Longitudinal associations of plasma TMAO and related metabolites with cognitive impairment and dementia in older adults: the Cardiovascular Health Study

Abstract

Background:

Animal studies suggest that gut microbiome metabolites such as trimethylamine N-oxide (TMAO) may influence cognitive function and dementia risk. However potential health effects of TMAO and related metabolites remain unclear.

Objective:

We examined prospective associations of TMAO, γ-butyrobetaine, crotonobetaine, carnitine, choline and betaine with risk of cognitive impairment and dementia among older adults aged 65 years and older in the Cardiovascular Health Study (CHS).

Methods:

TMAO and metabolites were measured in stored plasma specimens collected at baseline. Incident cognitive impairment was assessed using the 100-point Modified Mini-Mental State Examination administered serially up to 7 times. Clinical dementia was identified using neuropsychological tests adjudicated by CHS Cognition Study investigators, and by ICD-9 codes from linked Medicare data. Associations of each metabolite with cognitive outcomes were assessed using Cox proportional hazards models.

Results:

Over a median of 13 years of follow-up, 529 cases of cognitive impairment, and 522 of dementia were identified. After multivariable adjustment for relevant risk factors, no associations were seen with TMAO, carnitine, choline, or betaine. In contrast, higher crotonobetaine was associated with 20–32% higher risk of cognitive impairment and dementia per interquintile range (IQR), while γ-butyrobetaine was associated with ~25% lower risk of the same cognitive outcomes per IQR.

Conclusion:

These findings suggest that γ-butyrobetaine, crotonobetaine, two gut microbe and host metabolites, are associated with risk of cognitive impairment and dementia. Our results indicate a need for mechanistic studies evaluating potential effects of these metabolites, and their interconversion on brain health, especially later in life.

Introduction

The potential influence of dietary factors on brain health is of growing interest. Diet influences many risk factors that cause cognitive decline, such as hypertension that contributes to vascular-related cognitive dysfunction; and nutrition may also influence onset, progression, or thresholds for manifestation of degenerative brain diseases like Alzheimer’s disease [1–9]. However, which dietary components play a role in cognitive decline and dementia remain unclear.

The gut microbiome has recently gained attention as a potential factor that may influence cognitive function [10]. Among microbiome-related metabolites, trimethylamine-N-oxide (TMAO) is derived from microbiome-dependent metabolism of L-carnitine and choline, nutrients present in red meat, poultry, fish and eggs [11–14]. Mechanistic animal model studies show that dietary choline, carnitine, and TMAO directly accelerate atherosclerosis through alterations in tissue sterol metabolism, activation of vascular inflammatory pathways, and promotion of foam cell formation [13–18] Microbiome dependent metabolites, such as TMAO, could also play a role in cognitive function through mechanisms underlying dementia pathology and neuronal function, including for Alzheimer’s disease [19–21]. In a cross-sectional study of 410 older adults, those with mild cognitive impairment and Alzheimer’s disease had higher TMAO concentrations in cerebrospinal fluid, compared to individuals with no signs of cognitive impairment [6]. This study also found associations between TMAO levels and biomarkers of Alzheimer’s disease and neuronal degeneration. However, the prospective associations of TMAO with cognitive health have not been reported. It is also unknown whether TMAO precursors – such as carnitine-derived γ-butyrobetaine and crotonobetaine, or choline and its oxidation product betaine [12, 22] (Figure 1) – may influence cognitive function or dementia risk. γ-butyrobetaine, for example, is promoted as a supplement to help shift carnitine metabolism toward synthesis of acetyl-l-carnitine[23], which crosses the blood-brain barrier and is reported to assist with mental cognition [7–9, 24, 25]. However, the association of blood γ-butyrobetaine levels with cognitive decline and dementia in humans has not been studied. To elucidate potential effects of TMAO and its family of microbiome-related metabolites on brain health later in life, we investigated prospective associations of TMAO, γ-butyrobetaine, crotonobetaine, cartinine, choline, and betaine with incident cognitive impairment and dementia in the Cardiovascular Health Study (CHS).

Figure 1.

Pathways for generation of trimethylamine N-oxide (TMAO) and its intermediates. Black arrows represent transformations performed by the host, and red arrows, transformations performed by gut microbes. The endogenous biosynthesis of carnitine involves multiple steps from Lysine to γ-butyrobetaine, indicated by a chain of arrows. In healthy subjects, γ-butyrobetaine is also endogenously synthesized from lysine, independent of gut-microbiota.[45, 53] In contrast, production of TMAO and crotonobetaine are profoundly suppressed by antibiotic administration,[53] supporting a dominant role of gut microbial metabolism in their generation. (Figure reproduced with permission from Wang M. et al., Dietary meat, trimethylamine-N oxide-related metabolites, and incident cardiovascular disease among older adults, in press)

Methods

Study design and population

CHS is a multi-center prospective cohort study designed to investigate risk factors for cardiovascular diseases in older U.S. adults [26]. Community-based populations of individuals aged 65 years and older were enrolled from four U.S. communities (Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Allegheny County, Pennsylvania) including 5,201 subjects in 1989–1990 and an additional 667 African Americans in 1992–1993. Trained personnel performed annual study clinic evaluations including cognitive testing, physical examination, diagnostic testing and blood sampling, and questionnaires on medical history, health status, hospitalization, and lifestyle. The CHS Cognition Study, an ancillary study initiated to evaluate incident dementia in the cohort [2, 27], identified participants with prevalent dementia at the time of magnetic resonance imaging MRI and cognitive tests in 1992–94, which served as the baseline year for this analysis. All hospitalizations were ascertained every six months with detailed medical record review for cardiovascular events[28]. Study protocols were approved at the institutional review board of each participating university. All participants signed an informed consent.

