Human gut microbiota has been recognised as a significant contributor to host metabolism and health(Reference Tang, Backhed and Landmesser1,Reference de Vos, Tilg and Van Hul2) . Recently, several gut microbiota-derived metabolites have been associated with human health, including type 2 diabetes, CVD and cancer(Reference de Vos, Tilg and Van Hul2,Reference Witkowski, Weeks and Hazen3) . Among these metabolites, trimethylamine N-oxide (TMAO) stands out as a product generated by gut microbiota from dietary phosphatidylcholine, choline and L-carnitine, which are abundant in animal-based food(Reference Wang, Klipfell and Bennett4,Reference Koeth, Wang and Levison5) . Previous studies suggest that TMAO may increase mortality risk by contributing to the pathogenesis of CVD(Reference Tang, Wang and Levison6,Reference Abbasi7) , platelet function and thrombosis risk(Reference Zhu, Gregory and Org8), renal dysfunction(Reference Tang, Wang and Kennedy9), hyperglycaemia(Reference Chen, Henderson and Petriello10), cognitive impairment and carcinogenesis(Reference Chan, Law and Waye11).
A recent cohort study adopted serially measured plasma TMAO levels and changes in TMAO as exposures of interest and found positive associations between TMAO and both all-cause and cause-specific mortality(Reference Wang, Li and Wang12). Several meta-analyses of epidemiological studies have also indicated that higher TMAO levels were associated with an increased risk of all-cause mortality(Reference Schiattarella, Sannino and Toscano13–Reference Farhangi16). However, significant heterogeneity was detected across these investigations(Reference Farhangi16). Notably, most of the studies were conducted in western populations, while observations based on Chinese populations remain relatively scarce. A nationwide study revealed that 12·6 % of northeast China’s population (Heilongjiang, Jilin) and 11·4 % of north China’s population (Beijing, Tianjin) were at high CVD risk, compared with only 8·0 % in south China [1]. Report on Cardiovascular Health and Diseases in China 2023 shows that CVD accounted for 48·98 % and 47·35 % of deaths in rural and urban areas in 2021, respectively [2]. Therefore, CVD has become an important disease burden in rural areas of the north, and the research on TMAO and cardiovascular mortality may provide clues to the aetiology for explaining the high prevalence of CVD in rural Northern China.
Furthermore, in recent years, numerous studies have demonstrated that genetic susceptibility might interact with non-genetic factors, such as lifestyles and biomarkers, to influence on adverse health outcomes(Reference Li and Qi17,Reference Virolainen, VonHandorf and Viel18) . Genome-wide association studies have identified genetic loci associated with survival time(Reference Liu, Song and Li19) as well as TMAO levels(Reference Rhee, Ho and Chen20,Reference Jia, Dou and Gao21) . However, little is known about whether genetic factors could modify the association between TMAO levels and mortality risk. More evidence is needed to address the research gap.
There, in this prospective study of 500 participants from the Fangshan family-based cohort study in Beijing, China, we examined the associations between plasma TMAO levels and all-cause, cardiovascular and stroke mortality risk, with a particular focus on exploring the potential modification effect of genetic factors.
Methods
Study design and participants
The participants in the study were recruited from a family-based cohort study, which was conducted in Fangshan District, Beijing, China, from 2005 to 2020. The family-based cohort study enrolled participants using a proband-initiated contact method, which mainly focused on the genetic and environmental risk factors for cardiometabolic diseases, such as hypertension, diabetes and stroke. Detailed information about the cohort has been described in previous publications(Reference Tang, Hu and Chen22,Reference Wu, Tang and Wu23) .
In the current analysis, we included only 500 individuals with baseline TMAO measurements. The study design and participant selection process are outlined in online Supplementary Figure 1. This study was conducted in accordance with the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects were approved by the Medical Research Ethics Committee of the Peking University Health Science Center (the ethical review number: IRB00001052-13027). Written informed consent was obtained from all subjects.
