Trimethylamine N-Oxide and Related Gut Microbe-Derived Metabolites and Incident Heart Failure Development in Community Based Populations

Abstract

Background:

Growing evidence indicates that trimethylamine N-oxide (TMAO), a gut microbial metabolite of dietary choline and carnitine, promotes both cardiovascular disease (CVD) and chronic kidney disease risk. It remains unclear how circulating concentrations of TMAO and its related dietary and gut microbe-derived metabolites (choline, betaine, carnitine, γ-butyrobetaine, and crotonobetaine) affect incident heart failure (HF).

Methods:

We evaluated 11,768 participants from the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis with serial measures of metabolites. Cox proportional hazards models were used to examine the associations between metabolites and incident HF, adjusted for CVD risk factors.

Results:

2,102 cases of HF occurred over a median follow-up of 15.9 years. After adjusting for traditional risk factors, higher concentrations of TMAO (HR 1.15, 95%CI 1.09,1.20, p<0.001), choline (HR 1.44, 95%CI 1.26,1.64, p<0.001), and crotonobetaine (HR 1.24, 95%CI 1.16,1.32, p<0.001) were associated with increased risk for incident HF. After further adjustment for renal function (potential confounder or mediator), these associations did not reach Bonferroni- corrected statistical significance (p=0.01, 0.049, 0.006, respectively). Betaine and carnitine were nominally associated with higher incidence of HF (p<0.05). In exploratory analyses, results were similar for subtypes of HF based on left ventricular ejection fraction, and associations appeared generally stronger among Black and Hispanic/Latino vs White adults even though there were no interactions for any metabolites with race.

Conclusions:

In this pooled analysis of two well-phenotyped, diverse, community-based cohorts, circulating concentrations of gut microbe-derived metabolites such as TMAO, choline, and crotonobetaine were independently associated with higher risk of developing HF.

Trial Registration:

CHS (ClinicalTrials.gov ID: NCT00005133); MESA (ClinicalTrials.gov ID: NCT00005487)

INTRODUCTION

Heart failure (HF) is one of the leading causes of mortality worldwide, and represents downstream disease progression from several major cardiovascular diseases (CVD).¹ By identifying novel risk factors for HF, we are able to uncover new physiological risk pathways and develop corresponding preventive strategies.² In particular, research interest has focused on identifying targetable mechanistic pathways that allow for early pharmacologic treatments or dietary and lifestyle modifications.^3,4

Trimethylamine N-oxide (TMAO) is a gut microbiota-generated metabolite derived from dietary precursors like L-carnitine (primarily from red meat) and choline (primarily phosphatidylcholine, also called lecithin, abundant in egg yolks, meat, and other animal products), and to a lesser extent betaine (primarily from shellfish, wheat germ or bran, and spinach). Carnitine is converted into TMAO via a multi-step process including gut microbe generation of the carnitine-related compounds crotonobetaine and γ-butyrobetaine.^5–7 A number of other processes also contribute to carnitine homeostasis, including dietary intake and absorption, efficient reabsorption, and mechanisms that establish and maintain substantial concentration gradients between intracellular and extracellular carnitine pools.^8,9 In the large intestine, specific gut microbes consume dietary choline and carnitine, and directly or indirectly generate trimethylamine (TMA), which is then absorbed by the host and metabolized into TMAO in the liver by hepatic flavin monooxygenases.^10,11 Clinical studies among patients with prevalent HF have observed associations between circulating TMAO concentrations and severity of both acute decompensated HF and chronic HF.^12–19 Animal model studies have further demonstrated that a dietary choline-mediated increase in circulating TMAO is mechanistically linked to adverse cardiac remodeling and HF progression, and that this can be attenuated by pharmacologic inhibition of microbial TMA-generation enzymes or withdrawal of a high choline diet (both leading to reduced circulating TMAO concentrations).^20–22 Prospective studies have also shown that plasma TMAO is associated with higher risk of coronary artery disease,^23–25 a risk factor for HF.

