Assessment of Causal Direction Between Gut Microbiota–Dependent Metabolites and Cardiometabolic Health: A Bidirectional Mendelian Randomization Analysis
Abstract
We examined the causal direction between gut microbiota–dependent metabolite trimethylamine N-oxide (TMAO) or its predecessors and cardiometabolic diseases, such as risk of type 2 diabetes mellitus (T2DM), coronary artery disease (CAD), myocardial infarction (MI), stroke, atrial fibrillation (AF), and chronic kidney disease (CKD). We used genetic variants as instruments to test the causal associations. Genetically predicted higher TMAO and carnitine were not associated with higher odds of T2DM, AF, CAD, MI, stroke, and CKD after Bonferroni correction (P ≤ 0.0005). However, we observed that genetically increased choline showed a suggestive association with higher risk of T2DM (odds ratio 1.84 [95% CI 1.00–3.42] per 10 units, P = 0.05). In contrast, genetically predicted higher betaine (0.68 [0.48–0.95] per 10 units, P = 0.023) was suggestively associated with a lower risk of T2DM. We observed a suggestive association of genetically increased choline with a lower level of body fat percentage (β ± SE −0.28 ± 0.11, P = 0.013) but a higher estimated glomerular filtration rate (0.10 ± 0.05, P = 0.034). We further found that T2DM (0.130 ± 0.036, P < 0.0001) and CKD (0.483 ± 0.168, P = 0.004) were causally associated with higher TMAO levels. Our Mendelian randomization findings support that T2DM and kidney disease increase TMAO levels and that observational evidence for cardiovascular diseases may be due to confounding or reverse causality.
Introduction
Gut microbiota has been recently implicated as a novel endocrine organ that plays an important role in the development of cardiovascular disease (CVD) and type 2 diabetes mellitus (T2DM) through modulating blood levels of bioactive metabolites (1,2). There is a growing appreciation that a gut microbiota–dependent metabolite, trimethylamine N-oxide (TMAO), formed from ingested choline, betaine, and carnitine in humans (3,4), is predictive of cardiovascular events (1,5). Furthermore, a greater increase in TMAO, choline, and L-carnitine has been implicated in lesser improvements in insulin sensitivity (6), adiposity, and energy metabolism (7). Animal studies have shown that the TMAO meta-organismal pathway enhances the accumulation of cholesterol in macrophages and platelet hyperreactivity (1) and exacerbates impaired glucose tolerance by blocking the hepatic insulin signaling pathway (8), which promotes CVD and diabetes in animal models.
However, the results from human studies are not always consistent (9,10). Although a recent meta-analysis of 19 observational cohorts found that elevated levels of TMAO and its precursors are associated with an increased risk of major adverse cardiovascular events (11), available evidence from observational studies mainly relies on self-reported information and is susceptible to confounding or reverse causation bias. Therefore, the causality of these observations remains unclear. Documentation of such causality can inform potential lifestyle or drug treatment targets for prevention of cardiometabolic disease.
The Mendelian randomization (MR) approach, which uses genetic variants as an instrumental variable in epidemiological study, has been widely accepted to explore the potential causal effect of exposure on disease (12–16). Such an MR method is analogous to a randomized controlled trial, where genetic alleles are randomly assorted during conception, and is less likely to be affected by confounding or reverse causation (13,17). Therefore, we performed a bidirectional MR analysis to examine the causal direction between the gut microbiota–dependent metabolite TMAO or its predecessors (e.g., choline, betaine, and carnitine) and cardiometabolic diseases (e.g., T2DM, coronary artery disease [CAD], myocardial infarction [MI], stroke, atrial fibrillation [AF], and chronic kidney disease [CKD]) and related traits using summary data from genome-wide association studies (GWAS).