Assessment of trimethylamine N-oxide
Blood samples for all participants were collected at baseline and stored in a −80℃ refrigerator. TMAO concentrations in human serum were measured using an ELISA kit. Following the experimental standard procedure, 50 ul of blood serum, a standard sample and Horseradish Peroxidase-labeled detection antibody were added in sequence to the 96 microwell of the reagent template. The template was incubated for 30 min in a constant temperature bath and then thoroughly washed five times with distilled water. Subsequently, the substrate was added to initiate colour development, turning blue under peroxidase catalysis and yellow after the addition of the acidic stop solution. The intensity of the colour was directly proportional to the TMAO concentration. The absorbance (OD value) was measured at a wavelength of 450 nm using a microplate reader, and TMAO concentration was calculated based on a standard curve.
Assessment of mortality
The primary outcome of the study was all-cause mortality. The secondary outcomes were specific cause mortality, including CVD-related death and stroke-related death.
All the mortality data were obtained from the Death Surveillance System in the Center for Disease Prevention and Control in Fangshan District, Beijing, China. Detailed information about death was recorded in the system. Survival time was calculated from the baseline survey date to the date of death or censored at the end of the study period (31 December 2020) for participants still alive. The causes of death were classified using the International Classification of Diseases, Ninth Revision codes and International Classification of Diseases, Tenth Revision codes.
Assessment of covariates
Data on general demographic information, lifestyle factors, chronic disease status and medical history were collected via a standardised questionnaire by trained investigators. All participants completed physical measurements including weight, height, waist circumference and blood pressure. Height and weight were each measured three times by a trained investigator, and the average values were used for the analysis. BMI was calculated by dividing the weight by the square of the height: BMI (kg/m2) = weight (kg)/height (m2). The primary covariates considered were age (continuous), sex (female/male), intake of meat products (continuous), BMI (continuous), prevalent CVD (yes/no), prevalent hypertension (yes/no) and the use of antidiabetic drugs (yes/no). Daily consumption of fish, processed meat, fruits and vegetables (yes/no) was used to construct a dietary score, ranging from 0 to 4. Each healthy dietary habit was assigned 1 point, resulting in a total score ranging from 0 to 4. A diet score of ≤ 2 was classified as an unhealthy diet, while a score of > 2 was considered indicative of a healthy diet. Moreover, a healthy lifestyle score was generated based on five components: smoking status (yes/no), drinking status (yes/no), regular exercise (yes/no), BMI (< 24/≥ 24 kg/m2) and dietary pattern (diet score ≤ 2 for unhealthy, diet score> 2 for healthy). Each healthy lifestyle behaviour was assigned 1 point, resulting in a total healthy lifestyle score ranging from 0 to 5 points. This score was categorised into three groups: 0–1 points, 2–3 points and 4–5 points. Additionally, missing data for categorical variables were coded as a missing indicator, and mean values were imputed for continuous variables.
In the sensitivity analysis, models were further adjusted for healthy lifestyle score (0–1, 2–3 and 4–5 points) and waist circumstance (continuous) in addition to age (continuous), sex (females/males), prevalent CVD (yes/no), prevalent hypertension (yes/no) and the use of antidiabetic drugs (yes/no).
Definition of polygenetic risk score
Blood sample was collected from each participant for subsequent biochemical measurements and genotyping. Genotyping, quality control, phasing and imputation were performed. In brief, genetic data were obtained using Illumina genotyping arrays. Sex and relatedness were confirmed for all samples. In addition, duplicate samples or samples with a Mendelian error rate > 5 % were excluded. SNP were removed if they were duplicated or had an overall missing rate > 5 %, or had a minor allele frequency less than 1 %. All SNP were required to pass the Hardy–Weinberg equilibrium test. The 1000 Genomes phase 3 reference panels data were used as the main imputation reference panel, and gene phasing was conducted using shapeit (V2)(Reference Liu, Zhang and Graham24). Imputation SNP were removed if they had a minor allele frequency < 1 %, an overall missing rate > 5 %, or info < 0·7.