However, it remains unknown how TMAO and its associated metabolites may be related to the development of incident HF. To investigate this relationship, we examined serial biomarker measurements with adjudicated longitudinal HF events in the Cardiovascular Health Study (CHS)²⁶ and the Multi-Ethnic Study of Atherosclerosis (MESA)²⁷ – two multi-center, prospective, community-based cohort studies of older adults including diverse populations.

METHODS

Data Availability.

The data underlying this article were provided by the MESA (https://www.mesa-nhlbi.org) and CHS (https://chs-nhlbi.org) under license/by permission. Data can be shared on request to the corresponding author if permitted by MESA/CHS.

Study Design and Population.

The CHS is a multi-center, community-based, prospective cohort study designed to investigate risk factors for coronary heart disease and stroke in older adults.²⁸ This cohort consists of 5,201 community-dwelling individuals aged 65 years or older at enrollment, who were recruited from 4 communities (Sacramento, CA; Hagerstown, MD; Winston-Salem, NC; Pittsburgh, PA) in 1989–1990, plus an additional 687 predominantly Black participants who were recruited in 1992–1993. In the MESA study, 6,814 adults of diverse racial/ethnic backgrounds aged 45–84 years free of clinical CVD were recruited between 2000–2002 from 6 study sites (Baltimore County, MD; Chicago, IL; Forsyth County, NC; New York, NY; Los Angeles County, CA; St Paul, MN).²⁹ In both cohorts, race was classified by self-identification and was assessed as a social construct to investigate potential differences in associations of risk factors with disease outcomes. In CHS, TMAO was measured in 5,418 participants – of those, we excluded individuals with prevalent HF at baseline (n=314), or with recent antibiotic use (within 2 weeks preceding blood collection) (n=49), leaving 5,055 participants. In MESA, TMAO was measured in 6814 participants – of those, we excluded individuals with prevalent HF at baseline (n=6), recent antibiotic use (n=67), or loss of follow-up (n=28), leaving 6,713 participants. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting methods, findings, and study limitations for cohort studies. The investigational review boards of the respective clinical sites and the Data Coordinating Centers have approved the CHS and MESA data collection procedures, and all participants provided informed written consent.

Data Collection.

A detailed description of data collection procedures has been previously described and data from the two cohorts were harmonized.^28,29 Information on baseline clinical characteristics and lifestyle factors were collected at study clinic visits to assess participants’ sociodemographic characteristics, lifestyle, medical history, other health-related phenotypes, and clinical events. Medication use was assessed by a validated medication inventory. In both studies, plasma concentrations of glucose, insulin, lipids, and inflammatory biomarkers were assessed using enzymatic methods. Body mass index (BMI) was calculated from directly-measured body weight in kilograms divided by height in meters squared. Estimated glomerular filtration rate (eGFR) was calculated with an equation from the Chronic Kidney Disease Epidemiology Collaboration using serum creatinine and cystatin C concentrations.³⁰ Chronic kidney disease (CKD) was defined as eGFR <60 ml/min/1.73m². Physical activity (kilocalories per week) was assessed using a modified Minnesota Leisure-Time Activities questionnaire in CHS³¹, and the Typical Week Physical Activity Survey in MESA.³² We leveraged Alternative Healthy Eating Index (AHEI) to estimate dietary quality data.³³

Mass Spectrometry Quantification of TMAO and gut microbial-associated metabolites.

