Study cohorts

This study was embedded in the Rotterdam Study (RS), with a partial validation for neuroimaging associations in the Leiden Longevity Study (LLS). A brief description of both cohorts can be found in Methods S1 [14, 15]. Both studies comply with Declaration of Helsinki and are approved by the Medical Ethical Committee of the respective institutional review boards (registration number MEC 02.1015 and P01.113). All participants provided written informed consent.

The current study includes participants of the RS for whom plasma levels of TMAO and its precursors were measured at study entry (n = 3,933), originating from the fourth examination round (2002–2005) of the first subcohort (RS-I-4) and the second examination round (2012–2014) of the third subcohort (RS-III-2). Out of these eligible participants, we identified participants who were dementia-free at baseline and subsequently created subsamples of participants who had (1) complete data on neuropsychological tests (n = 3,143) or (2) a brain MRI during the same center visit (n = 2,047) or (3) follow-up for incident dementia (n = 2,517). For validation of neuroimaging associations in the LLS, we only included offspring or long-lived sibling-pairs and partners of the offspring, who were dementia-free and had data on plasma levels of TMAO and its precursors, as well as brain MRI (n = 318).

2.2 Assessment of TMAO and its precursors in plasma

Plasma samples were obtained from fasting participants of the RS and non-fasting participants of the LLS using ethylenediaminetetetacetic acid tubes. TMAO and its precursors were quantified by mass spectrometry in a similar manner for the RS and LLS. An extensive description of the analytical techniques can be found in Methods S2 [16].

Assessment of general cognition

Between 2002 and 2005, the protocol of the Rotterdam Study was extended with a comprehensive neuropsychological test battery for cognitive assessments. Three Stroop tests (reading, color naming and interference tasks), a letter-digit substitution task (LDST), a categorical Word Fluency Test (WFT), a Purdue pegboard (PPB) tests for the left hand, right hand and both hands and a 15-word verbal learning test based on Rey’s recall of words (15-WLT) were added to the protocol. As described in our previous publications [17], a summary measure of general cognition (‘G-factor’) was created using the first component of the principal component analysis that included the delayed recall score of the Stroop interference test, LDST, verbal fluency task, the PPB test and the 15-WLT. The variance in cognitive test scores explained by this G-factor was 50.3% for RS I-4 and 51.9% for RS III-2.

Neuroimaging protocol

MRI scans of the brain, including T1-weighted, FLAIR, and proton density–weighted sequences, were available in both the RS and the LLS studies for the purpose of evaluating brain volumetry, including total brain, gray matter, white matter, hippocampus, and white matter hyperintensities. Participants with large strokes hampering segmentation results were excluded. Diffusion tensor imaging was used for determining the mean diffusivity and fractional anisotropy. Additional information on preprocessing steps, segmentation methods and other technical details is presented in Methods S3 [18,19,20,21,22,23,24,25,26,27,28,29,30,31].

Dementia screening and surveillance

The ascertainment methods for dementia in the RS have been described in full in our previous publications [32]. Participants were cognitively screened using the Mini-Mental State Examination (MMSE) and the Geriatric Mental State (GMS) Schedule organic level, both at baseline and subsequent center visits. Participants scoring less than < 26 for MMSE or > 0 for GMS underwent further investigation and informant interview, including the Cambridge Examination for Mental Disorders of the Elderly. Data acquired from in-person screening was supported by data from the electronic linkage of the study database with medical records from general practitioners and the regional institute for outpatient mental health care, which facilitated continuous monitoring for dementia or cognitive disturbances between center visits.

The final diagnosis was established via a consensus panel of study physicians and was led by a consultant neurologist, who adhered to standard criteria for dementia (DSM-III-R) and Alzheimer’s disease (NINCDS-ADRDA) to determine presence, probability, and subtype of dementia. Follow-up was complete until January 1st 2018, and participants were censored within this follow-up period at date of dementia diagnosis, date of death, date of loss to follow-up, or January 1st 2018, whichever came first. Only participants from RS-I-4 were included for this analysis, as the number of individuals with incident dementia in the younger RS-III-1 sample was limited.

Covariables

Assessment of covariables (educational attainment, smoking status, history of stroke, history of coronary heart disease (CHD), medication use, blood pressure, body mass index, total cholesterol, HDL-cholesterol, diabetes, e.g.) was comparable between the RS and LLS, and detailed further in Methods S4.

Statistical analysisAssociations of TMAO and its precursors with cognition and brain MRI

Prior to analysis, plasma levels of TMAO and its precursors (in µmol/l) were log transformed to achieve normal distributions. The volume of white matter hyperintensities (cm3) was also log-transformed, as it had a left-skewed distribution. All structural imaging markers on brain MRI (cm3) were standardized to facilitate comparison. Cross-sectional associations between plasma levels of TMAO and its precursors with general cognition (G-factor) as well as brain MRI markers, were explored with linear regression models. All effect estimates are presented as mean differences, reflecting a log-unit change in the metabolite on the outcome, and its corresponding 95% confidence intervals (95% CIs). Due to slight variations in the assessment among samples, effect estimates were pooled using a random-effects meta-analysis using the inverse variance method and the DerSimonian-Laird estimator [33], with P-Het that represents the P-value for heterogeneity. The pooled estimates were also summarized in cluster plots to display how plasma levels of TMAO and its precursors vary across cognition levels and how they associate with various neuroimaging markers.

For all statistical models, the selection for potential confounders was guided by literature review and biological plausibility [34]. Model I was adjusted for the main potential confounders, including age, sex, education, use of lipid lowering medication and body mass index. Model II was additionally adjusted for total cholesterol, HDL-cholesterol, smoking, hypertension and history of coronary heart disease, which are likely confounders, but could also be mediators. For all models that included structural imaging markers on brain MRI, we additionally adjusted for intracranial volume and, if applicable, the time-interval between blood sampling and MRI. Models pertaining to white matter hyperintensities, fractional anisotropy and mean diffusivity were additionally adjusted for (normal appearing) white matter volume. Only participants with complete data on covariables were included.

Associations of TMAO and its precursors with incident dementia

Relationships between plasma levels of TMAO and its precursors with incident dementia and Alzheimer’s disease, were determined for participants of RS-I-4 using Cox proportional hazard models, from which we obtained hazard ratios (HRs) and 95% CIs. Models were adjusted for a similar set of potential confounders as described earlier.

Stratified, sensitivity and supplementary analyses

Plasma levels of TMAO and its precursors may be affected by use of antibiotics and proton pump inhibitors (PPIs), as well as impaired clearance by the kidneys. They can also differ by sex. This information was only available for a subset of participants followed for incident dementia in the RS. In this subset, we performed a stratified analysis, where we studied associations of TMAO and its precursors with incident dementia, separately for non-users and users of antibiotics or PPIs. As a sensitivity analysis, we studied associations between TMAO and its precursors with incident dementia, among those with impaired renal function (estimated glomerular filtration rate < 60 mL/min). We also studied whether associations differed by sex. In a supplementary analysis, we assessed the correlation between TMAO and its precursors using Pearsons’ correlation coefficients.

Significance thresholds and software used

Multiple testing correction was applied using the false discovery rate (FDR) of Benjamini-Hochberg [35], and a suggestive association was considered at a nominal significance threshold of α = 0.05 (P ≤ 0.05). Analyses were performed using R version 3.6.1 (packages tidyr, dplyr, lubridate, foreign, and survival) and IBM SPSS Statistics version 24.0 (IBM Corp, Somers, NY), after which the meta-analysis was performed using the R package meta.