular Disease

Due to advances in culture-independent sequencing technologies and bioinformatics, the vast majority of gut microbiome–related studies assess gut (often fecal) microbial community compositional changes associated with various disease states. This has led to a wealth of associative data, which, while helpful, are limited in terms of investigating causality. While it is difficult to define a truly pathogenic bacterial community, the term dysbiosis has been used to describe an imbalance of intestinal microbiota composition within a disease state or phenotype. Many investigators have reported an association between CVD phenotypes and changes in the relative abundance of specific microbial taxa or gut bacterial richness or diversity. For example, in early studies, bacterial DNA was detected in atherosclerotic plaques with signatures that match taxa associated with disease states,⁶ and microbial compositional changes have been reported in patients with numerous CVD risk factors, including hypertension, dyslipidemia, insulin resistance, and other metabolic phenotypes.^10,11 Changes in microbiota composition, diversity, and richness are associative, making it difficult to determine whether microbial changes are driving disease states or rather being driven by them. Moreover, typically-used sequencing techniques often do not reach species or strain level resolution, and analyses often exclude less abundant microbes. Studies focusing primarily on abundance can thus neglect gain-of-function microbial pathways that contribute to disease, despite arising from microbes that can represent only a small proportion of the microbial community. A recent example of this is shown in a series of microbial transplantation studies designed to demonstrate a direct role of gut microbial cut(choline utilization) C—a major microbial gene responsible for choline→trimethylamine (TMA) transformation^12,13—in host thrombosis potential¹⁴ (Figure 2). These studies used germ-free (GF) mice colonized with a synthetic polymicrobial community lacking choline→TMA functional capacity, and the addition of a human commensal (Clostridium sporogenes) genetically engineered to either possess or lack a functional microbial cutC gene. While only present at a low abundance (only ≈0.1%) within the large intestine gut microbial community, the gain-of-function C sporogenes mutant elevated circulating TMA and TMAO levels within the host, and both enhanced rate of thrombus formation and reduced time to cessation of blood flow following arterial injury in vivo.¹⁴

Whole genome sequencing coupled with bioinformatics analyses can also be used to infer potential functional capacities of gut microbial communities. For example, deep sequencing and systematic analyses of biosynthetic gene clusters identified the presence of common antibiotic genes in human commensals.¹⁵ But just as in studies in vertebrate animals, the mere presence of a gene within bacteria does not provide information about its expression and activity. What’s more, the majority of microbial genes have unknown functions. Further, microbial protein expression is highly dynamic, as bacterial gene regulation is greatly influenced by environmental cues. This is demonstrated in recent work showing that dietary addition of a single type of fiber can cause dramatic changes in the expression of genes required for bacterial metabolism of fiber glycans.¹⁶ So while studies and approaches that focus on microbial gene abundance and inference of potential functional capacity can be valuable tools for hypothesis generation, they need to be experimentally validated.¹⁷

In its healthy state, intestinal barrier function is maintained by physical factors including tight junctions between epithelial cells, mucus production, and mucosal immunity. And in heart failure patients, we often observe bowel wall edema and impaired barrier function.^18,19 Following the leaky gut concept, impaired gut barrier function leads to translocation of bacterial products into host circulation, which can result in a proinflammatory state. Multiple studies show patients with heart failure have alterations in intestinal integrity and that elevated levels of proinflammatory cytokines in the blood correlate with symptom severity and poorer outcomes.^20,21 In the presence of heart failure, venous fluid overload, adaptive sympathetic activation resulting in redistribution of the systemic circulation, and low cardiac output contribute to bowel wall edema and reduced mucosal perfusion.²² Intestinal hypoperfusion in heart failure alters mucosal function, as evidenced by increased paracellular permeability and augmented intestinal bacterial biofilm formation.²³ When the gut barrier is impaired, lipopolysaccharide (LPS) originating from Gram-negative bacteria can enter the host circulation, where it is mainly recognized by TLR (toll-like receptor) on the surface of immune cells.²⁴ Upon binding of bacterial ligands, TLR signaling induces the release of proinflammatory cytokines that orchestrate a proinflammatory state in the host.

Enhanced levels of LPS and other bacterial wall products, presumably derived from gut microbiota, have been mechanistically linked to modulation of inflammation, immunity, and vascular function (Figure 1). Patients with decompensated heart failure have higher endotoxin levels in the blood compared with stable patients.²⁵ Translocation of LPS from the bowel in this setting is supported by higher endotoxin concentrations in the hepatic vein as compared with blood directly sampled from the ventricles.²⁶ The detection of gut microbiota and metaorganismal metabolites at heightened levels in subjects with CVD, or risk for incident adverse CVD events, may, in part, reflect alterations of the host barrier function (Figure 1). In a recent observational study, circulating LPS concentrations were predictive of major adverse cardiac events in a cohort of patients with atrial fibrillation, suggesting that endotoxin translocation impacts CVD complications.²⁷ It is worth noting that a Mediterranean diet was negatively associated with endotoxemia in this study, suggesting the involvement of the gut microbiota in gut barrier function of the host. Indeed, in preclinical models, gavage of Western diet–fed ApoE (Apolipoprotein E)^−/− mice with live Akkermansia muciniphila decreased intestinal permeability and lowered fecal and circulating LPS levels, which was associated with reduced aortic atherosclerosis independent of lipid metabolism.²⁸ While the mechanism of Akkermansia-induced changes in barrier function needs to be explored, a recent proof-of-concept study in humans found that administration of pasteurized (but not live) A muciniphila for 3 months led to reduction in plasma LPS in obese individuals with metabolic syndrome.²⁹ Although the intervention did not change body weight, the authors observed improvement in insulin sensitivity and dyslipidemia in the Akkermansia-treated group. Interestingly, structural differences of LPS subtypes from different gut microbial species have been associated with altered TLR recognition and their effects on host innate immunity.³⁰ For instance, gavage with the gram-negative Bacteroides (vulgatus and dorei) that produce penta- and tetra-acylated lipid A, in contrast to the hexa-acylated lipid A of E coli, reduced colonic inflammation, endotoxemia, and atherosclerosis in ApoE^−/− mice.³¹