TMAO and related biomarkers

Circulating plasma concentrations of TMAO, choline, betaine, carnitine, γ-butyrobetaine, and crotonobetaine were measured in stored plasma specimens collected at baseline (i.e. 1989–90 for the original cohort and 1992–93 for the minority cohort; n=5,148) and in 1996–97 (n=3,338). All compounds were quantified by stable isotope dilution liquid chromatography with online electrospray ionization tandem mass spectrometry (LC/MS-MS) using a triple quadrupole mass spectrometer (both LCMS-8050 and LCMS-8060, Shimadzu Corporation, Kyoto, Japan). Analyses were performed in positive ion mode with multiple reaction monitoring (MRM) using the following characteristic parent to daughter ion transitions for endogenous analytes: m/z 76.10 → 59.10 (TMAO), m/z 104.00→ 60.05 (choline), m/z 118.10→ 58.10 (betaine), carnitine (m/z 162.00→ 103.00), γ-butyrobetaine (m/z 146.00→ 87.05), and crotonobetaine (m/z 144.00→ 58.10). Stable isotope–labeled internal standards used were d9-TMAO, d9-choline, d9-betaine, d3-carnitine, d9-butyrobetaine, and d9-crotonobetaine (the former four internal standards were purchased from Cambridge Isotope Laboratories, Tewksbury, MA, USA, and the latter two were synthesized, as previously described [12, 29]). Internal standards dissolved in methanol were added to plasma samples at same time as cold methanolic protein precipitation. The isotope labeled internal standards were similarly monitored in MRM mode using parent to daughter ion transitions as follows: m/z 85.00 → 66.15, m/z 113.10 → 69.20, m/z 127.00 → 66.10, m/z 165.00 → 103.05, m/z 155.00 → 87.05 and m/z 153.00 → 66.15 for d9-TMAO, d9-choline, d9-betaine, d3-carnitine, d9-butyrobetaine and d9-crotonobetaine, respectively [12, 29]. Laboratory coefficient of variations (both intra-day, and interday) for TMAO, choline, betaine, and carnitine were <6%, and for γ-butyrobetaine and crotonobetaine were <8%. For the present analyses, we included the 3,775 participants with circulating TMAO measures who were also free of clinical stroke or dementia in 1992–93, the baseline year for this investigation.

Cognitive Outcomes

Our primary outcomes were incidence of cognitive impairment and dementia. Cognitive impairment was assessed using standardized serial cognitive testing based on the Modified Mini-Mental State Examination (3MSE), a validated tool assessing different aspects of cognition including memory, orientation, calculation and verbal fluency [30]. Cognitive tests were serially administered during annual in-person visits between 1992–93 and 1999–2000 and again in 2005–2006. For participants who did not attend any particular study visit (4–17% each year), cognitive testing was performed over the phone using the Telephone Interview for Cognitive Status (TICS) [31] and the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE). TICS or IQCODE were collected annually between 1992–93 and 1998–1999, in 2005–2006, and again annually from 2007–2008 until 2015. Missing 3MSE scores were estimated using equations developed and validated in CHS as previously described [31]. Analysis of incident cognitive impairment (n= 3,178) included cognitive scores estimated based on 3MSE in combination with TICS and IQCODE scores. Participants with baseline subclinical cognitive impairment (3MSE scores <80; n=274) were excluded from these analyses. Individuals were classified as having new cognitive impairment if they had low 3MSE scores (≤80) on two consecutive annual visits or on a single visit with no subsequent cognitive information (n=529).

Analyses of incident dementia (n=3,775) included an additional 1,289 CHS subjects not participating in the ancillary Cognition Study (319 cases). Among those in the CHS Cognition Study, potential dementia cases were identified during in-clinic visits based on the participants’ self-report of cognitive deficits affecting their daily life, a history of normal intellectual function before the onset of cognitive abnormalities, and assessment of cognitive impairment in two cognitive domains that did not necessarily include memory[27]. Potential dementia cases were initially evaluated at each CHS study site, with subsequent adjudication by a central dementia committee based on review of medical records, including participant’s neuropsychological and brain imaging tests[27, 32]. Among other CHS participants, dementia cases were identified from hospitalization ICD-9 codes, requiring at least two separate ICD codes indicating a diagnosis of dementia, Alzheimer’s disease, or related disorders.

Covariates

Information on sociodemographics, medical history, prescription medications, anthropometric measures, blood pressure, and laboratory values were obtained using standardized procedures during study clinic examinations. Leisure activity was assessed by using a modified Minnesota Leisure-Time Activities questionnaire that evaluated the frequency and duration of 15 activities [33]. Diet was assessed in 1989–1990 with the use of a validated 99-item food-frequency questionnaire adapted from the National Cancer Institute [34], and again in 1995–1996 with the use of a validated Willett food-frequency questionnaire [35]. Deoxyribonucleic acid was collected from consenting participants, from which apolipoprotein E genotype was assessed [36]. Height, weight, waist circumference, and resting seated blood pressure were directly measured during study clinic examinations by trained personnel using established methods [26]. Serum creatinine was measured using a colorimetric method (Ektachem 700,Eastman Kodak, Rochester, NY) and calibrated to isotope dilution mass spectrometry [37]. Cystatin C was quantified from frozen samples using a BNII nephelometer 10 (Siemens; Deerfield, Ill, USA) [38]. Estimated glomerular filtration rate (eGFR) was calculated using the CKD-EPI Creatinine-Cystatin C equation [39]. Missing covariate data (<2% for most; 8–12% for dietary factors) were imputed with the use of single imputation adjusting for multiple demographic and risk factor variables. Prior analyses in CHS have shown no appreciable difference in results between single and multiple imputation methods for these factors [40].