To assess long-term (usual) exposure, TMAO and related metabolites were measured in serial samples over time in both cohorts. Plasma was collected in ethylenediaminetetraacetic acid (EDTA)-plasma tubes, processed, and stored in central repositories at −80°C. In CHS, metabolites were measured in plasma samples collected at baseline (1989–1990 or 1992–1993) and again in samples collected in 1996–1997, with 5,418 participants having at least one measure and 2,995 having both. In MESA, metabolites were measured in plasma samples collected at baseline (2000–2002) and again in 2005–2007, with 6,814 participants having at least one measure and 5,624 having both. Each metabolite and its heavy isotope labeled isotopologues (internal standard) were quantified using a stable-isotope dilution assay coupled with high-performance liquid chromatography, with online electrospray ionization tandem mass spectrometry on a Shimadzu 8050 mass spectrometer as previously described.^5,34,35 All laboratory measurements were performed at the Cleveland Clinic Lerner Research Institute, with laboratory coefficients of variation <10% for each metabolite with excellent stability data for archived specimens over long periods of storage time.³⁴ Investigators performing mass spectrometry analyses were blinded to patient clinical data.

Ascertainment of Heart Failure.

The primary outcome was incident HF. Detailed methods for assessment and confirmation of HF in these cohorts have been previously described. ^27–29,36 Briefly, potential HF events were identified during regular study examinations and interim telephone interviews. The incidence of HF was then determined by centralized CHS and MESA Cardiovascular Events Committees, based on diagnosis from a physician as well as review of medical records, including consideration of symptoms, signs, chest X-ray findings, other cardiac imaging, and treatment of HF.^27–29,36 Leveraging the availability of endpoint adjudication for incident HF cases in both cohorts, we further investigated the role of plasma concentrations of TMAO and other associated metabolites in the development of incident HF with preserved ejection fraction (HFpEF, defined as LVEF ≥50%) or incident HF with reduced ejection fraction (HFrEF, defined as LVEF <50%).² Follow-up for adjudicated events for MESA occurred through 2018 and for CHS occurred through 2015.

Statistical Analysis.

Cox proportional-hazards models estimated the multivariable-adjusted association of concentrations of each metabolite with incident HF. Participants were followed from baseline (time of earliest available sample) to their first diagnosis of HF, death, or loss of follow-up. To assess long-term exposure to each metabolite, we used time-varying cumulative averages, with the measure at baseline related to incident HF until the time of the second measure, and the average of the measure at baseline and the second measure related to incident HF thereafter. With its skewed distribution, TMAO and other metabolites was log (base-2) transformed, so that hazard ratios (HRs) for one unit increases represent the difference in risk per each doubling of TMAO and other metabolites. Our primary model adjusted for the following variables: age (years), sex (male/female), race (Black vs. other), enrollment site (4–6 sites per study), education (<high school, high school graduate, some college, college graduate), self-reported annual income (<12k, $12k-24.9k, $25k-49.9k, >$50k), BMI (kg/m²), smoking status (never/former/current), physical activity (total kcals per week), systolic blood pressure (beats per minute), treated hypertension (yes/no), low density lipoprotein cholesterol (mg/dL), fasting glucose (mg/dL), prevalent diabetes (yes/no), atrial fibrillation (yes/no), and coronary heart disease (yes/no). A second model further adjusted for eGFR, which could be either a mediator or confounder of the TMAO-HF association given that TMAO (as well as most of the metabolites) is cleared by the kidneys (confounder), and experimental models demonstrate that TMAO damages the kidney and reduces eGFR (mediator, i.e., in the potential causal pathway).³⁷ Adjustment covariates were updated at the time of the second metabolite measurement. We evaluated cubic splines to assess linearity, and scaled Schoenfeld residuals were used to evaluate potential violation of the proportional hazards assumption.

To explore whether associations of TMAO and other metabolites with incident HF were modified by covariables such as age, sex, BMI, or eGFR, we evaluated an additional model that included each variable x metabolite multiplicative term, and analyses were stratified by race/ethnicity and by prevalent CKD (eGFR <60 ml/min/1.73m²). Since CVD could be both a confounder and a mediator of associations, we also performed sensitivity analyses excluding individuals with prevalent CVD at baseline (and without serial updating of metabolite concentrations for individuals who developed incident CVD prior to the second measurement). Single imputation was used to impute missing values of adjustment covariates, using information on age, sex, race, study site, education, income, self-reported health status, smoking, alcohol intake, physical activity, BMI, waist circumference, treated hypertension, insulin, oral hypoglycemic agents, C-reactive protein, and eGFR as previously described.³⁸ To correct for multiple comparisons, the significance level of main-effects tests was set as 2-tailed α = 0.0083 (0.05/6 metabolites); and the significance level of interaction tests was set at α = 0.0021 (0.05/24 [6 metabolites × 4 effect modifiers]). Analyses were performed using STATA, version 16.0 (StataCorp LLC).