It should be noted that factors beyond gut leakiness contribute to CVD. There are many disease conditions under which enhanced gut leakiness is present, but not all such diseases show heightened associations with CVD risks. For example, while the presence of heart failure–induced bowel wall edema has been linked to endotoxemia and CVD progression, gut barrier defects caused by colitis and inflammatory bowel diseases are not classically associated with heightened CVD risks. These observations point to a more complex relationship between bowel wall integrity, changes in gut microbial communities, and the relationship between host systemic inflammation and altered susceptibility for development of CVD. We speculate that different mechanisms of gut leakiness (eg, inflammatory, bowel wall edema) may differentially impact gut microbiome and pathophysiological processes that modulate susceptibility for the development of CVD. Improved understanding of the role of gut bacterial proinflammatory factors in triggering systemic inflammatory cascades may help to provide novel therapeutic strategies to improve care and risk stratification among patients with CVD.

Numerous lines of evidence link multiple different facets of inflammation to heightened risks for CVD.^32,33 The role of inflammatory pathways in CVD has recently been reaffirmed by the CANTOS clinical trial (Canakinumab Anti-inflammatory Thrombosis Outcome Study). Administration of canakinumab—an antibody against IL (interleukin)-1β—showed that inhibition of the IL-1β pathway reduces cardiovascular event risk independent of lipid level lowering.³⁴ Thus, treatment and prevention of CVD with immunomodulatory therapies seems promising. But targeting inflammatory pathways bears the risk for opportunistic infections, which may limit its use in patients with multiple comorbidities, as is often observed in patients with CVD. Identification of gut microbiota that elicit host immune responses that participate in CVD pathogenesis may provide a therapeutic avenue to ameliorate inflammation-driven CVD phenotypes.

A major function historically attributed to bile acids (BAs) has been to facilitate emulsification and adsorption of fat-soluble dietary nutrients. However, BAs are composed of a diverse array of structurally specific species whose concentrations differ by many orders of magnitude. More recent studies have shown structurally specific and distinct BAs also play additional roles, including but not limited to modulation of host lipid metabolism, glucose/insulin metabolism, and inflammation (Figure 1).

Initially synthesized from cholesterol in the host liver, primary BAs, which only represent a small fraction of the total BA pool, are then secreted into the intestinal (duodenum) lumen, where subsequent gut microbiota–dependent modifications participate in the generation of a remarkably large array of BA species. The body maintains a large pool of hydrophobic BAs through reuptake in the ileum and through negative feedback. These negative feedback mechanisms are triggered by activation of the FXR (farnesoid X receptor) and cholesterol 7α hydroxylase³⁵ or by expression of intestinal bile transporters.³⁶ BAs modulate gut microbial composition via potent antimicrobial properties, immune responses,³⁷ and FXR,³⁸ and bile obstruction can lead to bacterial overgrowth syndromes.³⁹ Gut microbiota modify primary BAs via bile salt hydrolysis and BA 7α dehydroxylation, yielding secondary BAs, many of which have hormone-like functions. Following systemic adsorption, some secondary BAs in turn impact host physiology though interaction with a variety of host nuclear receptors including FXR, LXR (liver X receptor), PXR (pregnane X receptor), and specific GPCRs (G-protein–coupled receptors) like TGR5 (takeda G-protein–coupled receptor 5; Figure 1).^40–44 Perturbations to the dynamic interactions among diet, gut microbiota, and specific BAs may thus contribute to cardiometabolic phenotypes and disease susceptibly. For example, altered levels of BAs in plasma are associated with insulin resistance in type 2 diabetes mellitus,^45,46 and modulation of BA signaling may contribute to metabolic improvements during some antidiabetic treatments.⁴⁷ As such, measuring systemic levels of BAs may aid in the assessment of potential gut microbiota contributions to cardiometabolic diseases.⁴⁶ A better understanding of how perturbation in BA profiles are associated with future development of disease states or responses to therapy may enhance our understanding of CVD pathogenesis. Functional studies to determine whether candidate BAs are mechanistically linked to these processes are promising areas of future investigation.