Statistical Analysis

We evaluated associations between each biomarker and incident cognitive impairment and dementia using Cox proportional-hazards models, with time-at-risk until the first event or censoring including death or the last date of adjudicated follow-up (PROC PHREG procedure using counting process for time dependent variables in SAS[41]). Time-varying exposures were assessed, with biomarker levels measured in 1989–1993 samples related to incident events through 1996–97, and the average of two biomarker levels measured in 1989–1993 and 1996–97 samples related to events occurring after 1996–97. Metabolite concentrations were evaluated in quintiles as indicator variables, and continuously by using the interquintile median range (IQR, defined as the difference between the median of the fifth and first quintiles). Statistical significance of trends across quintiles was tested by assigning participants the median value in each quintile and assessing this variable continuously. For metabolites with a significant association across quintiles or per IQR, potential nonlinear associations were assessed using restricted cubic splines.

The proportional hazards assumption was tested by using extended Cox models with product terms of each metabolites and covariates and time-dependent log of time to event. We found evidence for violation of the proportional hazards assumption for age, age², years of education, and study site, which were addressed by incorporating these as risk-set stratification variables [42]. To evaluate and minimize potential confounding, we adjusted for all major risk factors related to cognitive outcomes in older adults or associated with TMAO concentrations and/or its dietary sources. Multivariable models included adjustments for age (years), age², sex, race (white, nonwhite), and study site (4 sites) (model 1); additional adjustment for education (years of education through 12th grade (continuously), any education beyond 12th grade [yes or no]), income (<$12,000, $12,000–$24,999, $25,000–$49 999, >$50,000/year), APOE ε−4 genotype (at least one ε−4 allele vs none), and most recent information on time-varying smoking status (current, former, never), leisure-time physical activity (kcal/week), and intakes of alcohol (drinks/week), fruits (servings/day), and vegetables (servings/day) (model 2). In order to evaluate influence of factors that could be either mediators or confounders of the relationship between these biomarkers and cognitive outcomes, we evaluated a model with additional adjustment for time-varying estimated glomerular filtration rate (eGFR), prevalent CHD, atrial fibrillation and heart failure (model 3), each of which could be in the downstream causal pathway of effects of TMAO and its precursors. We explored potential effect modification by APOE ε−4 status, sex, and categories of eGFR (eGFR >= 60 vs < 60) by including multiplicative interaction terms with each biomarker level in statistical models. We explored potential interactions with circulating carnitine concentrations using similar methods. P values were 2-sided, with P<0.05 indicating significance; except for exploratory subgroup comparisons which were Bonferroni-corrected for multiple comparisons (3 effect modifiers × 6 biomarkers = 0.05/18 = 0.0027). Analyses were performed using Stata, release 14.0 (StataCorp), and SAS, version 9.4 (SAS Institute).

Results

At baseline, mean±SD age was 71.6±4.8 years, 65% of participants were female, and 17% were non-white. Carnitine and betaine had highest mean plasma concentrations (mean±SD, 37.3±8.1 and 37.1±12.8, respectively), followed by choline (9.7±3.4), TMAO (7.2±10.9), γ-butyrobetaine (1.0±0.3), and crotonobetaine (0.02±0.01) (Table 1). Intercorrelations of these metabolites and with potential dietary sources were low to moderate, with Spearman partial correlation coefficients (r) ranging from 0.04 to 0.36 (Table 2).

Table 1.

Baseline characteristics of 3.775 US adults free of dementia and stroke in the Cardiovascular Health Study

Sociodemographic Factors
Age (SD), years	71.6 (4.8)
Race, % white	83
Sex, % male	35
Years of Education through 12th grade	11.1 (1.8)
High School or higher, %	73
Annual income
<12,000	24
$12,000–$24,999	36
$25,000–$49,999	27
>$50,000	13
Medical History
Coronary Heart Disease, %	16.5
Atrial Fibrillation, %	2.1
Heart Failure, %	3.0
Lifestyle
Former Smoker, %	47
Current Smoker, %	12
Diet
Mean intake of fish (SD), servings/week	1.7 (1.4)
Mean intake of meat (SD), servings/day	0.8 (0.6)
Mean intake of eggs (SD), servings/week	0.8 (1.0)
Alcohol, drinks/week	2.6 (7.4)
Plasma concentrations of TMAO and related metabolites, mean (SD) μmol/L
TMAO	7.4 (10.9)
Betaine	37.1 (12.8)
γ-butyrobetaine	1.0 (0.3)
Carnitine	37.3 (8.1)
Choline	9.7 (3.4)
Crotonobetaine	0.02 (0.01)

Values are mean (SD) for continuous variables and percent for categorical variables at analysis baseline (1989–1990). TMAO: trimethylamine N-oxide; SD, standard deviation.