RESULTS

Figure 1 depicts the flow chart of participants who were included from CHS and MESA cohorts in this analysis, as well as the pooled analysis. Table 1 shows the baseline characteristics of participants in this pooled analysis of CHS (n = 5,055) and MESA (n = 6,713) with measurements of metabolites (n = 11,768, see respective measurement ranges and sample size distributions of each metabolite in Table S1). Overall, participants with elevated concentration of TMAO and related metabolites were older, more likely to be male, and had more comorbid conditions including cardiovascular diseases, and renal insufficiency. In contrast, there were no consistent relationships with other metabolic markers, such as glucose, LDL, or BMI.

Figure 1. STROBE Diagram for Patient Characteristics.

Abbreviations: CHS = Cardiovascular Health Study; MESA = Multiethnic Study of Atherosclerosis; TMAO = trimethylamine N-oxide; HF = heart failure.

Table 1.

Baseline Characteristics

	CHS	MESA	Pooled
	(n=5,055)	(n=6,713)	(n=11,768)
Age (years)	73 ± 6	62 ± 10	67 ± 10
Male	40%	47%	44%
Race  White	84%	39%	58%
Black	16%	28%	22%
Chinese	0%	12%	7%
Hispanic/Latino	0%	22%	13%
Other	1%	0%	0%
Diabetes mellitus	15%	13%	14%
Coronary heart disease	18%	0%	8%
Atrial fibrillation	3%	1%	2%
Systolic blood pressure (mmHg)	136 ± 21	127 ± 21	131 ± 22
Treated hypertension	47%	37%	41%
Smoking (past/current)	42%/12%	37%/13%	39%/13%
Body mass index (kg/m2)	27 ± 5	28 ± 5	28 ± 5
Physical Activity (total Kcals/week)	1,163 ± 1566	1,559 ± 2362	1389 ± 2067
AHEI score	39.0 ± 12.1	41.0 ± 11.6	40.2 ± 11.8
Glucose (mg/dL)	110 ± 36	97 ± 30	103 ± 33
LDL cholesterol (mg/dL)	131 ± 35	117 ± 32	123 ± 34
eGFR (ml/min/1.73m2)	69 ± 17	90 ± 18	81 ± 21
TMAO (μM)	4.8 [3.2, 7.8]	3.6 [2.4, 5.6]	4.1 [2.7, 6.5]
Choline (μM)	9.5 [81, 11.2]	9.2 [7.8, 10.8]	9.3 [7.9, 11.0]
Betaine (μM)	36.1 [28.8, 44.3]	35.6 [28.2, 43.5]	35.8 [28.4, 43.8]
Carnitine (μM)	36.7 [31.7, 42.2]	35.6 [30.7, 40.7]	36.1 [31.2, 41.4]
γ−Butyrobetaine (μM)	0.99 [0.81, 1.21]	0.88 [0.72, 1.06]	0.92 [0.76, 1.12]
Crotonobetaine (μM)	0.023 [0.010, 0.030]	0.023 [0.018, 0.030]	0.030 [0.017, 0.030]

Values presented as mean ± standard deviation, median [interquartile ranges] or percentages. Abbreviations: HF = heart failure; CHS = Cardiovascular Health Study; MESA = Multiethnic Study of Atherosclerosis; AHEI = Alternative Healthy Eating Index; LDL = low-density lipoprotein; eGFR = estimated glomerular filtration rate; TMAO = trimethylamine N-oxide.