A polygenetic risk score (PRS) for longevity was generated using eleven SNP extracted from a previous genome-wide association study (online Supplementary Table 1)(Reference Liu, Song and Li19). A cohort study of 15 651 Chinese individuals reported 1881 SNPs in multiple genome-wide association studies of longevity and replicated those SNP that were significantly associated with longevity and mortality. A classic weighted method was applied to calculate the PRS. Each SNP was recoded as 0, 1 or 2 according to the number of risk alleles and then multiplied by the risk estimate (β coefficient) on longevity obtained from the previous genome-wide association study to calculate the PRS: PRS = (β1 × SNP1) + (β2 × SNP2) +…+ (β11 × SNP11) × (11/sum of the β coefficients). A higher score indicated a higher genetic predisposition to longevity. Participants were divided into low-risk and high-risk groups based on the median PRS.
Statistical analysis
Mean and standard deviation were used to describe continuous variables, while frequency and proportion were calculated for categorical variables for baseline characteristics according to different quartiles of TMAO, respectively. TMAO levels were normal distributed, so no more transformation was made for TMAO levels. For the primary analysis, cox regression model was used to analyse the association of TMAO level with both all-cause mortality and cause-specific mortality. The proportionality assumption was tested, and the random Schoenfeld residuals support the proportionality assumption. HR and 95 % CI were calculated, using the lowest quartile of TMAO as the reference group. In model 1, we adjusted for socio-demographical factors, including age (continuous) and sex (females/males). Family structure was included in the model as random effects terms using the frailty function in survival analysis. In model 2, prevalent CVD (yes/no), prevalent diabetes (yes/no) and the use of antidiabetic drugs (yes/no) were added as covariates in the basis of model 1. In model 3, intake of meat products (continuous) and BMI (continuous) were further adjusted for. Linear trend tests were performed by including the median value of each category of TMAO level as a continuous variable in the models. Moreover, in the sensitivity analysis, healthy lifestyle score, waist circumstance and chronic diseases were further adjusted to test the robustness of results.
Subgroup analyses were conducted based on potential modifiers, including age, sex, BMI, prevalent hypertension, prevalent CVD or the use of antidiabetic drug. These analyses aimed to determine whether these factors modified the association between TMAO and all-cause mortality. HR and 95 % CIs were calculated for each interquartile range of TMAO. In addition, we assumed that PRS for longevity might modify the association between TMAO and mortality risk. Therefore, we also investigated the effect modification of PRS for longevity on the association between TMAO and mortality risk. The product terms between TMAO and each prior potential modifiers or PRS were included in separate multivariable models to test for the potential effect modifications on mortality. The P value for the product term of the exposure variable and the interaction factor was used to assess whether there was a significant interaction between the two terms.
All the analysis were conducted using R software (version 4.2.0). P < 0·05 was considered statistically significant with a two-sided hypothesis.
Results
Characteristic of participants
Table 1 presents the baseline characteristics of participants according to TMAO levels. A total of 500 participants (aged 56·4 (sd 12·8) years) with TMAO data were included in the analysis. Of those, 45·8 % of the participants were females. The average concentration of TMAO was 4·1 (sd 3·4) µmol/l. Participants with higher TMAO levels were more likely to be overweight or obesity and less likely to use antidiabetic drugs. At baseline, 303 CVD were identified, including 192 stroke and 214 coronary heart disease. During a median follow-up of 7·38 years, a total of thirty-eight participants died. Among them, twenty deaths were due to CVD causes and nineteen deaths were from stroke.
Baseline characteristic of the participants according to trimethylamine N-oxide quartiles (Mean values and standard deviations; numbers and percentages)
TMAO, trimethylamine N-oxide.
Association of trimethylamine N-oxide with all-cause mortality
In the model adjusted for age, sex and family ID, the HR (95 % CI) of all-cause mortality was 1·30 (0·44, 3·89), 1·75 (0·65, 4·76), and 2·39 (0·91, 6·25), respectively, in higher quartile groups of TMAO levels compared with the lowest quartile group (P trend = 0·052). After adjusting for prevalent hypertension, prevalent CVD and the use of antidiabetic drugs, the HR (95 % CI) were 1·34 (0·44, 4·08), 1·67 (0·60, 4·65) and 2·48 (0·93, 6·60) (P trend = 0·047). In model 3, which further adjusted for BMI and meat intake, the highest TMAO group showed a 145 % higher risk of all-cause mortality compared with the lowest TMAO group (Figure 1).