Table 2:

Intercorrelations between cumulative average of plasma TMAO, choline, betaine, carnitine, γ-butyrobetaine, crotonobetaine, and dietary habits among 3,775 older U.S. adults¹

	Choline	Carnitine	Betaine	γ-butyrobetaine	Croton obetaine	Red meat	Fish	Eggs
TMAO	0.16	0.06		0.20	0.33		0.09
Choline		0.18	0.23	0.27	0.33	0.05
Carnitine			0.22	0.36	0.25	0.08		0.04
Betaine				0.29	0.10			0.04
γ-butyrobetaine					0.32	0.09		0.04
Crotonobetaine						0.07		0.04

¹Estimated using partial Spearman correlation analysis adjusting for age, sex, race. Only statistically significant coefficients (p-value < 0.05) are shown. Cumulative average of TMAO and its related precursors at each time point, with 50% weight assigned to 1996–1997 measurement and 50% weight for 1989–1993 measures when available.

Dietary habits were assessed using average of 1989–1990 and 1995–1996 measures.

TMAO, Trimethylamine N-oxide

Sociodemographic, clinical, and dietary risk factors at baseline according to plasma concentrations of TMAO and its precursors are shown in Supplemental Tables 1, 3 and 4. Overall, participants with higher metabolite concentrations were more likely to be male, have prevalent coronary heart disease (CHD) and atrial fibrillation, and lower eGFR. Dietary habits were more similar across quintiles of these plasma metabolites, except fish intake was slightly higher across increasing quintiles of TMAO concentrations, and red meat intake across increasing quintiles of plasma choline concentrations.

Incident Cognitive Impairment

During 42,130 person-years of follow-up, 529 new cases of cognitive impairment were identified (12.6 cases per 1,000 person-years). In multivariable models adjusting for sociodemographic and lifestyle factors, TMAO, betaine, carnitine, and choline concentrations were not significantly associated with incidence of cognitive impairment. However, plasma levels of a key carnitine- and gut microbiota-derived TMAO precursor, γ-butyrobetaine, was associated with lower risk of cognitive impairment, with 38% lower risk in the highest vs. lowest quintile (HR: 0.62; 95% CI: 0.44, 0.88) (Table 3, Figure 2). Conversely, plasma crotonobetaine, which is both a likely host generated intermediate in carnitine synthesis (from γ-butyrobetaine), and also a gut microbiota-derived intermediate in carnitine transformation (from γ-butyrobetaine) into TMAO, was associated with higher risk of cognitive impairment, with 47% higher risk when comparing extreme quintiles (1.47; 1.09,1.97). Assessed continuously per IQR, crotonobetaine was associated with 32% higher risk (1.32; 1.06, 1.65). These associations were not materially altered by further adjustment for fish and red meat consumption, as well as for factors that could be confounders or mediators such as eGFR, prevalent CHD, atrial fibrillation and heart failure. In semi-parametric analyses, we found little evidence of nonlinear associations, except for a borderline P-nonlinearity (P=0.04) for γ-butyrobetaine (Figure 1). Crotonobetaine is believed to be an intermediate in the multi-step gut microbial transformation of dietary L-carnitine into trimethylamine (TMA), the precursor of TMAO [43]. Specifically, the CoA derivative of crotonobetaine is a proposed intermediate during microbial conversion of L-carnitine into γ-butyrobetaine[43, 44]. In exploratory analyses we therefore assessed the ratio of plasma γ-butyrobetaine:crotobetaine, where a higher ratio might reflect greater conversion of crotonobetaine to γ-butyrobetaine. This ratio was associated with 29% lower risk of cognitive impairment across quintiles (0.71; 0.51, 0.97) and 25% lower risk per IQR (0.75; 0.60, 0.95).

Table 3.

Risk of incident cognitive impairment according to according to plasma concentrations of TMAO, choline, betaine, carnitine, γ-butyrobetaine, and crotonobetaine^* in 3,178 Older US Adults (n=529 events)