Almost a decade ago, Wang et al⁹ first causally linked gut microbiota with oral intake of nutrient precursors, TMAO production, and CVD risk (Figures 1 and 2). A combination of untargeted metabolomics and mechanistic animal model studies was used to uncover small molecules whose levels in blood associate with CVD risk in humans, and impact CVD-relevant phenotypes in animal models. Several analytes linked to phosphatidylcholine metabolism, including TMAO, were identified, and TMAO was shown to both predict CVD risks in multiple clinical cohorts and facilitate accelerated atherosclerosis (as did nutrient precursors with an intact host gut microbiome) in animal models.⁹ Through these and other studies, generation of TMAO in humans and mice has been shown to occur via a multistep, metaorganismal pathway starting with the dietary precursors choline,⁹ phosphatidylcholine,^8,9 and carnitine.⁷ These are most abundant in foods found in a Western diet, including red meat, egg yolks, and other animal products (Figure 2). Notably, plasma or serum levels of every additional TMA nutrient precursor identified (ie, shown to generate TMA and TMAO in hosts via a gut microbiota–dependent fashion), including betaine,⁹ γ-butyrobetaine,⁷⁴ and trimethyllysine,⁷⁵ have all similarly shown associations with incident CVD risks in large-scale clinical studies, and these associations appear to be mediated by TMAO (since their clinical prognostic value becomes attenuated with TMAO in the model).

Gut microbial metabolism of TMA-containing nutrient precursors begins with specific microbial TMA lyases that generate TMA—an odorous gas—as a product. The major microbial choline TMA lyase is thought to be encoded by the microbial cutC/D genes (cut gene cluster genes C [catalytic] and D).¹² The TMA produced is then transported to liver via the portal vein and readily metabolized by host hepatic FMOs (flavin monooxygenases; mainly FMO3⁷⁶) into TMAO.⁹ In the circulation, TMA levels are typically negligible. When radiolabeled TMA or TMAO was orally administered to human volunteers, 95% of the dose label was excreted via the kidneys with the majority being TMAO.⁷⁷ TMAO has been shown to enhance atherosclerosis in most, but not all, mechanistic and animal model studies.^7,9,78–82 It has also been shown to promote platelet reactivity and thrombosis potential,^14,83–85 vascular inflammation and inflammasome activation,^86–88 heightened heart failure,^89–91 and chronic kidney disease^92–95–related phenotypes in animal models (Figure 2).

Circulating levels of TMAO have been shown to associate with CVD and predict outcomes in the presence of multiple CVD phenotypes, including peripheral artery disease,⁹⁶ coronary artery disease,⁹⁷ acute coronary syndrome,^97–99 and heart failure.^100–104 Notably, the prognostic value of TMAO withstands adjustment for traditional risk factors, highlighting its potential as a biomarker for risk stratification beyond what has been considered traditional risk. While not all studies have observed the relationship between heightened TMAO levels and incident CVD risks, examination of extant clinical studies with TMAO in multiple meta-analyses have concluded that a strong relationship indeed exists between elevated circulating TMAO levels and both CVD risk and mortality in multiple cohorts on different continents.^105–107 In many studies, a plasma TMAO cutoff value of >6 μM predicted heightened risk of adverse cardiac events.¹⁰⁸ And in one recent meta-analysis comprising >25 000 subjects, a 7.6% increase in all-cause mortality was noted for each 10-μmol/L increment of TMAO.¹⁰⁵

Several factors impact circulating TMAO levels within subjects.^19,108 First, the gut microbial community composition is critical since there is an initial obligatory role of gut microbes in TMA(O) generation.^7,9 Second, renal functional decline leads to less efficient excretion and thus heightened levels of TMA(O).⁹⁵ But elevated levels are also frequently observed among subjects with normal kidney function.¹⁰⁵ In all subjects, choline is a major and continuous nutrient precursor, since beyond diet, choline content in bile is remarkably high and thus baths gut microbes in both omnivore and vegans/vegetarians alike (the origins of the word choline are in the Greek word khole, for bile, since choline was first isolated by Adolph Strecker from pig and ox bile in 1862).¹⁰⁹ However, carnitine, which is found in high levels in red meat (and some energy drinks and over-the-counter supplements), also serves as a nutrient precursor and can account for significant elevation in TMAO levels, particularly in some omnivores.^7,110,111 Large-scale clinical observational studies show that subjects with heightened circulating carnitine levels have higher risk for incident CVD events (heart attack, stroke, and death). The prognostic value for plasma carnitine levels, like other TMA precursors, seems to be mediated by TMAO, as inclusion of TMAO in statistical models attenuates the prognostic value of carnitine (but TMAO remains a robust predictor), and carnitine supplementation accelerates atherosclerosis development in animal models.^7,112

In a recent human dietary intervention study examining protein source (red meat versus white meat versus nonmeat), substantially higher levels of circulating TMAO were observed when subjects consumed a red meat diet (equivalent to 8 oz of steak daily for 1 month).¹¹¹ Although there is a modest increase in choline content in omnivorous versus vegan diets, as noted above, a substantial amount of choline is introduced into the gut in the form of bile (choline in the form of phosphatidylcholine is a major component of bile). Thus, there is far more modest overall difference in choline exposure to the gut microbial community of vegan/vegetarian versus omnivore. By contrast, omnivorous diets show markedly higher carnitine content since vegan/vegetarian diets are virtually devoid of carnitine. In line with this, a major source of the observed elevation in plasma TMAO levels in subjects following 1 month of either a red meat–rich diet versus white meat or nonmeat diet (predominantly vegetarian protein source) was shown to arise from carnitine. In addition to having enhanced nutrient density of TMA(O) precursors (including carnitine), isotope tracer studies showed the red meat–rich diet induced functional remodeling of the gut microbial community to enhance carnitine→TMA transformation but not choline→TMA generation.¹¹¹ Interestingly, the chronic exposure to a red meat–rich diet also induced a functional change in the kidneys, with reduction in the fractional renal excretion rate for TMAO, despite no change in glomerular filtration rate.¹¹¹ It is also notable that numerous studies support a dose-dependent association between meat consumption and CVD risks and mortality.¹¹³