Risk of all-cause mortality associated with trimethylamine N-oxide quartiles. Model 1 adjusted for age (continuous) and sex (females/males). Family ID was added in the model as random effects terms. Model 2 additionally adjusted for prevalent cardiovascular disease (yes/no), prevalent hypertension (yes/no) and the use of antidiabetic drugs (yes/no) in the basis of model 1. Model 3 additionally adjusted for BMI (continuous) and intake of meat products (continuous) in the basis of model 2.
Association of trimethylamine N-oxide with cause-specific mortality
We also explored the association between TMAO concentration and cause-specific mortality. No significant association was observed between TMAO and CVD-related mortality (online Supplementary Figure 2). In contrast, we observed an association between TMAO concentration and stroke mortality. In the age, sex and family ID-adjusted model, the HR (95 % CI) of stroke mortality was 2·24 (0·41, 12·26), 2·47 (0·48, 12·77) and 4·42 (0·94, 20·88), respectively, in higher quartile groups of TMAO levels compared with the lowest quartile group (P trend = 0·044). After adjusting for prevalent hypertension, prevalent CVD and the use of antidiabetic drugs, the HR (95 % CI) was 1·91 (0·38, 9·50), 1·95 (0·41, 9·35) and 4·32 (0·96, 19·51) (P trend = 0·018). In model 3, which further adjusted for BMI and intake of meat products, the HR (95 % CI) was 1·93 (0·40, 9·39), 1·91 (0·41, 8·96) and 4·16 (0·94, 18·52) in higher groups compared with the lowest group of TMAO (P for trend = 0·046, Figure 2). Additionally, in the sensitivity analysis, the results remained largely unchanged with further adjustment for healthy lifestyle score and waist circumstance (online Supplementary Figure 3).
Risk of stroke-cause mortality associated with trimethylamine N-oxide quartiles. Model 1 adjusted for age (continuous) and sex (females/males). Family ID was added in the model as random effects terms. Model 2 additionally adjusted for prevalent cardiovascular disease (yes/no), prevalent hypertension (yes/no) and the use of antidiabetic drugs (yes/no) in the basis of model 1. Model 3 additionally adjusted for BMI (continuous) and intake of meat products (continuous) in the basis of model 2..
Subgroup analyses
Associations of TMAO levels with all-cause mortality were not modified by age, sex, BMI, waist circumstance, prevalent hypertension, prevalent CVD or the use of antidiabetic drug (P interaction > 0·05, Table 2). In addition, we tested the effect modification of PRS of longevity and the results showed that PRS modified the association of TMAO with all-cause mortality (P interaction = 0·008, Table 3). We also conducted subgroup analyses according to high or low PRS in all three models. All three models provided robust evidence that PRS showed significant effect modification on the association of TMAO with all-cause mortality. The HR (95 % CI) of all-cause mortality for per interquartile range TMAO level was 2·00 (1·06, 4·57) among participants with low PRS and 1·00 (0·71, 1·40) among those with high PRS (Table 3).
Subgroup analyses of risk of all-cause mortality associated with serum trimethylamine N-oxide levels (Hazard ratios and 95 % confidence intervals)
TMAO, trimethylamine N-oxide; HPT, hypertension.
TMAO is used as a continuous variable in the regression model, and the unit of TMAO is µmol/l. Models were adjusted for age (continuous), sex (females/males), intake of meat products (continuous), BMI (continuous), prevalent CVD (yes/no), prevalent hypertension (yes/no) and the use of antidiabetic drugs (yes/no). Family ID was added in the model as random effects terms.
Risk of all-cause mortality associated with serum trimethylamine N-oxide levels stratified by polygenetic risk score (Hazard ratios and 95 % confidence intervals)
TMAO, trimethylamine N-oxide.
TMAO is used as a continuous variable in the regression model, and the unit of TMAO is µmol/l. Model 1 adjusted for age (continuous) and sex (females/males). Family ID was added in the model as random effects terms. Model 2 additionally adjusted for prevalent cardiovascular disease (yes/no), prevalent hypertension at baseline (yes/no) and the use of antidiabetic drugs (yes/no) in the basis of model 1. Model 3 additionally adjusted for body mass index (continuous) and intake of meat products (continuous) in the basis of model 2.