	Quintiles of plasma metabolites
	Q1	Q2	Q3	Q4	Q5	P-trend	Hazard Ratio (95% CI) for 1-IQR unit
TMAO
Basic	1.00	1.15 (0.85, 1.55)	1.18 (0.88, 1.57)	1.09 (0.81, 1.46)	1.26 (0.93, 1.72)	0.26	0.98 (0.86, 1.12)
Multivariable	1.00	1.15 (0.85, 1.56)	1.17 (0.87, 1.56)	1.10 (0.82, 1.47)	1.26 (0.92, 1.72)	0.26	1.00 (0.87, 1.14)
Multivariable+	1.00	1.14 (0.84, 1.55)	1.15 (0.86, 1.54)	1.10 (0.82, 1.47)	1.22 (0.89, 1.67)	0.35	0.98 (0.85, 1.13)
Betaine
Basic	1.00	0.88 (0.68, 1.15)	1.05 (0.81, 1.36)	0.94 (0.70, 1.26)	0.99 (0.74, 1.32)	0.90	0.96 (0.76, 1.22)
Multivariable	1.00	0.89 (0.68, 1.16)	1.08 (0.84, 1.40)	0.95 (0.71, 1.27)	0.98 (0.73, 1.32)	0.91	0.95 (0.75, 1.22)
Multivariable+	1.00	0.90 (0.69, 1.18)	1.10 (0.85, 1.43)	0.97 (0.72, 1.29)	0.99 (0.74, 1.33)	0.87	0.96 (0.75, 1.22)
γ-butyrobetaine
Basic	1.00	0.73 (0.56, 0.93)	0.78 (0.60, 1.00)	0.89 (0.67, 1.17)	0.63 (0.45, 0.90)	0.06	0.77 (0.57, 1.03)
Multivariable	1.00	0.71 (0.55, 0.91)	0.79 (0.61, 1.01)	0.88 (0.67, 1.16)	0.62 (0.44, 0.88)	0.05	0.76 (0.56, 1.01)
Multivariable+	1.00	0.71 (0.55, 0.91)	0.80 (0.62, 1.03)	0.88 (0.66, 1.15)	0.62 (0.44, 0.89)	0.05	0.75 (0.56, 1.01)
Carnitine
Basic	1.00	1.15 (0.88, 1.50)	0.99 (0.75, 1.30)	1.33 (1.02, 1.75)	0.99 (0.72, 1.37)	0.58	1.13 (0.86, 1.47)
Multivariable	1.00	1.13 (0.86, 1.47)	0.99 (0.75, 1.31)	1.29 (0.99, 1.70)	0.95 (0.69, 1.30)	0.84	1.03 (0.80, 1.34)
Multivariable+	1.00	1.12 (0.86, 1.47)	0.99 (0.75, 1.30)	1.27 (0.96, 1.67)	0.94 (0.68, 1.30)	0.91	1.02 (0.79, 1.33)
Choline
Basic	1.00	1.02 (0.77, 1.36)	0.85 (0.64, 1.13)	1.09 (0.82, 1.44)	0.99 (0.73, 1.36)	0.85	0.97 (0.79, 1.18)
Multivariable	1.00	1.02 (0.76, 1.36)	0.85 (0.64, 1.14)	1.10 (0.83, 1.46)	0.95 (0.69, 1.30)	0.95	0.94 (0.77, 1.15)
Multivariable+	1.00	1.03 (0.77, 1.37)	0.85 (0.63, 1.13)	1.10 (0.83, 1.46)	0.96 (0.70, 1.32)	0.99	0.94 (0.77, 1.15)
Crotonobetaine
Basic	1.00	1.05 (0.82, 1.34)	0.99 (0.76, 1.29)	0.84 (0.63, 1.13)	1.54 (1.15, 2.06)	0.05	1.33 (1.07, 1.65)
Multivariable	1.00	0.99 (0.78, 1.27)	0.98 (0.76, 1.28)	0.83 (0.62, 1.11)	1.47 (1.09, 1.97)	0.09	1.32 (1.06, 1.65)
Multivariable+	1.00	1.00 (0.78, 1.28)	0.98 (0.76, 1.28)	0.83 (0.62, 1.11)	1.43 (1.06, 1.92)	0.13	1.29 (1.03, 1.62)
Butyrobetaine/Crotobetaine ratio
Basic	1.00	0.75 (0.56, 1.00)	0.78 (0.59, 1.02)	0.66 (0.50, 0.86)	0.69 (0.50, 0.95)	0.04	0.76 (0.60, 0.95)
Multivariable	1.00	0.75 (0.56, 1.00)	0.78 (0.60, 1.03)	0.66 (0.51, 0.87)	0.71 (0.51, 0.97)	0.05	0.75 (0.60, 0.95)
Multivariable+	1.00	0.76 (0.57, 1.03)	0.80 (0.61, 1.	0.68 (0.52, 0.90)	0.73 (0.53, 1.00)	0.07	0.73 (0.58, 0.93)

^*long-term exposure to trimethylamine was assessed by using cumulative average of TMAO measures, i.e., FA levels in 1989–1993 were related to risk from 1989–96; the average of TMAO levels in 1992 and 1996, to risk from 1996 onwards.

Basic models included sex, race (whites or nonwhite) and enrollment site (4 sites) and time-varying age (years), and age²; Multivariable model included additional adjustment for education (years of education through 12th grade, any education beyond 12th grade), income (<$12 000, $12 000-$24 999, $25 000-$49 999, or >$50 000/year), APOE status (at least one ε−4, none, unknown), smoking status (never, current or former smokers), alcohol intake (drinks/week), leisure-time physical activity (kcal/week). Multivariable+ includes additional adjustment for red meat intake (servings/day), fish (servings/day, total energy consumption (kcal/day), eGFR, prevalent CHD, atrial fibrillation and heart failure.

Figure 2.

Multivariable-adjusted hazard ratios (HR) of incident cognitive impairment according to concentrations of plasma γ-butyrobetaine (top) and plasma crotonobetaine (bottom), evaluated using restricted cubic splines. Solid lines and shaded area represent the HR estimate and 95% confidence interval, respectively. The dotted vertical lines correspond to the 10th, 25th, 50th, 75th, and 90th percentiles. The top and bottom 1% of participants were omitted. P values for linearity: 0.04 (γ-butyrobetaine) and 0.07 (crotonobetaine)

Incident Dementia

During 46,409 person-years of follow-up, 522 new dementia cases were identified (11.4 cases per 1,000 person-years). Findings were concordant with those for incident cognitive impairment. There were no statistically significant associations between TMAO, betaine, cartinine or choline and onset of dementia. However, even following adjustment for sociodemographic and lifestyle factors, higher γ-butyrobetaine concentrations were inversely associated with risk of incident dementia (across quintiles: 0.68; 0.49, 0.95; per IQR: 0.75; 0.57, 0.98); while higher crotonobetaine concentrations were associated with a higher risk per IQR (1.20; 1.04, 1.40) (Table 4, Figure 3). A higher ratio of plasma γ-butyrobetaine: crotobetaine was associated with 29% lower risk across quintiles (0.71; 0.51, 0.97)). Findings were similar in multivariable analysis further adjusting for fish and meat intake, eGFR, and prevalent CHD, atrial fibrillation, and congestive heart failure. Little evidence was identified for potential nonlinear associations assessed using restricted cubic splines (data not shown).