Research Design and Methods
Study Design
The MR approach must satisfy three assumptions (18) (Fig. 1): 1) the genetic variant selected as the instrumental variable is associated with the gut microbiota–dependent metabolite TMAO and its predecessors; 2) the genetic variant is not associated with any unmeasured confounders of the gut microbiota–dependent metabolites and the cardiometabolic relationship; and 3) the genetic variant is associated with cardiometabolic events only through gut microbiota–dependent metabolites, not through other pathways.
Schematic representation of an MR analysis. MR can be used to test the hypothesis that exposure (gut microbiota–related metabolites) causes T2DM, CAD, MI, AF, stroke, and CKD. The three assumptions of MR are as follows: 1) Genetic variants must be associated with gut microbiota–related metabolites; 2) genetic variants must not be associated with confounders; and 3) genetic variants must influence disease outcomes only through gut microbiota–related metabolites, not through other pathways.
Gut Microbiota–Dependent Metabolites and Instrumental Variable Selection
The gut microbiota–dependent metabolites include TMAO, choline, betaine, and carnitine. We searched PubMed for GWAS of the gut microbiota–dependent metabolites and cardiometabolic diseases and identified genetic variants for each metabolite (19) and disease (20–24). For each metabolite, we selected genetic variants (single nucleotide polymorphisms [SNPs]) at thresholds for suggestive genome-wide significance (P < 5 × 10−5) from the published GWAS (19) (Supplementary Table 1). Details on GWAS where we extracted summary-level data are presented in Table 1 and Supplementary Table 1. Contributing studies received ethical approval from their respective institutional review boards. Informed consent was obtained from all participants of contributing studies.
Description of modifiable factors or cardiometabolic disease
Cardiometabolic Diseases and Data Sources
For disease outcomes, summary-level data were extracted from the Diabetes Genetics Replication and Meta-analysis (DIAGRAM) Consortium for T2DM (n = 149,821) (20); the Atrial Fibrillation Consortium (AFGen) for AF (21); the Coronary Artery Disease Genomewide Replication and Meta-analysis (CARDIoGRAM) plus the Coronary Artery Disease (C4D) Genetics (CARDIoGRAMplusC4D) Consortium for CAD (60,801 patients and 123,504 control subjects) and MI (43,676 patients and 128,197 control subjects), respectively (22); the National Institute of Neurological Disorders and Stroke Genetics Network (SiGN) and International Stroke Genetics Consortium (ISGC) for stroke (37,792 patients and 397,209 control subjects) (23); and the Chronic Kidney Disease Consortium (CKDGen) (n = 133,814) for CKD (24).
Patients with T2DM were diagnosed according to 1999 World Health Organization criteria of fasting plasma glucose concentration ≥7.0 mmol/L or 2-h plasma glucose concentration ≥11.1 mmol/L, by report of diabetes medication use, or on the basis of medical record review (25). Ascertainment of AF in UK Biobank included samples with one or more of the following codes: 1) noncancer illness code, self-reported (1471, 1483); 2) operation code (1524); 3) diagnoses main/secondary (ICD-10 codes I48, I48.0–4, I48.9); 4) underlying (primary/secondary) cause of death (ICD-10 codes I48, I48.0–4, I48.9); 5) diagnoses main/secondary (ICD-9 code 4273); and 6) operative procedures main/secondary (OPCS codes K57.1, K62.1–4) (21). CKD was classified in patients with an estimated glomerular filtration rate based on creatinine (eGFRcreat) <60 mL/min/1.73 m2 (24). Cases of CAD were determined with a broad definition, including MI, acute coronary syndrome, chronic stable angina, and coronary artery stenosis >50% (22).