During a median follow up of 12.6 years (maximum: 25.0 years) in CHS and 16.7 years in MESA (maximum: 18.5 years), a total of 1,693 and 409 cases of incident HF, respectively, were identified. Among the 2,102 total incident HF cases, 1,281 occurred among people without prevalent CKD at baseline, and 821 among people with prevalent CKD. Among the 1,343 cases with sufficient information to determine left ventricular ejection fraction, 777 (57.9%) were adjudicated to have HFpEF (640 in CHS, 137 in MESA), while 566 (42.1%) were adjudicated to have HFrEF (433 in CHS, 133 in MESA).

After multivariable-adjustment and Bonferroni correction for 6 metabolites (2-tailed α = 0.0083), higher plasma concentrations of TMAO, choline, and crotonobetaine were each found to be significantly associated with increased risk of incident HF (Table 2). For each doubling of concentrations, TMAO was associated with 15% higher risk, choline with 44% higher risk, and crotonobetaine with 24% higher risk. Betaine and carnitine were nominally associated with higher incidence of HF (p<0.05) but did not remain significant following Bonferroni correction. Findings were similar when using restricted cubic splines, and we noted that there was evidence that the associations of choline and crotonobetaine were non-linear (Figure 2).

Table 2.

Association of serial measures of gut microbial metabolite concentrations and incident HF, based on pooled analyses among N=11,768 adults in the CHS and MESA cohorts

	Adjusted for age, sex, race, site			Multivariable adjusted			Multivariable adjusted w/ eGFR (Potential mediator model)
All HF (2,102 cases)
	HR	95% CI	p	HR	95% CI	p	HR	95% CI	p
TMAO	1.20	(1.15, 1.26)	<0.001	1.15	(1.09, 1.20)	<0.001	1.07	(1.02, 1.12)	0.01
Choline	1.71	(1.50, 1.95)	<0.001	1.44	(1.26, 1.64)	<0.001	1.15	(1.00, 1.33)	0.049
Betaine	1.06	(0.96, 1.16)	0.27	1.14	(1.03, 1.26)	0.010	1.11	(1.00, 1.22)	0.044
Carnitine	1.24	(1.06, 1.45)	<0.001	1.20	(1.02, 1.40)	0.023	1.07	(0.92, 1.25)	0.34
γ−Butyrobetaine	1.09	(0.98, 1.23)	0.12	1.12	(1.00, 1.26)	0.052	0.90	(0.81, 0.99)	0.032
Crotonobetaine	1.33	(1.24, 1.42)	<0.001	1.24	(1.16, 1.32)	<0.01	1.10	(1.03, 1.18)	0.006
HFpEF (777 cases)
	HR	95% CI	p	HR	95% CI	p	HR	95% CI	p
TMAO	1.19	(1.11, 1.28)	<0.001	1.14	(1.05, 1.23)	0.001	1.09	(1.01, 1.18)	0.035
Choline	1.63	(1.32, 2.03)	<0.001	1.34	(1.08, 1.67)	0.008	1.19	(0.95, 1.49)	0.14
Betaine	1.05	(0.89, 2.03)	0.54	1.13	(0.96, 1.33)	0.14	1.12	(0.95, 1.31)	0.19
Carnitine	1.20	(0.94, 1.54)	0.15	1.07	(0.84, 1.36)	0.58	1.01	(0.79, 1.28)	0.96
γ−Butyrobetaine	0.91	(0.76, 1.10)	0.33	0.90	(0.75, 1.08)	0.28	0.78	(0.67, 0.92)	0.002
Crotonobetaine	1.25	(1.13, 1.39)	<0.001	1.17	(1.05, 1.30)	<0.003	1.10	(0.98, 1.23)	0.10
HFrEF (566 cases)
	HR	95% CI	p	HR	95% CI	p	HR	95% CI	p
TMAO	1.19	(1.09, 1.30)	<0.001	1.14	(1.05, 1.25)	0.003	1.06	(0.97, 1.17)	0.21
Choline	1.60	(1.23, 2.06)	<0.001	1.40	(1.08, 1.83)	0.012	1.12	(0.85, 1.48)	0.43
Betaine	1.03	(0.86, 1.24)	0.72	1.11	(0.92, 1.33)	0.27	1.09	(0.90, 1.31)	0.38
Carnitine	0.97	(0.71, 1.33)	0.85	1.00	(0.73, 1.37)	0.98	0.90	(0.66, 1.23)	0.51
γ−Butyrobetaine	1.11	(0.90, 1.36)	0.34	1.15	(0.94, 1.42)	0.18	0.93	(0.78, 1.12)	0.46
Crotonobetaine	1.30	(1.14, 1.47)	<0.001	1.23	(1.08, 1.40)	0.002	1.09	(0.95, 1.25)	0.22