Table 4.

Risk of incident dementia according to plasma concentrations of TMAO, choline, betaine, carnitine, γ-butyrobetaine, and crotonobetaine^* in 3,775 Older US Adults (n=522 events)

	Quintiles of plasma metabolites
	Q1	Q2	Q3	Q4	Q5	P-trend	Hazard Ratio (95% CI) for 1-IQR unit
TMAO
Basic	1.00	0.97 (0.73, 1.28)	0.92 (0.70, 1.21)	0.84 (0.64, 1.11)	1.05 (0.79, 1.40)	0.59	1.01 (0.91, 1.13)
Multivariable	1.00	1.02 (0.77, 1.35)	0.95 (0.72, 1.26)	0.85 (0.64, 1.12)	1.10 (0.83, 1.47)	0.51	1.00 (0.90, 1.11)
Multivariable+	1.00	1.01 (0.76, 1.34)	0.93 (0.71, 1.24)	0.83 (0.63, 1.10)	1.07 (0.80, 1.44)	0.63	1.00 (0.90, 1.11)
Betaine
Basic	1.00	1.01 (0.77, 1.34)	1.07 (0.81, 1.41)	1.07 (0.80, 1.42)	1.04 (0.77, 1.40)	0.75	1.07 (0.85, 1.35)
Multivariable	1.00	0.99 (0.75, 1.31)	1.06 (0.81, 1.39)	1.05 (0.79, 1.39)	0.98 (0.72, 1.32)	0.96	1.07 (0.84, 1.35)
Multivariable+	1.00	1.00 (0.76, 1.32)	1.07 (0.81, 1.41)	1.05 (0.79, 1.40)	0.98 (0.73, 1.33)	0.99	1.08 (0.85, 1.36)
γ-butyrobetaine
Basic	1.00	0.80 (0.62, 1.04)	0.74 (0.57, 0.97)	0.86 (0.65, 1.13)	0.72 (0.52, 1.00)	0.11	0.79 (0.60, 1.04)
Multivariable	1.00	0.81 (0.63, 1.05)	0.76 (0.58, 0.99)	0.85 (0.65, 1.12)	0.68 (0.49, 0.95)	0.05	0.75 (0.57, 0.98)
Multivariable+	1.00	0.81 (0.63, 1.05)	0.75 (0.58, 0.99)	0.84 (0.64, 1.11)	0.67 (0.48, 0.93)	0.04	0.73 (0.56, 0.96)
Carnitine
Basic	1.00	0.88 (0.68, 1.15)	0.96 (0.74, 1.25)	0.95 (0.73, 1.24)	0.98 (0.73, 1.32)	0.95	0.95 (0.72, 1.25)
Multivariable	1.00	0.95 (0.72, 1.25)	1.02 (0.77, 1.34)	1.00 (0.76, 1.32)	0.98 (0.72, 1.34)	0.97	1.02 (0.79, 1.30)
Multivariable+	1.00	0.95 (0.72, 1.25)	1.02 (0.77, 1.34)	0.98 (0.74, 1.30)	0.98 (0.72, 1.33)	0.95	1.00 (0.78, 1.28)
Choline
Basic	1.00	0.89 (0.67, 1.18)	0.74 (0.56, 1.00)	0.92 (0.70, 1.22)	0.98 (0.73, 1.30)	0.78	1.05 (0.90, 1.22)
Multivariable	1.00	0.89 (0.66, 1.19)	0.76 (0.56, 1.02)	0.92 (0.69, 1.24)	0.94 (0.70, 1.27)	1.00	1.03 (0.86, 1.23)
Multivariable+	1.00	0.89 (0.66, 1.20)	0.75 (0.56, 1.02)	0.91 (0.68, 1.22)	0.93 (0.69, 1.26)	0.93	1.02 (0.85, 1.22)
Crotonobetaine
Basic	1.00	1.01 (0.78, 1.32)	1.10 (0.84, 1.43)	1.06 (0.81, 1.40)	1.33 (1.01, 1.75)	0.05	1.25 (1.08, 1.45)
Multivariable	1.00	1.07 (0.81, 1.41)	1.11 (0.84, 1.46)	1.07 (0.80, 1.42)	1.33 (0.99, 1.77)	0.07	1.20 (1.04, 1.40)
Multivariable+	1.00	1.07 (0.81, 1.40)	1.10 (0.84, 1.45)	1.05 (0.79, 1.41)	1.30 (0.97, 1.73)	0.11	1.23 (1.05, 1.44)
Butyrobetaine/Crotobetaine ratio
Basic	1.00	0.71 (0.54, 0.94)	0.75 (0.57, 0.97)	0.73 (0.56, 0.95)	0.63 (0.46, 0.86)	0.02	0.80 (0.63, 1.02)
Multivariable	1.00	0.77 (0.58, 1.02)	0.75 (0.57, 0.99)	0.75 (0.57, 0.98)	0.59 (0.43, 0.81)	0.01	0.81 (0.63, 1.03)
Multivariable+	1.00	0.76 (0.57, 1.02)	0.76 (0.58, 1.00)	0.76 (0.58, 0.99)	0.60 (0.43, 0.82)	0.01	0.82 (0.65, 1.05)

^*long-term exposure to trimethylamine was assessed by using cumulative average of TMAO measures, i.e., FA levels in 1989–1993 were related to risk from 1989–96; the average of TMAO levels in 1992 and 1996, to risk from 1996 onwards.