Cardiometabolic Traits and Data Sources
We searched PubMed for GWAS of cardiometabolic traits. Summary-level data were extracted from the Genetic Investigation of Anthropometric Traits (GIANT) Consortium for adiposity indicators such as BMI (26), body fat (27), and waist-to-hip ratio–adjusted BMI (28); the Meta-Analyses of Glucose and Insulin-Related Traits Consortium (MAGIC) (25,29) for glycemic traits such as fasting glucose, log-transformed fasting insulin, 2-h glucose, log-transformed HOMA of β-cell function (HOMA-B), log-transformed HOMA of insulin resistance (HOMA-IR), log-transformed proinsulin, and hemoglobin A1c (HbA1c); the Global Lipids Genetics Consortium (GLGC) (n = 188,577) for lipids such as HDL (30), LDL (30), total cholesterol (30), and triglycerides (30); the CKDGen (n = 133,814) for kidney function indicators such as eGFRcreat, which was estimated using the four-variable MDRD equation, and eGFR based on cystatin C (eGFRcys), which was estimated as 76.7 × (serum cystatin C) − 1.19 (24); and GWAS summary data for leptin (31) and heart rate (32).
Statistical Analysis
Linkage Disequilibrium Assessment and Pleiotropy Assessment
To verify that the SNPs selected in this study met assumptions 1 and 2, we examined the genetic association with each metabolite, measured linkage disequilibrium (LD) among all the SNPs for the same metabolite (33), and selected independent genetic variants for each metabolite (13). We chose the variant with the lowest P value for association with each metabolite if genetic variants were in LD; therefore, the SNPs selected did not violate assumption 2. All SNPs for the same metabolite showed strong association (F statistic >21.10, the strength of the instrument), thus meeting assumption 1. We used MR-Egger regression to assess the presence of pleiotropic effects on cardiometabolic outcomes (18). Using the MR-Egger method, the SNP’s effect upon each metabolite is plotted against its effect upon outcomes, and an intercept distinct from the origin provides evidence for pleiotropic effects.
MR Analysis
The estimates of the causal effect of each metabolite on outcomes were analyzed using inverse-variance weighting (IVW), which provides a combined estimate of the causal estimate from each SNP. IVW is equivalent to a two-stage least squares or allele score analysis using individual-level data and is hence considered here as conventional MR (34). Results are presented as odds ratios (ORs) (95% CIs) per 10 units of genetically predicted increase in each metabolite. In addition, results for causal effects of diseases on metabolites are presented as β ± SE per each unit higher in log odds of disease. Furthermore, complementary approaches such as simple median method, weighted median method, the mode-based estimate (MBE), and MR-Egger for multiple genetic variants were used to examine causal effect (34–37). Detailed information on these MR methods has been described previously (34–37). Finally, the MR-Egger regression test was used to evaluate the pleiotropic effects (18). Using the MR-Egger regression method, the effect of an instrumental variable on the exposure is plotted against its effect on the outcome, and an intercept distinct from the origin provides evidence for pleiotropic effects. The slope of the MR-Egger regression can provide pleiotropy-corrected causal estimates (18). In the present MR analyses, median weighted, MBE, and MR-Egger methods were considered as sensitivity analyses for MR investigations with multiple genetic variants (34–36). Power calculations for MR were conducted on the basis of mRnd (http://cnsgenomics.com/shiny/mRnd).
Analyses were performed using R 3.2.3 software (R Project for Statistical Computing). The threshold of statistical significance was P ≤ 0.0005 (0.05/100) after Bonferroni correction. P ≤ 0.05 but above the Bonferroni-corrected significance threshold was considered as suggestive of evidence for a potential association.
Results
Characteristics for Selected SNPs
The characteristics of the selected SNPs for each gut microbiota–dependent metabolite are presented in Table 1 and Supplementary Table 1. To examine assumptions 2 and 3, we tested whether any of the selected SNPs were influenced by LD. We chose the variant with the lowest P value for association with each metabolite if genetic variants were in LD. Our genetic analysis showed that 15.4% of betaine, 17.1% of carnitine, 8.0% of choline, and 9.6% of TMAO were explained by their SNPs (Supplementary Table 2). MR power calculation showed that we had 87%, 84%, 78%, and 81% power to test significant (P < 0.05) causal effects (OR 1.3) of betaine, carnitine, choline, and TMAO, respectively, on cardiometabolic events.