Hazard ratios (HRs) and 95% confidence intervals (95% CIs) are per doubling of metabolite concentrations (log base-2), with serial measures and updating of adjustment covariates. Multivariable model includes age (years), sex (male/female), race (Black vs. other), enrollment site (4–6 sites per study), education (<high school, high school graduate, some college, college graduate) , self-reported annual income (<12k, $12k-24.9k, $25k-49.9k, >$50k), body mass index (kg/m²), smoking status (never/former/current), physical activity (total kcals per week), systolic blood pressure (mmHg), treated hypertension (yes/no), low density lipoprotein cholesterol (mg/dL), fasting glucose (mg/dL), and prevalent diabetes (yes/no), atrial fibrillation (yes/no), or coronary heart disease (yes/no). Abbreviations: TMAO = trimethylamine N-oxide; HF = heart failure; CHS = Cardiovascular Health Study; MESA = Multiethnic Study of Atherosclerosis; HFpEF = heart failure with preserved ejection fraction; HFrEF = heart failure with reduced ejection fraction.

Figure 2. Restricted cubic spline plots of gut microbial metabolites with incident HF Risk.

Cubic spline plots showing associations between incident HF risk and A) TMAO, B) choline, C) betaine, D) carnitine, E) γ-butyrobetaine, and F) crotonobetaine. The x-axis represents the metabolite concentrations expressed in log scale. The y-axis represents the HR relative to the median metabolite value (reference level). The curves were truncated symmetrically at 2.5th and 97.5th percentiles to represent the middle 95% of the data. The shading represents the 95% confidence interval, and data ranged from 1–97.5th percentiles The dotted lines on each plot represent the thresholds for the 5^th, 25^th, 50^th, 75^th, and 95^th percentile values. All associations were adjusted for age (years), sex (male/female), race (Black vs. other), enrollment site (4–6 sites per study), education (<high school, high school graduate, some college, college graduate) , self-reported annual income (<12k, $12k-24.9k, $25k-49.9k, >$50k), body mass index (kg/m2), smoking status (never/former/current), physical activity (total kcals per week), systolic blood pressure (mmHg), treated hypertension (yes/no), low-density lipoprotein cholesterol (mg/dL), fasting glucose (mg/dL), and prevalent diabetes (yes/no), atrial fibrillation (yes/no), or coronary heart disease (yes/no). P values for nonlinearity were >0.05 for all metabolites. Abbreviations: TMAO = trimethylamine N-oxide; HR = hazard ratio.

When we assessed TMAO and related metabolites and incidence of HFpEF or HFrEF, confidence intervals were wider due to smaller numbers of cases, but results were generally similar for each subtype of HF. For each doubling of concentrations, TMAO was associated with 14% higher risk of HFpEF or HFrEF; crotonobetaine, with 17% higher risk of HFpEF and 23% higher risk of HFrEF – all remained statistically significant following Bonferroni correction (Table 2). The other four metabolites were not significantly associated with these HF subtypes following Bonferroni correction. Exclusion of individuals with prevalent CVD at baseline had little effect on the main findings, slightly strengthening the associations for both HFpEF and HFrEF (data not shown).