Basic models included sex, race (whites or nonwhite) and enrollment site (4 sites) and time-varying age (years), and age²; Multivariable model included additional adjustment for education (years of education through 12th grade, any education beyond 12th grade), income (<$12 000, $12 000-$24 999, $25 000-$49 999, or >$50 000/year), APOE status (at least one ε−4, none, unknown), smoking status (never, current or former smokers), alcohol intake (drinks/week), leisure-time physical activity (kcal/week). Multivariable+ includes additional adjustment for red meat intake (servings/day), fish (servings/day, total energy consumption (kcal/day), eGFR, prevalent CHD, atrial fibrillation and heart failure.

Figure 3.

Multivariable-adjusted hazard ratios (HR) of incident dementia according to concentrations of plasma γ-butyrobetaine (top) and plasma crotonobetaine (bottom), evaluated using restricted cubic splines. Solid lines and shaded area represent the HR estimate and 95% confidence interval, respectively. The dotted vertical lines correspond to the 10th, 25th, 50th, 75th, and 90th percentiles. The top and bottom 1% of participants were omitted. P values for non-linearity 0.87 (γ-butyrobetaine) and 0.59 (crotonobetaine).

Sensitivity Analysis

In order to evaluate whether associations of γ-butyrobetaine and crotobetaine with cognitive outcomes were confounded or mediated by carnitine, we included additional adjustment for circulating carnitine concentrations. Further adjustment for carnitine did not appreciably change our results. For example, multi-adjusted HR (95% CI) of incident cognitive impairment for each IQR unit difference after adjusting for carnitine was 0.69 (0.50, 0.97) for γ-butyrobetaine and 1.34 (1.05, 1.72) for crotobetaine, while HRs (95% CI) of incident dementia for each IQR unit difference was 0.65 (0.44, 0.83) for γ-butyrobetaine and 1.19 (1.00, 1.42) for crotobetaine.

Exploratory interactions

Associations of circulating γ-butyrobetaine, crotobetaine, and other TMAO-related metabolites with incident cognitive impairment and dementia were similar across subgroups by sex, APOE e-4 status, or eGFR status (P for interaction=NS each). Similarly, there was no statistically significant interactions circulating carnitine concentrations (P-interaction > NS)

Discussion

In this large, prospective investigation among older, community-based US adults, we examined whether the gut microbial metabolite TMAO and its key precursors were associated with cognitive impairment and dementia. Notably, while circulating levels of TMAO and its major nutrient sources, choline and L-carnitine, had no significant association, circulating levels of crotonobetaine and γ-butyrobetaine, intermediates in both carnitine biosynthesis in mammals, and in gut microbial metabolism involved in dietary carnitine transformation into TMAO, were associated with both incident cognitive impairment and incident dementia. Higher circulating levels of crotonobetaine were positively associated with incidence of cognitive impairment and dementia, whereas circulating levels of γ-butyrobetaine, a related and key metabolite that is interconverted with crotonobetaine by gut microbiota, were inversely associated with risk of cognitive impairment and dementia. In exploratory analyses, the ratio of these two molecules, reflecting their interconversion (i.e. crotonobetaine into γ-butyrobetaine), was associated with lower risk of adverse cognitive outcomes. No significant associations were observed for plasma betaine levels (a choline-derived metabolite via host enzymes, and another potential nutrient precursor for gut microbiota-dependent TMAO generation) and either cognitive outcome. Collectively the present studies provide a novel assessment of how TMAO and its precursor metabolites, objectively assessed using serial biomarker measures, relate to incident cognitive impairment and dementia later in life

Previous animal and human studies show that intestinal microbiota can convert dietary carnitine, a nutrient abundant in red meat, into both crotonobetaine and γ-butyrobetaine [45–47]. γ-Butyrobetaine is then further converted by distinct gut microbiota into TMA, the precursor to TMAO[43]. Mechanistic studies involving transplantation of genetically engineered microbes, and animal model studies with targeted inhibition of TMAO generation, support a contributory role of TMAO to both cardiovascular and metabolic diseases and their adverse outcomes [16, 17, 46, 48–52]. In contrast, mechanistic effects of crotonobetaine and γ-butyrobetaine have not been as widely studied. Following ingestion of a red meat rich diet, gut microbes convert carnitine into the intermediate γ-butyrobetaine [12, 14, 43, 53]. Animal model studies show supplemental γ-butyrobetaine in the diet accelerates atherosclerosis and increases TMAO levels [45]. In addition, elevation in TMAO levels by oral γ-butyrobetaine ingestion has also been reported to increase thrombosis potential in animal models of arterial injury [43]. γ-Butyrobetaine generation by gut microbiota has thus far been predominantly linked to adverse cardiovascular effects. However, γ-butyrobetaine can also be synthesized endogenously, as it is the proximal precursor to L-carnitine biosynthesis in mammals [54–56]. The enzyme γ-butyrobetaine hydroxylase, which is abundant in kidney, liver and brain tissues, catalyzes the transformation of γ-butyrobetaine into L-carnitine[57]. What the impact of endogenously synthesized γ-butyrobetaine is on cognitive function remains unknown. Similarly, the potential impact of crotonobetaine on the host, is essentially unexplored.