A causal contribution of gut microbiota to atherosclerosis susceptibility was first demonstrated with the discovery of TMAO as a gut microbiota–derived factor and the initial functional studies demonstrating both direct provision of TMAO accelerated atherosclerosis in murine models and that suppression of gut microbiota–dependent conversion of nutrient precursors (choline) into TMA and TMAO (with antimicrobial/antibiotics) blocked choline diet–dependent enhancement in atherosclerosis (Figures 2 and 3).⁹ In atherosclerosis-prone ApoE^−/− mice, dietary supplementation with choline led to augmented atherosclerotic lesion burden, higher aortic expression of scavenger receptors (CD36 [cluster of differentiation] and scavenger receptor A), increased cholesterol-laden macrophage foam cell formation,⁹ and impaired in vivo reverse cholesterol transport.⁷ In addition, TMAO suppressed BA pool size and, therefore, cholesterol clearance in the host.⁷ Consistent and complementary to these findings, FMO3 knockdown has been shown to impair TMA transformation into TMAO, thus reducing plasma TMAO levels and concomitantly restoring cholesterol balance.¹¹⁴ Early microbial transplantation studies of cecal microbial communities from high TMA-producing inbred C57BL/6J mice into atherosclerosis-resistant NZW/LacJ recipients demonstrated the transmissibility of dietary choline-induced TMA and TMAO generation and atherosclerosis.¹¹⁵ Not all TMAO precursor feeding studies, however, have shown similar results, supporting the notion that differences in the microbial communities present in the host can impact the final phenotypes observed.¹¹⁶

The striking association between circulating TMAO levels in subjects and thrombotic event risks, such as heart attack and stroke, has been witnessed across numerous large-scale clinical cohorts.^105,117 This has prompted mechanistic studies in both humans and mice to explore the role of TMAO in thrombosis. Zhu et al found that TMAO alters human platelet calcium signaling, heightening their responsiveness to submaximal stimulation by agonists (eg, thrombin, ADP, collagen). Consequently, heightened thrombosis potential has been observed in both whole blood and in vivo arterial injury models.^{14,83–85,118} In a subsequent human feeding study, healthy volunteers (both omnivore and vegan/vegetarian) that were orally supplemented with choline exhibited higher levels of TMAO and concomitant enhanced platelet responsiveness and aggregation.^77,118 Importantly, higher TMAO levels were dose dependently associated with increased platelet aggregation responsiveness, even in subjects on low-dose aspirin. This suggests that in subjects with elevated TMAO, the antiplatelet effects of aspirin may be attenuated, highlighting the possible involvement of TMAO in on-treatment platelet reactivity and so-called aspirin resistance.

The mechanistic involvement of the metaorganismal TMAO pathway in platelet function and in vivo thrombosis potential has also been examined through both genetic gain- and loss-of-function manipulations, including to the host gene FMO3 (Figure 2). Multiple studies have confirmed through both genetic gain- (as global FMO3 transgene) and loss-of-function (via antisense oligonucleotide to FMO3 and via global FMO3 knockout) studies in mice that manipulation of TMA and TMAO levels in vivo alters platelet responsiveness, rate of clot generation, and thrombosis potential.^83,84,119 Moreover, cecal microbial transplantation experiments confirmed that the prothrombotic phenotype mediated by a choline-rich diet was a transmissible trait.⁸³ Early studies by Cracium et al¹² first identified the cut gene cluster in human commensals encoding the catalytic and regulatory gene products CutC and CutD. The presence of the microbial cutC/D genes in human microbiota is associated with the ability to generate TMA from choline¹²⁰ and with subsequent TMAO accumulation.¹²¹ Importantly, studies using GF mice colonized with synthetic microbial communities±a genetically engineered gain- or loss-of-function cutC mutant human commensal confirmed that a functional microbial cutC gene is sufficient to transmit TMA and TMAO generation, as well as in vivo thrombosis potential. Microbial cutC may thus represent a therapeutic target for preventing thromboembolic complications.¹⁴

Recent studies have shown that beyond impacting platelet function, TMAO induces expression of TF (tissue factor)—the initiator of the extrinsic clotting—in endothelial cells in vitro.¹²² Vascular TF promotes thrombosis and vascular inflammation,¹²³ particularly in patients with type 2 diabetes mellitus who have higher levels of circulating TMAO.^124–126 Animal model studies are still needed to validate a contribution of gut microbiota and TMAO generation to alterations in TF pathway in vivo and to altered thrombosis potential. A recent study reported that the absence of microbiota was associated with reduced hepatic VWF (von Willebrand factor) synthesis and reduced thrombus growth after carotid artery injury in C57BL/6 GF mice as compared with conventionally raised littermates.¹²⁷ However, the role of metaorganismal metabolites, in particular TMAO, in the VWF-depending thrombosis has yet to be determined.