Discussion
In this prospective family-based cohort study, we observed that higher serum TMAO levels were associated with increased risks of all-cause and stroke mortality, but not CVD mortality. The associations were modified by genetic susceptibility to longevity, where the adverse effect of TMAO on mortality risk was more prominent in individuals with low genetic risk of longevity.
The estimates of TMAO with all-cause mortality are comparable with previous meta-analyses(Reference Heianza, Ma and Manson14). Heianza et al included a total of nineteen prospective studies and found that higher concentrations of TMAO were associated with a pooled relative risk of 1·63 (95 % CI: 1·36, 1·95) for all-cause mortality(Reference Heianza, Ma and Manson14). In addition, a recent meta-analysis indicated that being in the highest category of TMAO increased the risk of all-cause mortality by 47 %(Reference Farhangi16). Another meta-analysis enrolled 26 167 subjects from seventeen clinical studies showed that elevated TMAO levels were associated with a higher incidence of all-cause mortality(Reference Schiattarella, Sannino and Toscano13). Notably, we did not observe association between TMAO and CVD mortality in the current analysis. The results on the link between TMAO and CVD mortality were inconsistently reported. Several studies have established an association between TMAO concentration and CVD mortality(Reference Qi, You and Li15). However, there were several inconsistencies regarding the association between TMAO and increased CVD mortality in large population-based studies(Reference Wang, Li and Wang12,Reference Nagata, Wada and Tamura25) . For instance, a study included 6785 participants from the Multi-Ethnic Study of Atherosclerosis found that the association between TMAO and CVD mortality was not statistically significant with further adjustment for estimated glomerular filtration rate(Reference Wang, Li and Wang12). Although the present study failed to show any correlation between TMAO and the risk of CVD-related death, interestingly, a significant association was observed for stroke mortality. Compelling evidence has suggested that TMAO was a risk factor for stroke incidence and mortality(Reference Farhangi16,Reference Nam26,Reference Haghikia, Li and Liman27) . However, considering the low number of stroke deaths and limited statistical power in the current analysis, the results should be interpreted with caution.
TMAO is a gut microbiota-related metabolite generated through gut microbiota-dependent production of trimethylamine from nutrient precursors and its subsequent oxidation by liver enzyme FMO3(Reference Wang, Klipfell and Bennett4,Reference Koeth, Wang and Levison5,Reference Tang, Wang and Levison28) . Red meat is rich in choline and carnitine, which are important nutrient precursors involved in trimethylamine and TMAO generation(Reference Wang, Klipfell and Bennett4,Reference Koeth, Wang and Levison5) . An intervention study showed that chronic consumption of red meat can increase systemic TMAO levels(Reference Wang, Bergeron and Levison29). The current analysis and previous epidemiological and experimental studies have suggested that TMAO might lead to an increased risk of mortality(Reference Wang, Klipfell and Bennett4,Reference Koeth, Wang and Levison5,Reference Zhu, Gregory and Org8,Reference Schiattarella, Sannino and Toscano13–Reference Qi, You and Li15,Reference Tang, Wang and Levison28,Reference Meyer and Shea30) . These findings indicated that high TMAO levels are likely major pathways linking red meat to mortality, beyond other mechanisms, which has implications for healthy diet interventions in public health practice.