Gut Microbiota–Dependent Metabolites and Cardiometabolic Disease
Genetically predicted higher TMAO was not associated with a higher odds of T2DM (OR 0.96 [95% CI 0.59–1.57], P = 0.863), AF (0.99 [0.79–1.26], P = 0.961), CAD (1.00 [0.70–1.43], P = 0.986), MI (1.08 [0.73–1.60], P = 0.708), stroke (0.94 [0.53–1.65], P = 0.83), and CKD (1.08 [0.59–1.99], P = 0.794) per 10 units after Bonferroni correction (Fig. 2). Likewise, genetically predicted higher carnitine was also not associated with each cardiometabolic disease. However, we observed that genetically increased choline was suggestively associated with higher risk of T2DM (1.84 [1.00–3.42] per 10 units, P = 0.05) but not with cardiovascular events and CKD. In contrast, genetically predicted higher betaine (0.68 [0.48–0.95] per 10 units, P = 0.023) was suggestively associated with a lower risk of T2DM (Fig. 2).
MR of gut microbiota–related metabolites and risk of T2DM, CAD, MI, AF, stroke, and CKD.
In addition, the associations were consistent in sensitivity analyses that used the simple median and weighted median methods (Supplementary Table 3). The MR-Egger regression test was used to examine the presence of a pleiotropic effect. The intercepts (SE) from MR-Egger regression testing of each outcome were centered at the origin with a CI including the null, suggesting that the results were not influenced by pleiotropy (Supplementary Table 3).
Gut Microbiota–Dependent Metabolites and Cardiometabolic Traits
Genetically predicted higher TMAO and betaine were not associated with any cardiometabolic factors after Bonferroni correction (Table 2). However, we observed a suggestive association (P < 0.05) of genetically increased choline with a lower level of body fat percentage (β ± SE −0.28 ± 0.11, P = 0.013) but a higher eGFRcys (0.10 ± 0.05, P = 0.034). Additionally, genetically predicted higher carnitine was suggestively associated with a lower level of triglycerides (−0.17 ± 0.07, P = 0.012) (Table 2). Results were similar in sensitivity analyses that used the simple median and weighted median methods (Supplementary Table 4). The results were not influenced by directional pleiotropy (P > 0.05). We further examined the genetic association of TMAO SNPs with potential confounders, such as alcohol intake and dietary factors (Supplementary Table 5). We did not find that TMAO SNPs were associated with any potential confounders, thus validating assumption 2.
MR analysis of gut microbiota–dependent metabolite and cardiometabolic traits
Causal Effects of Cardiometabolic Diseases on Gut-Dependent Metabolites
We further examined the causal effects of cardiometabolic diseases on gut-dependent metabolites. We found that T2DM was causally associated with lower betaine (β ± SE −0.111 ± 0.035, P = 0.002) and higher TMAO levels (0.13 ± 0.036, P < 0.0001) per 1-unit higher log odds. CKD was also causally associated with higher TMAO levels (0.483 ± 0.168, P = 0.004) per 1-unit higher log odds (Table 3 and Supplementary Table 6).
MR analysis of cardiometabolic diseases on gut microbiota–dependent metabolites
Discussion
In the present MR study, we did not observe a significant association of genetically predicted higher gut microbiota–dependent metabolite TMAO and its predecessor with cardiometabolic disease and related traits. However, we observed a suggestive association of genetically increased choline with a higher risk of T2DM. Furthermore, genetically increased choline showed a suggestive association with a lower body fat percentage but a higher level of eGFR, whereas genetically predicted higher carnitine was suggestively associated with a lower level of triglycerides. We further found that T2DM and CKD were causally associated with higher TMAO levels. Our findings support that T2DM and CKD increase TMAO levels and that the previously observed association of TMAO and CVD may be due to confounding or reverse causality.