A variety of experimental mechanistic studies have reported that L-carnitine and its most common metabolite, acetyl-L-carnitine (ALCAR), have neuroprotective effects in various disorders including Alzheimer’s disease, hypoxia-induced ischemia, and traumatic brain injury, through inflammatory-oxidative mechanisms including reduced inflammation and lipid peroxidation [9, 58], oxidative stress [59–62] and oxidative DNA damage and cell death [58]. Recent studies hypothesize stronger benefits for ALCAR compared to those of L-carnitine for its role in energy status and in the endogenous production of metabolites that are essential for cell growth and neurotransmission [9]. Our findings support the need to investigate a possible relationship between higher plasma γ-butyrobetaine and increased synthesis of L-carnitine and its metabolite ALCAR. In addition, crotonobetaine is generated from gut microbiota during L-carnitine transformation into γ-butyrobetaine [14, 43]. Whether or not crotonobetaine is harmful for neurocognitive function deserves further examination. Similarly, as suggested by our findings, higher γ-butyrobetaine levels could be a marker of brain protection via increased metabolism of crotonobetaine. Our results indicate a need for mechanistic studies evaluating potential effects of crotonobetaine, γ-butyrobetaine, and their interconversion (both by host enzymes, and gut microbial sources) on brain health, especially later in life.

An important finding in the present prospective study is that we did not observe a significant association between circulating levels of TMAO itself, or three of its dietary precursor nutrients – carnitine, choline, and betaine – with cognitive health later in life. The specificity of our findings for γ-butyrobetaine and crotonobetaine – one of the Bradford-Hill criteria for considering causality [63] – rather than all TMAO-related molecules helps support their potential role in cognitive impairment and dementia. In this cohort, high plasma TMAO, but not other metabolites was associated with higher incidence of atherosclerotic cardiovascular disease among older adults, especially within those with impaired renal function [64]. The absence of any association for TMAO here provides further support for its specific relationship with cardiovascular diseases, with which it is significantly associated in this cohort [64] as well as others[15, 65–68].

Very few studies have evaluated associations between TMAO or its precursors and brain health. In a recent prospective investigation, serum TMAO concentrations were associated with early neurological deterioration after acute ischemic stroke in a clinical sample of 362 patients (mean age, 62.5 ± 9.6 years)[69]. In contrast to our investigation of older adults in generally good health (and that specifically excluded subjects with prior history of stroke), this prior study focused on associations between TMAO and neurological decline following ischemic stroke. In a cross-sectional study of 410 adults (mean age, 61.9 ± 7.9 years), cerebrospinal fluid TMAO levels were associated with mild cognitive impairment, dementia, and biomarkers of Alzheimer’s disease pathology (i.e. phosphorylated tau [p-tau], p-tau/Aβ42), and neuronal degeneration (i.e. total tau and neurofilament light chain protein) [6]. Among 171 human microbial metabolites evaluated in a recent large network-based cross-sectional analysis, TMAO showed the strongest association with Alzheimer’s disease, with 9 genetic pathways identified as leading to greater susceptibility to cognitive impairment [19]. However, these cross-sectional studies have numerous potential weaknesses. For example, they could not establish temporality, and cognitive impairment or dementia could change dietary habits and even microbial composition over time, leading to reverse causation. In addition, like many untargeted analyses, the metabolites were quantified using semiquantitative methods. In contrast, our findings greatly extend and expand these previous limited results by quantitatively (stable isotope dilution LC/MS/MS methodology) evaluating measures of plasma TMAO, γ-butyrobetaine, crotonobetaine, carnitine, choline, and betaine in a large cohort of older, community-dwelling adults. Our studies leverage access to well-defined cognitive outcomes, and were performed utilizing prospective analyses, which establish temporality and minimize reverse causation. Finally, our studies utilize serial plasma measures over extended follow-up to better assess these chronic endpoints with long induction periods.

Potential limitations of the present study should be considered. We cannot exclude the possibility of residual confounding by unmeasured or imprecisely measured confounders. On the other hand, all findings were multivariable adjusted for numerous major risk factors for brain health including sociodemographics, lifestyle, dietary covariates, and prevalent diseases. Missing 3MSE scores could result in underestimation of cognitive impairment in our study population. In addition, some dementia outcomes may have been missed, which would probably be random with respect to metabolite assessment, and may lead to potential attenuation of true associations. CHS included mostly white older adults, and results may thus not be generalizable to younger populations or other racial groups. Conversely, the broad, community-based recruitment strategy in CHS increases generalizability to the general population of older US adults.

In conclusion, our prospective investigation of older adults provides novel evidence that two gut microbe and host metabolites linked to carnitine pathways, γ-butyrobetaine and crotonobetaine, and their ratio, are associated with incident risk of cognitive impairment and dementia later in life.

Abbreviations:

TMAO

trimethylamine N-oxide

eGFR

estimated glomerular filtration rate

APOE

apolipoprotein E

CHD

coronary heart disease

CHS

Cardiovascular Health Study

BMI

body-mass index

3MSE

Modified Mini-Mental State Examination

IQCODE

Informant Questionnaire on Cognitive Decline in the Elderly