With further adjustment for eGFR, which could be either a confounder or mediator (i.e., in the causal pathway) of the TMAO-HF relationship, the association between plasma TMAO and incident HF was no longer statistically significant with Bonferroni correction (HR 1.07, 95% CI 1.02–1.12, p=0.01, Table 2). A similar attenuation was observed with crotonobetaine (HR 1.10, 95% CI 1.03–1.18, p=0.006, Table 2).

In analyses exploring potential heterogeneity according to other participant characteristics, the relationships between all metabolites and incident HF were generally not significantly different according to age, sex, BMI, eGFR, except in the case of choline and crotonobetaine, which differed with eGFR (Table S2). When stratified according to race, associations of metabolites with incident HF appeared stronger in Black and Hispanic/Latino participants than in White participants (Table S3), but when tested for interaction of Black vs. White participants we did not find evidence of an interaction for any of the metabolites (Table S2). In addition, when we adjust for AHEI scores in Model 2, our findings are largely unchanged compared with our main analysis (Table S4). For sensitivity analysis in individuals with two measures, only the most recent value was used as the exposure, results were similar to those that update the 2nd measurement with the cumulative average (Table S5).

DISCUSSION

While elevated TMAO and related metabolites have been associated with HF, no prior study has investigated their associations with the development of incident HF. In this pooled analysis of two large, community-based prospective cohorts derived from diverse U.S. adults without HF at baseline, we investigated how serial plasma concentrations of TMAO and related metabolites were associated with incidence of HF. We observed significant associations of elevated concentrations of three specific gut microbiota related metabolites - TMAO, crotonobetaine, and choline - with increased risk of incident HF, which remained significant after adjustment for a range of CVD, sociodemographic, lifestyle, medical, and biochemical risk factors. These associations were generally similar when assessing the development of incident HFpEF and HFrEF. The associations of TMAO and crotonobetaine (but not choline) with incident HF did not reach Bonferroni-corrected statistical significance when further adjusted for renal function. The findings were generally consistent across age, sex, race/ethnicity, BMI, and baseline renal function, and metabolites were associated with incident HF independently in White, Hispanic/Latino, and Black participant subgroups.

Dietary intake, which is subsequently filtered through the gut microbiota, represents humans’ largest environmental exposure. Over the past decade, our group has demonstrated that metabolites produced by gut microbes following processing of dietary nutrients such as carnitine or choline may have mechanistic links to the formation of atherosclerosis and cardio-renal diseases, both in animal models and in humans.^{6,12,13,20,21,37,39–42} Relevant to the present study, a recent meta-analysis has demonstrated the prognostic value of TMAO in clinical samples of patients with established HF.⁴³ The present findings on incident HF from pooled analysis of two community-based cohorts further supports and greatly expands the potential links between HF development and microbe-derived metabolites including TMAO, choline, and carnitine. While the precise mechanisms remain largely unknown, animal models of HF have demonstrated that pharmacologic inhibition of choline trimethylamine lyase has the capacity to inhibit the exacerbated fibrotic responses in both heart and kidneys to a choline-rich diet.^21,37,44 In addition, some individuals are disposed to generate higher concentrations of TMAO following a red meat diet (rich in carnitine), while a choline rich diet may also raise circulating TMAO concentrations.^35,45 In our recent work with CHS, we have identified TMAO, choline, carnitine, and γBB as independently associated with mortality.⁴⁶ Our current findings suggest that the diet-microbiome axis may also be relevant for HF, supporting the need for future studies evaluating dietary and pharmacologic modification in individuals with elevated gut microbial metabolites like TMAO as a potential effective strategy in mitigating future HF development.