Gut microbiota alteration has been implicated in the development of diabetes (38). Previous observational findings suggested that higher plasma TMAO is associated with an increased odds of newly diagnosed T2DM (39). Animal studies showed that TMAO exacerbates impaired glucose tolerance and hyperglycemia through causing inflammation or blocking the hepatic insulin signaling pathway in adipose tissue (8). Likewise, a weight loss diet intervention study demonstrated that a greater increase in TMAO is related to lesser improvements in the glycemic traits (6). However, previous studies have yielded inconsistent associations between plasma TMAO and T2DM (39–41). Our MR analysis also did not confirm the causal effect of genetically increased TMAO on risk of T2DM. Interestingly, we found that T2DM increases TMAO levels. In addition, a recent genome-wide scan suggested that human gut microbiota is influenced by a host gene that has been associated with diabetes (42), suggesting that a TMAO-T2DM association may share a genetic basis rather than a causal relationship. We cannot exclude the possibility that the suggestive genome-wide SNP (P < 5.0 × 10−5) used in the MR analysis might introduce a weak instrument bias (19), thus diluting the association. Future investigation for identifying genome-wide significant SNPs for TMAO and its causal role in the development of T2DM is required.
Furthermore, a recent prospective study indicated that a higher intake of phosphatidylcholine, the precursor to the generation of TMAO, is independently associated with an increased risk of T2DM in three U.S. populations (43). Interestingly, our MR results showed a suggestive association of genetically increased choline with a higher risk of T2DM. Our findings were supported by a previous weight loss diet intervention study that demonstrated that greater decreases in choline and L-carnitine are significantly associated with greater improvements in fasting insulin concentrations and insulin resistance (6), underscoring the importance of changes in choline and L-carnitine in improving insulin sensitivity (6).
Our findings did not corroborate the results from a previous meta-analysis of data from 19 prospective studies, which provided quantitative pooled estimates of the associations of circulating TMAO level with the incidence of cardiovascular events (11). It was estimated that participants with high TMAO levels had a 62% increased risk for the development of cardiovascular events compared with those with low TMAO levels (11). However, our MR results showed that genetically increased TMAO is not associated with a higher risk of cardiovascular events, such as CAD, MI, stroke, and AF, suggesting that observational evidence might result from confounding and reverse causality. Although we measured LD among all selected SNPs using northern Europeans from Utah samples from the 1000 Genomes Project (33), we cannot exclude the possibility that our results might be affected by unmeasured confounders. Furthermore, the MR-Egger regression showed that our results are not influenced by pleiotropy (18). Nevertheless, it is possible that the results reflect a shared genetic basis between gut microbiota–related metabolites and CVD rather than a causal relationship.
Despite strong observational evidence for an association of predecessors of TMAO, such as carnitine, choline, or betaine, with risk of CAD, MI, stroke, and AF, our results disagree with an earlier meta-analysis of observational studies showing a positive association of L‐carnitine, choline, or betaine with risk of cardiovascular events (11). These null findings suggest that our findings do not support a causal role of predecessors of TMAO in the development of CVD, but we have to acknowledge that more variants, especially suggestive genome-wide SNPs, may lead to greater pleiotropy and dilute the association in our analysis. However, we found that genetically predicted higher carnitine was suggestively associated with a lower level of triglycerides. The potential mechanism may be that dietary carnitine is metabolized in the liver by intestinal bacteria to produce TMAO by the enzyme flavin monooxygenase 3 (3,4), which is reported to be a key integrator of hepatic cholesterol and lipid metabolism (1,5).