Vascular inflammation is critically involved in the pathogenesis of atherosclerosis and thrombotic complications. Seldin et al⁸⁶ found that acute infusion of physiological levels of TMAO in mice heightens vascular inflammation (Figure 2), as supported by aortic endothelial cell activation (recovered by laser microdissection), including activation of MAPK (mitogen-activated protein kinase) signaling and NF-κB (nuclear factor-kappa B) nuclear translocation, leading to subsequent proinflammatory gene expression. Complementary findings (TMAO stimulated vascular inflammation; Figure 2) have been observed in vitro using primary human aortic endothelial cells and vascular smooth muscle cells.⁸⁶ After an acute injection, mouse aortas also showed increased expression of vascular adhesion molecules, such as E selectin or ICAM (intercellular cell adhesion molecule)-1, even when TMAO had been cleared from circulation, implying sustained vascular inflammation. Further, TMAO was reported to increase oxidative stress and vascular senescence, which was characterized by impaired cell proliferation and migration in human umbilical vein endothelial cells.¹²⁸ In other animal model studies, gut microbiota suppression with oral poorly absorbed antibiotics was associated with decreased TMAO levels, improved endothelial function, reduced arterial stiffness, and lower oxidative stress.¹²⁹ Recent investigations have suggested TMAO can impact inflammation via priming and activation of the NLRP3 (NOD-, LRR-, and pyrin domain-containing protein 3) inflammasome in endothelial cells, as well as the arterial vascular wall in mice, involving mitochondrial reactive oxygen species production, TXNIP (thioredoxin-interacting protein), and lysosomal destabilization.^87,130,131 However, the exact mechanisms by which TMAO induces inflammasome activity have yet to be explored.

The receptor for TMA was identified long ago as the highly sensitive olfactory receptor TAAR5 (trace amine-associated receptor 5).¹³² TAAR5 shows high specificity for TMA and does not recognize TMAO. While TMA alone may contribute to proinflammatory signaling in the vasculature,⁸⁶ the role of TAAR5 in CVD requires further investigation. Recently, a receptor for TMAO has been reported and shown to participate in TMAO-dependent effects on glucose and insulin metabolism.¹³³ Chen et al¹³³ showed that TMAO directly binds to PERK (protein kinase R-like endoplasmic reticulum kinase), a main component of the unfolded protein response—a signaling pathway that adapts the cell to ER (endoplasmic reticulum) stress (Figure 1). Dietary supplementation of TMAO in C57BL/6J wild-type animals induced hepatic PERK expression, accompanied by increased Fox01 (forkhead box protein 01)—a key transcription factor in insulin singling and impaired glycemic control. Genetic manipulation demonstrated that the absence of hepatic PERK blunted the TMAO-induced increases in Fox01 expression and improved glycemic indices. Whether or not PERK plays a role in TMAO-mediated effects on atherosclerosis or thrombosis remains unknown and is an important area for further exploration. It is also interesting to note that ER stress is implicated in the pathogenesis of many CVD phenotypes. Whether TMAO and PERK participate in these associations also remains to be explored.

Due to its small size and combination of hydrophilic and hydrophobic properties, TMAO behaves as a chaotropic agent, with the ability to alter protein conformation and potentially serve as an allosteric modulator to proteins.¹³⁴ These features may have important physiological functions in the host (eg, impacting the protein unfolding or ER stress response within cells). In some aquatic animals, including a subset of deep sea fish, TMAO is reported to act as an osmolyte and to protect against pressure-induced protein destabilization.^135,136 Indeed, some bony fish can use large amounts of TMAO for osmoregulation, and TMAO plasma levels within the fish increase with depth of habitat, reaching levels up to 400 mmol/kg in snailfish that were caught at 7000-m depth in the South Pacific Ocean.¹³⁷ TMAO has been shown to stabilize proteins to elevated hydrostatic pressure and is thought to both impact protein conformational changes, like those that occur with allosteric regulation, and modulate intracellular molecular crowding effects.^136,138,139 Thus, TMAO accumulation both in deep sea creatures and in mammals may represent an adaptive mechanism to impact protein stability and intracellular signaling processes.

Other studies have reported that TMAO is involved in the adaptive freeze avoidance response to extreme cold—in other words, it acts as antifreeze—such as within Newfoundland smelt.¹⁴⁰ These fish have the ability to elevate plasma osmolarity by seasonal accumulation of TMAO and other organic solutes, which depresses the freezing point of body fluids and allows them to survive at subfreezing temperatures. It has recently been hypothesized that TMAO-induced protein stabilization may play a role to protect cardiomyocytes from hydrostatic pressure fluctuations during heart failure.¹⁴¹ By contrast, high concentrations of TMAO may also impair function, as was reported for the activity of the actomyosin motor.¹⁴²

TMAO may also be involved in tissue osmolality in vertebrates. When measuring concentrations in various mouse tissues, we observed that TMAO levels in kidney tissue largely exceeded those observed in corresponding paired plasma samples collected at the same time from the same animals (Hazen, unpublished data, 2016). In the kidneys, an osmotic gradient arising from the corticomedullary boundary to the inner medullary tip is normally maintained as a mechanism that allows reabsorption of water and concentration of the urine—a process also called countercurrent multiplication.¹⁴³ It seems plausible that high levels of TMAO in the kidneys may function analogously to urea, playing a role in osmoregulation and renal function. Understanding the molecular participants involved in TMAO secretion and cellular transport in the kidneys and other tissues is an important potential area of future investigation, as it may reveal novel targets for therapeutic intervention. For example, one could theorize a small-molecule drug that targets renal TMAO transport/secretion might function as a diuretic, facilitating TMAO urinary excretion and reducing both blood pressure and CVD risks.