TMAO concentrations might be associated with mortality risk through several mechanisms, including atherosclerosis, thrombosis, hyperglycaemia and inflammation(Reference Zhen, Zhou and He31). Animal studies showed that supplementation with TMAO reduced in vivo reverse cholesterol transport, augmented macrophage cholesterol accumulation and increased atherosclerosis(Reference Wang, Klipfell and Bennett4,Reference Koeth, Wang and Levison5) . Additionally, Brunt et al used experimental approaches in mice and humans and found that TMAO promoted endothelial dysfunction via oxidative stress(Reference Brunt, Gioscia-Ryan and Casso32). TMAO was also related to platelet hyperreactivity and thrombosis process(Reference Zhu, Gregory and Org8). Furthermore, TMAO might activate nucleotide-binding domain, leucine-rich-containing family, pyrin domain-containing-3 inflammasome and nuclear factor-κB (NF-κB) signals, thereby inducing vascular inflammation(Reference Zhang, Li and Yang33,Reference Seldin, Meng and Qi34) . Experimental and animal studies indicated that TMAO impaired glucose tolerance(Reference Gao, Liu and Xu35,Reference Dumas, Rothwell and Hoyles36) and activated PERK to promote metabolic dysfunction(Reference Chen, Henderson and Petriello10). In addition, TMAO has been associated with renal disease, neurodegenerative disease and cancer, all of which contribute to mortality risk. In animal studies, TMAO was showed to reduce glomerular filtration rate, renal tubulointerstitial fibrosis and elevated cystatin C levels(Reference Tang, Wang and Kennedy9,Reference Gupta, Buffa and Roberts37) . TMAO has also been hypothesised to promote ageing-related cognitive decline and carcinogenesis through inflammatory and oxidative stress(Reference Li, Ke and Zhan38,Reference Brunt, LaRocca and Bazzoni39) . However, we were unable to test direct associations of TMAO with chronic kidney disease, dementia and cancer due to limited data in the current analysis. Therefore, further studies are needed to confirm these hypotheses.
We found that sex, age or BMI did not modify the association between TMAO and mortality risk. However, intriguingly, there was a statistically significant interaction between TMAO and genetic risk of longevity with all-cause mortality. The association between TMAO and mortality risk was stronger in participants with lower longevity PRS compared with higher longevity PRS. The findings were in line with prior evidence that genetic susceptibility might modify the relations between environmental factors and human health outcomes(Reference Heianza and Qi40–Reference Wang, Heianza and Sun42). Additionally, individuals with a higher genetic predisposition to mortality should pay more attention to the consumption of animal foods including red meat. To be noted, the PRS of longevity was calculated based on genetic variants that constituted only a small proportion of the longevity variation. Therefore, results of the gene–TMAO interaction analysis need further validation.
To the best of our knowledge, the current study is the first prospective study to assess associations between TMAO levels and risks of all-cause, CVD and stroke mortality, with considering the gene–environment interactions. The novel finding on the interaction between TMAO levels and genetic susceptibility to death may prompt more appropriate prevention strategies by considering gut microbiota-derived metabolites and genetics together. However, several potential limitations should be considered. First, although a range of covariates have been adjusted, residual confounding may still exist since some confounders such as prevalent kidney disease history were not considered due to a lack of data. Second, due to the limited sample size and cases, the results should be interpreted with caution, as they may only provide preliminary insights for future studies, especially in stratified analyses. Larger studies are needed to validate these findings and clarify the underlying mechanisms. Third, only a single measurement of TMAO was available in the current study. Further studies with repeated measurements are needed to confirm the findings. Fourth, energy intake information was not available in the current study. Future studies with more detailed dietary data are needed to further clarify the association between TMAO and mortality. Finally, the present study was based on northern Chinese populations, thus the generalisation of the gene–TMAO interaction results to other populations should be interpreted with caution.
Conclusions
The study indicates that serum TMAO level is potentially associated with long-term mortality risk in rural areas of northern China, especially for stroke-related deaths. It provides novel evidence that genetic variations might modify the association, highlighting the importance of considering genetic modifications in epidemiological studies. Future studies with large sample size are encouraged to provide more robust findings and uncover the underlying mechanisms.
Supplementary material
For supplementary materials referred to in this article, please visit https://doi.org/10.1017/S0007114525104224
Acknowledgements
This work was supported by National Natural Science Foundation of China (Grant No. 82204135, 81872695, 81230066, 82473716), Beijing Municipal Natural Science Foundation (7232237), the China Postdoctoral Science Foundation (Grant No. BX2021021, 2022M710249), Special Fund for Health Scientific Research in Public Welfare (Grant No. 201502006) and the China Cohort Consortium (Grant No. CCC2020001).
T.W., Y.W, Y.H. and M.W. designed the entire research framework and obtained research funding. M.W. and H.P. wrote the first draft manuscript and performed data analysis. R.Y., H.Y., T.H., X.Q., J.L. and D.C. participated in the experiments and revised the manuscript. All the authors agreed to submit the final version.
The authors declare no conflict of interest.