Interestingly, an observational study has reported that TMAO, cleared by the kidney, is elevated in patients with impaired renal function (44) and portends poorer long-term survival (45), suggesting that circulating TMAO might play a role in kidney dysfunction (9,44–46). In the present MR study, our findings did not support a causal role of TMAO in kidney dysfunction. Surprisingly, we found that genetically increased choline showed a suggestive association with a higher level of eGFR. Importantly, we demonstrate that kidney disease increases TMAO levels. Given that the gut microbiota–mediated metabolism of choline and carnitine each ultimately produces TMAO (3,4), we speculate that there is a threshold in the observed association between choline and kidney function and that the optimal level of choline may be beneficial for improving kidney function. However, chronic high dietary exposure to choline or carnitine that increases TMAO appears to be toxic and directly contributes to progressive renal fibrosis and dysfunction (45); further investigation on the mechanism is thus required.
This study has several strengths. First, we have systematically assessed for the first time in our knowledge the causal role of gut microbiota–dependent metabolites in the development of T2DM, CAD, MI, stroke, AF, and CKD. Second, the present MR analysis used summary-level data from large GWAS, which might avoid bias from reverse causation and reduce confounding. Furthermore, the large sample size from GWAS summary statistics has sufficient power for reliable and lifelong causal estimation. Third, the consistent causal estimation across five methods, such as the conventional IVW, simple median, weighted median, MBE, and MR-Egger methods, suggests robustness of our findings.
However, several limitations merit consideration. First, we assumed that the associations between continuous metabolites and outcomes are linear, but our findings might suggest that there is a threshold in observed association. Therefore, further investigation is warranted on the causal role of gut microbiota–dependent metabolites in the development of cardiometabolic disease. Second, we used the suggestive genome-wide SNPs for the metabolites; thus, we cannot exclude the possibility that our findings might have been affected by weak instrument bias, which depends on the strength of the genetic instrument through the F statistic. Third, we do not have diet or lifestyle information; therefore, we cannot test whether the instrumental variables are associated with these factors. However, we used diet/lifestyle information from UK Biobank and demonstrated that the genetic variants are not associated with confounders. Fourth, we cannot exclude the possible diet-gene or gene-environment interactions on outcomes, which may influence the observed results. Finally, completely ruling out an alternative direct causal pathway is a challenge for all MR analyses, particularly for metabolites determined by both gut microbiota and multiple genetic variants.
In summary, our findings from an MR approach support that TMAO and kidney disease increase TMAO levels and that observational evidence for gut microbiota–dependent metabolites and CVDs may be due to confounding or reverse causality. The human genome influences both gut microbiota and chronic disease; therefore, we cannot exclude the possibility that such an association may have a genetic basis rather than a causal relationship. A large-scale genome-wide scan for genetic variants of gut microbiota–related metabolites and further investigation in understanding the potential role of gut microbiota–dependent metabolites in the development of cardiometabolic disease are required.
Article Information
Acknowledgments. The authors thank Dr. Robert Gerszten (Division of Cardiovascular Medicine, Beth Israel Deaconess Medical Center, Boston, MA), Dr. Ming-Huei Chen (Population Sciences Branch, National Heart, Lung, and Blood Institute’s Framingham Heart Study, Framingham, MA), and Dr. Qiong Yang (Department of Biostatistics, Boston University School of Public Health, Boston, MA) for sharing the GWAS summary statistics of gut-dependent metabolites.
Funding. The study was supported by Peking University Start-up grant BMU2018YJ002, High-performance Computing Platform of Peking University, and Beijing Technology and Business University grant 88442Y0033.
The funding organizations had no role in the preparation of the manuscript.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. J.J., P.D., M.G., X.K., C.L., Z.L., and T.H. contributed to the statistical analysis, performed the data analysis, wrote the manuscript, critically reviewed the manuscript during the writing process, and approved the final version to be published. J.J., P.D., and T.H. designed the research. J.J., Z.L., and T.H. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Data and Resource Availability. All data used in the current study were obtained from GWAS summary statistics, which were publicly released by genetic consortia.
Footnotes
This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db19-0153/-/DC1.
- Received February 12, 2019.
- Accepted May 29, 2019.
- © 2019 by the American Diabetes Association.
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at http://www.diabetesjournals.org/content/license.
References
In this Issue