Many studies have shown that high TMAO levels are associated with risk for thrombotic complications. However, these same properties may theoretically confer benefits to the host during situations with high bleeding risk, such as the delivery of a baby. It is thus interesting that one study reported TMAO levels may increase during pregnancy.¹⁴⁴ Studies by Bennett et al found that human liver samples from female subjects showed higher FMO1/3 expression and, therefore, an increased capacity for TMAO generation, than those from males. Likewise, gonadectomized male mice treated with testosterone show reduced hepatic FMO3 expression, while ovariectomized female mice treated with estrogen exhibit increases in hepatic FMO3 expression.⁷⁶ These results suggest that high levels of TMAO in the terminal stages of pregnancy may in theory better equip women to avoid severe blood loss during delivery. In addition, beyond FMO3 expression, microbial community alterations are known to occur during pregnancy.¹⁴⁵

While purely speculative, evolutionary drive may have selected for metabolic changes in hosts that lead to harboring of gut commensal communities with the ability to produce TMA or other gut microbiota–generated prothrombotic metabolites, thus enabling hosts to better cope with environmental stressors that lead to hemorrhage (such as parturition). However, these features may have become detrimental for individuals living in modern Western societies associated with a CVD-prone environment and less need to survive traumatic injuries. Rodents show a significant sexual dimorphism with respect to FMO3 expression, and TMAO levels, with females showing higher levels (and greater atherosclerosis capacity^9,76). Although gene expression studies in humans have suggested gender differences in hepatic FMO3 expression, plasma levels of TMAO thus far reported have failed to show sex differences, and the prognostic value of TMAO appears to be similar in both men and women.¹⁰⁵ However, it should also be noted that virtually all clinical TMAO studies reported involve cohorts that are of postmenopausal age. Future investigation should include a full exploration of sex-specific differences in TMAO levels in between younger men and women.

The pathogenesis of type 2 diabetes mellitus goes beyond glycemic control, and traditional risk factors, including level of glucose control, poorly stratify CVD risk among diabetics. To investigate this, Nemet et al¹⁴⁶ used untargeted metabolomics to identify novel metabolites that associate with incident risk for major adverse cardiac event, are increased in type 2 diabetes mellitus subjects, and do not significantly correlate with glycemic control. The candidate analyte showing the strongest association with major adverse cardiac events (m/z 265.1188) was subsequently identified as phenylacetylglutamine (PAG)—a phenylalanine-derived metabolite (Figure 4). The association of PAGln with incident risk for major adverse cardiac events like heart attack, stroke, and death was further validated in an independent and nonoverlapping cohort comprised of 4000 stable cardiovascular subjects and shown to be independent of traditional CVD risk factors in both diabetics and nondiabetics alike.¹⁴⁶ Additional functional studies revealed that the association of PAG with incident CVD risks likely occurs because the metabolite impacts host physiology and fosters CVD-related phenotypes. Moreover, PAG was shown to be generated via gut microbes during metabolism of phenylalanine, as illustrated in Figure 4. Genetic engineering studies in microbes coupled with transplantation into GF mice confirmed gut microbial (and some of the gut microbial genes) contribution to host platelet reactivity and in vivo thrombosis potential. And through multiple gain-of-function and loss-of-function genetic and pharmacological studies, PAG was shown to interact with GPCRs, including the α- and β-ARs (adrenergic receptors).¹⁴⁶ ARs are crucially involved in heart disease¹⁴⁷ and platelet function.¹⁴⁸ However, until the discovery of PAG, AR signaling had not yet been implicated in gut microbiota-derived factors driving CVD. The new studies by Nemet et al¹⁴⁶ also showed that adverse CVD-related phenotypes observed with PAG administration at physiological levels were attenuated with the β-blocker carvedilol.

It is striking that gut microbiota appear to elicit adverse cardiovascular phenotypes in the host via modulation of ARs. Such a finding may help explain some of the beneficial effects of clinical β-blocker treatment. Interestingly, ARs may also in turn modulate microbial abundance as observed in β-AR knockout mice.¹⁴⁹ The novel discovery of a microbiota-AR signaling axis is particularly interesting considering the widespread implication of ARs in cardiovascular physiology and metabolism. There is no reason to assume that the subset of ARs identified in this study are the only receptors modulated by PAG. It remains to be determined whether other members of the AR gene family, some of which are also known to participate in the regulation of cardiovascular homeostasis are similarly modulated by PAG. It is intriguing to speculate that some phenotypes exhibited by microbiota are directly or indirectly mediated via ARs. This is indirectly supported by the fact that the absence of microbiota alters cardiovascular homeostasis, such as blood pressure regulation,¹⁵⁰ myocardial repair following postinfarction cardiac repair,⁶⁶ or thrombosis growth¹²⁷—functions that all involve ARs. More studies are needed to further characterize PAG-mediated functions in the host. PAG also appears to represent another potential gut microbiome pharmacological target for future efforts.¹⁵¹

In response to accumulating evidence that the microbiota affects susceptibility for CVD and cardiometabolic diseases, researchers have begun to develop microbiota-directed interventions to improve clinical outcomes (Figure 3). The microbiome-host axis comprises many different layers, including dietary precursors, microbial communities, and metaorganismal pathways that generate bioactive metabolites recognized by host receptors—all of which represent potential therapeutic targets to modulate community output and host phenotype. Figure 3 illustrates several different therapeutic approaches for targeting the gut microbiome to exert a beneficial effect in the host. We have already mentioned the use of antimicrobial agents such as poorly absorbed antibiotics as a valuable tool for demonstrating involvement of gut microbiota in both animal models and humans⁹ but as a poor choice as a long-term therapeutic due to the development of antibiotic resistance. While many studies have reported associations between atherosclerotic plaque and the presence of pathogens such as Cytomegalovirus, Chlamydia, and Helicobactor pylori,^152–155 multiple prospective randomized trials with antibiotics have thus far failed to demonstrate clinical benefit.^156–158 Further, the impact of antibiotic treatment on microbial communities is hard to predict, often because the microbial community that recolonizes after cessation of antibiotics can be variable, depending upon many factors, including the microbes one is exposed to as the antibiotics are metabolized and excreted. Currently, there is no clear evidence that antibiotics have efficacy in the treatment of CVD in humans. Use of antibiotics thus seems better suited to eradicating true pathogens than as a chronic long-term preventive intervention.