One novel feature of the present study is the assessment of not only plasma TMAO, but also 5 other closely-related metabolites. We found that both crotonobetaine and choline were associated with increased risk of HF independent of traditional cardiac risk factors and, for crotonobetaine, renal function (potential confounder or mediator). While the role of endogenous pathways of carnitine metabolism and microbial contributions of TMAO precursors such as γBB are better described,^5–7 far less is known about the metabolic pathways and consequences of crotonobetaine, which is also metabolized by the intestinal microbiota.^5,6 Under anaerobic conditions and in the absence of preferred substrates, some bacteria can use carnitine and its catabolic product crotonobetaine as final electron acceptors.⁸ Enterobacteria, especially Escherichia coli, Salmonella typhimurium, and Proteus vulgaris, can form L-carnitine by hydration of the double bond of crotonobetaine via crotonobetaine reductase under anaerobic conditions.⁴⁷ Our group has previously demonstrated that a red meat rich diet increases circulating crotonobetaine concentrations compared to eucaloric diets where the primary protein source is instead either white meat or vegetarian, independent of dietary fat content or renal excretion rates.³⁵ Only one prior small study has investigated carnitine-related metabolites in patients with acute HF, in which concentrations of crotonobetaine were below the levels of detection for the assays being used.¹⁸ Our new findings indicate that the potential contributory roles of crotonobetaine and choline in the development of HF warrant further studies.

This investigation has several strengths. The prospective design among individuals free of prevalent HF allows assessment of temporality and minimizes reverse causation, while the cohort design minimizes selection bias. The large sample size with adequate event rates provides statistical power for determining the association between TMAO and related metabolites and incident HF. Circulating TMAO was measured in serial samples and incorporated using time-varying modeling and improving classification. The time-varying Cox regression models appropriately allow adjustment for exposures and covariate values, including age. Because TMAO levels were measured prospectively, before development of disease, any remaining within-individual variation is likely to attenuate findings toward the null, reducing the strength of observed associations. The availability of a wealth of well-measured covariates, including sociodemographic, measured anthropometrics, blood pressure, cholesterol concentrations, inflammatory markers, education, income, and dietary habits, allows adjustment for major confounding factors. Both CHS and MESA populations consist of community-dwelling adults enrolled in longitudinal cohort studies, increasing generalizability while any remaining within-individual variation is likely to attenuate findings toward the null.

Potential limitations should also be considered. Residual confounding cannot be excluded in any observational analysis, even though our findings were robust to adjustment for major known HF risk factors as well as other sociodemographic and lifestyle habits. Only a subset of cases had sufficient data on left ventricular systolic function to classify HFpEF or HFrEF, although this mirrors clinical experience in practice. In addition, the number of cases in certain racial/ethnic subgroups, such as Chinese, were too small to perform meaningful analysis of racial differences. In these large cohorts followed for many years, we did not have information on fecal gut microbial composition, liver flavin monooxygenase 3 activity, or genetic data available for this analysis, which could have provided additional evidence to inform the observed associations.

In conclusion, here we present a pooled analysis of two well-phenotyped community-based cohorts of diverse U.S. adults without HF at baseline, wherein incident HF was associated with circulating concentrations of gut microbe-derived metabolites including TMAO, crotonobetaine and choline.

CLINICAL PERSPECTIVES.

What is New?

• In two prospective, diverse, community-based cohorts, circulating levels of the gut microbial metabolites TMAO and crotonobetaine, and the related metabolites carnitine and choline were associated with heart failure risk.

• These associations were independent of each other and hold true following adjustment for traditional risk factors but were attenuated after further adjustment for renal function.

What are the clinical implications?

• TMAO and related metabolites may help to predict future development of heart failure.

• Targeting TMAO generation may serve as a therapeutic tool for heart failure prevention.

ABBREVIATIONS

CHS

Cardiovascular Health Study

CKD

chronic kidney disease

CVD

cardiovascular disease

GFR

glomerular filtration rate

HF

heart failure

HFpEF

heart failure with preserved ejection fraction

HFrEF

heart failure with reduced ejection fraction

γBB

gamma butyrobetaine

MESA

Multiethnic Study of Atherosclerosis

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

TMAO

trimethylamine N-oxide