In a recent study that screened over 1000 commonly prescribed nonantibiotic drugs for their impact on a broad selection of human gut commensals, nearly a quarter of the tested medications demonstrated antibiotic-like activity, significantly inhibiting the growth of at least 1 human commensal.¹⁵⁹ These results suggest that commonly prescribed medications may impact human gut microbial communities, thereby impacting host phenotypes through indirect effects mediated by changes in the gut microbiome. One well-studied example of this is the widely used antidiabetic drug metformin. Recent studies by Bäckhed et al¹⁶⁰ show that some of the antidiabetic effects of metformin therapy appear to be mediated by alterations in gut microbiota composition, since microbial transplantation studies reveal transmission of antidiabetic effects in recipient GF mice following fecal transfer from metformin treated donors. In addition to the potential for common medications to impact gut microbial composition and function, thus potentially contributing to drug effects in the host, it is becoming increasingly clear that there is broad diversity in gut microbial metabolism of medications, leading to potential altered responsiveness to drugs. The diverse drug-microbiome interactions that vary between individuals is another area where further studies are needed, and shows promise for possible use in drug development efforts and personalized medicine.¹⁶¹

Defined microbial compositions (probiotics) and nonmicrobial substances that may alter microbial community structure (prebiotics) have also been proposed to improve CVD (Figure 3). Indeed, there are many preclinical and some clinical intervention studies using either prebiotics or probiotics that have shown promising results. However, these studies tend to be relatively small in size, and their adoption into clinical practice would require further study. The use of probiotics and prebiotics has recently been extensively reviewed, including discussions of the many promises and potential challenges of their development and use.^162–167 While a complete review of this topic is beyond the scope of the present review, below we discuss several preclinical studies involving probiotics, particularly where a mechanistic role for alteration in gut microbiome related processes to improved host phenotype were reported. One example includes a recent rodent hypertension study in which either a high-fiber diet (a prebiotic) or acetic acid alone (a gut microbial product) resulted in reduced blood pressure and adverse cardiac remodeling.¹⁶⁸ In a different mouse model involving ApoE^−/− mice on a n-3 polyunsaturated fatty acid–depleted diet for 12 weeks, supplementation of dietary inulin-type fructans was shown to reverse endothelial dysfunction in carotid arteries via activation of the NO synthase/NO pathway.¹⁶⁹ Beyond changes in vascular function, probiotic use in animal studies of heart failure has also shown promise. For example, administration of Lactobacillus rhamnosus GR-1 in rodents improved systolic and diastolic left ventricular function following coronary artery ligation.¹⁷⁰ This study was of interest because it was shown that while the GR-1 strain did not colonize to distal intestines where most anaerobes reside, beneficial cardiac remodeling was none-the-less observed. In another recent study, provision of A muciniphila was associated with reduced aortic atherosclerosis in the hypercholesterolemic ApoE^−/− mouse model.²⁸ This probiotic is of interest because of its use in human clinical investigations. For example, provision of A muciniphila in a randomized placebo controlled double-blinded interventional exploratory study was reported to promote a reduction in plasma LPS in obese individuals with metabolic syndrome.²⁹

While prebiotic and probiotic interventions have shown ability to favorably alter metabolic profiles in some human studies, results are highly variable, and animal model findings have not yet been translated to evidence of clinical efficacy. It is still unclear whether most of the used microbes survive the acidity of the stomach, if they colonize the colon where most of the gut microbiota reside, and whether the beneficial effects are mediated by the ingested microbes, caused by shifts in the community structure or possibly even caused by secondary effects on host immune education and function. One difficulty with probiotic studies where secondary effects on microbial communities likely play a role in beneficial effects if observed is the vastness of the gut microbial community and the interindividual differences in community structure. These factors are thought to give rise to variable responses to probiotics or prebiotics in general.¹⁷¹ Thus, results of probiotic or prebiotic administration have been difficult to predict. Moreover, it is worth noting that the current selection of probiotics seems to be primarily driven by abundance-based analyses of microbiota composition, wherein microbial community members whose proportions are highly associated with beneficial phenotypes are the focus of interest. However, the keystone commensal organizing a community architecture, or providing a key gain of function as discussed above, can be a low abundance component and is often not easily detected by current conventional sequencing depths of analyses.

Perhaps the most obvious potential therapeutic intervention for targeting the gut microbiome is diet (Figure 3). The TMAO pathway is an excellent example of this, given the nutrient precursors are more abundant in a Western diet, and diets rich in phosphatidylcholine and carnitine are associated with heightened levels of TMAO, whereas vegetarian or vegan diets have reduced nutrient precursors.^111,118 Notably, however, TMAO levels appear to be driven more so by the gut microbial community composition than by the dietary intervention, and significant variation in TMAO production among individuals on a given diet is observed.^14,111 A low choline- or carnitine-containing diet can be generated using a primarily vegetarian or vegan selection of food, and such diets are rational recommendations for subjects with high TMAO levels. But such dietary recommendations are harder to envision with other gut microbiota-generated metabolites, where either the nutrient precursors are numerous (eg, SCFA) or are essential amino acids (eg, phenylalanine and PAG), and so cannot be easily avoided. However, even with phenylalanine, there are dietary choices that can be made to reduce intake. For example, those who experience the inborn error of metabolism called phenylketonuria have a host enzyme deficiency that makes eating foods abundant in phenylalanine harmful (ie, proteins). Consequently, throughout life, a phenylketonuric diet is highly recommended, which has an overall low phenylalanine content.¹⁷² The impact of adopting a phenylketonuria diet in subjects with high PAG levels (eg, subjects with diabetes mellitus, renal disease) has yet to be examined but is of considerable interest.

Recent work has examined the selective nonlethal targeting of gut microbial enzymes for TMA generation as a therapeutic approach for the treatment or prevention of CVD (Figure 3). Key to this approach is the development of a small-molecule inhibitor that is nonlethal to the microbe and thus does not trigger as great a selective pressure as an antibiotic for development of resistance. The first study of this type targeting the gut microbiome for the treatment of CVD used a choline structural analogue, 1,3 dimethylbutanol. Through a series of studies, 1,3 dimethylbutanol was shown to serve as a nonlethal microbial enzyme inhibitor of choline→TMA transformation and to reduce TMAO production in vivo without affecting microbial fitness.⁷⁸ When fed to animals, 1,3 dimethylbutanol inhibited choline diet–dependent TMAO generation, reduced macrophage foam cell formation, and inhibited aortic atherosclerotic plaque development.⁷⁸ Next-generation choline TMA lyase inhibitors have since been developed that selectively target and accumulate in microbiota, thereby limiting systemic exposure in the host. The choline TMA lyase suicide substrate inhibitor fluoromethylcholine was shown to be over 10 000-fold more potent an inhibitor of cutC than 1,3 dimethylbutanol and markedly blocks microbial choline catabolism.⁸⁵ Interestingly, owing to their highly polar nature, fluoromethylcholine and related halomethylcholines were shown to be poorly absorbed into the host, limiting systemic exposure and thus chances of side effects. In addition, cutC inhibition by fluoromethylcholine and iodomethylcholine was shown to result in microbial cytosolic choline elevation, which appears to be sensed as an abundant fuel source by the microbe. This then induces expression of the entire cut gene cluster, including both cutC and the microbial choline transporter. This leads to active microbial uptake of fluoromethylcholine and the substrate of cutC, choline. A positive feedback loop is thus created, whereby the more microbial catabolism of choline into TMA is inhibited, the greater the elevation in cytosolic choline within the gut microbe (sensed as an abundant fuel source), and the greater the sequestering of choline from the intestinal lumen into the inhibited gut microbe. As intestinal luminal choline is depleted, microbial community TMA production is globally inhibited, even from neighboring community members who might otherwise not be potently inhibited.⁸⁵

Although human clinical studies with choline TMA lyase inhibitors have not yet been reported, numerous efforts are ongoing in this area. Microbiota-targeting nonlethal small-molecule inhibitors will require the same sort of safety testing as any other drug. But with compounds that have reduced systemic absorption, there is a theoretical potential benefit of limiting adverse side effects from off-target inhibitory activities in the host. Thus, pharmacological interventions aimed at drugging the microbiome with nonlethal small-molecule inhibitors represents a novel therapeutic approach for both the treatment and prevention of cardiometabolic diseases that will need to be validated in clinical intervention studies.

Although our knowledge about how microbiota impact CVD is still rudimentary, the rate at which new discoveries are emerging is impressive. As outlined, there is overwhelming evidence that gut microbiota-derived processes in general are linked to numerous CVD-relevant phenotypes, including but not limited to atherosclerosis, platelet reactivity and thrombosis potential, blood pressure, lipid metabolism, adiposity, glucose homeostasis, and vascular inflammation. Investigative approaches have included a wide array of microbiota transplantation studies, both animal and human dietary interventions in colonized versus antibiotic suppressed states (or GF mice), direct provision (dietary or via infusion) of specific gut microbiota metabolites, and both genetic and pharmacological studies that have targeted multiple components of metaorganismal pathways (including gut microbial genes and both host transformational enzymes and end-organ receptors). New therapeutic approaches that target gut microbes for the treatment and prevention of CVDs represent exciting areas of investigation. The development of nonlethal microbial inhibitors that target specific pathways, yet show limited systemic exposure in the host, are just one of the new and potentially promising therapeutic approaches. Yet others include but are not limited to dietary interventions, probiotics, or prebiotics that hopefully can be used to someday terraform the microbial community to alter its functional output to the betterment of the host. As with any therapeutic, large, prospective interventional studies will be needed to validate novel gut microbiome-targeted therapeutics.

This work is supported by grants from the NIH and Office of Dietary Supplements (P01HL147823 and R01HL103866), the German Research Foundation (WI 5229/1-1), and an award from the Leducq Foundation. S.L. Hazen reports being partially supported by a gift from the Krieger Fund. M. Witkowski, T.L. Weeks, and S.L. Hazen all contributed to the drafting of the manuscript and critical revisions to the manuscript.