Impact of gut microbiota on metabolic syndrome and its comprising traits: a two-sample mendelian randomization study Authors: Yaodong Zhang, Jinhai Fan Published: 2024 ABSTRACT BackgroundThe prevalence of metabolic syndrome is on the rise globally. Understanding the etiology and discovering potential treatment target have become a priority. Observational data have linked gut microbiota with metabolic syndrome and its comprising traits. However, whether these relations underlie causal effects remains unclear.MethodsUsing Inver Variance Weighted (IVW) as primary analysis method, we performed two-sample Mendelian Randomization (MR) analyses to explore the causal relationship between gut microbiota and metabolic syndrome with its comprising traits. Methods including MR-Egger regression, MR Pleiotropy RESidual Sum and Outlier (MR-PRESSO), Weighted Mode, and Weighted Median were chosen for additional MR analysis to test the robustness of MR results. Cochran’s IVW Q test and leave-one-out IVW analysis tested the heterogeneity among instrumental variables (IVs). Steiger filtering was utilized to exclude all IVs with reverse causality. Genome-wide association study (GWAS) data used in this study were all from the largest respective GWAS studies available.ResultsOut of 1172 tests, a total of 16 associations with evidence of causality were identified after sensitivity analyses, but only 3 remained after multiple testing correction. Class Melainabacteria (β = 0.02, adjusted P = 0.01) with affiliated order Gastranaerophilales (β = 0.02, adjusted P = 1.20*10− 3) and genus Eubacterium hallii (β = 0.03, adjusted P = 0.03) showed a positive effect on abdominal obesity. All effect sizes were small (abs(β) < 0.1). All causal relationships identified were unidirectional.ConclusionsGiven the study’s limitations, we found little evidence supporting a large causal effect, i.e. absolute effect size > 0.1, of gut microbial taxa abundance on metabolic syndrome and its comprising traits. This study also suggests that previously reported associations between gut microbiota and metabolic syndrome with its comprising traits may not necessarily lead to causal relations.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13098-024-01520-8. Introduction Metabolic syndrome (MetS) is marked by a cluster of comorbid conditions including elevated fasting blood glucose (FBG), low high density lipoprotein cholesterol (HDL-C), elevated triglycerides (TG), high diastolic or systolic blood pressure, and increased waist circumference (WC). The diagnosis of MetS can be established once at least three out of five aforementioned conditions are present [1]. The prevalence of MetS has surged globally in recent years, posing a huge burden on public health and medical resources worldwide [2]. MetS is a known risk factor of many metabolic diseases, including coronary artery disease, type 2 diabetes, non-alcoholic fatty liver disease. Factors like sedentary lifestyles, poor dietary habits, genetic predisposition and socioeconomic status are related to the high prevalence of MetS, yet the exact etiology of MetS remains to be elucidated [3]. The human gut microbiota (GM) plays a crucial role in various metabolic and immune functions, including host immunity, food digestion, intestinal endocrine functions, and maintaining intestinal permeability. These functions are partly mediated by byproducts of microbiota metabolism, bridging the crosstalk between gut microbiome and various organ in human body [4]. Existing literature has shown that the main pathophysiology of MetS is chronic low-grade inflammation dominated by insulin resistance [5]. On one hand, gut dysbiosis can disrupt the gut barrier and allow bacteria to enter the circulation, causing low levels of inflammation [6]. On the other hand, GM metabolism can produce short chain fatty acids (SCFAs) such as acetate, propionate and butyrate. Animal studies have shown that SCFAs can improve glucose metabolism in skeletal muscle [7, 8]. SCFAs also play an anti-inflammatory role by enhancing regulatory T-cells and strengthening the gut barrier, which reduces endotoxin-induced inflammation and insulin resistance [9, 10]. Human microbiota-associated (HMA) study, in which human feces from healthy individual and patients were transplanted to rodents, are often used to study the impact of human GM on various disease process. Dozens of HMA studies has more or less linked gut dysbiosis to traits related to metabolic syndrome-related traits, suggesting GM may play a role in the pathogenesis of MetS [11]. However, according to Walter et al., current HMA rodent models are flawed and thus conclusions from such studies were ‘implausible’. HMA study cannot guarantee adequate replication of human GM in rodents, since many human-associated microbes fail to be established with rodent hosts. Many studies inappropriately rely on small numbers of human donors and replicate their microbiomes across multiple mice, using each mouse as an independent data point. This practice, known as “pseudoreplication”, artificially inflates sample size and increases the likelihood of false-positive results [11]. Growing amount of observational studies also reported difference in composition of GM between healthy individual and patients [12–15]. However, such study is unfit for causal inference because it is prone to reverse causality and unmeasured confounding factors. Mendelian randomization (MR) uses single nucleotide polymorphism (SNP) strongly associated with certain phenotype as proxy to infer causality between an exposure (GM traits) and an outcome (MetS and its five comprising traits). It can mitigate part of biases in observational studies such as unmeasured confounding and reverse causality. Coupled with data from genome-wide association study (GWAS), MR is a suitable tool to explore the potential causal role of GM in MetS. In this study, we performed two-sample MR analyses of GM and MetS with its five components to investigate the causal effect genetically predicted GM taxa abundance on MetS and its five components. The objective is to contribute more insights into the intervention for MetS. Materials and methods Study design A brief description of this two-sample MR study design is shown in Fig. 1. In this study we investigated the causal relationship of GM on MetS. For MetS, this analysis used the National Cholesterol Education Program Adult Treatment Panel III (NCEP/ATP III) diagnostic criteria, which contain five diagnostic elements [16]. Firstly, we extracted summary statistics from meta-analysis of GWAS for MetS, hypertension, WC, FBG, HDL-C, TG to identify valid genetic instrument variables (IVs). Then, we extracted data associated with IVs from the summary statistics of the largest GWAS meta-analysis of GM. After data harmonization, we performed MR analysis using a set of methods to ascertain the causal relationship between GM and MetS with its five components. MR depends on three key assumptions: (1) genetic instruments are significantly associated with exposure of interest; (2) genetic instruments are not related to any confounding factors of the exposure-outcome association; (3) genetic instruments affect the outcome only via the exposure [17]. This study was conducted in adherence to the STROBE-MR guidelines as shown in S1 File. Detailed information about data source used in this study can be found in Table 1. Fig. 1Study design Study design Data source For GM, We employed the largest-scale GM genome-wide meta-analysis (GWAS) to date obtained from the international consortium MibioGen [18]. This project studied the genome-wide genotypes and 16 S ribosomal RNA gene sequencing of 18,340 participants in 24 cohorts, mostly of European ancestry (more than 72%). Each cohort has surveyed the gut microbiome via 16 S rRNA sequencing and genotyped their participants with full-genome SNP arrays. All the data have been processed using identical analytical pipeline. Five levels (phylum, class, order, family, genus) and 211 bacterial taxa in total were identified, including 15 unknown bacterial taxa were excluded in this study. Table 1Data source of all GWAS data used in this studyGMMetSWCHypertensionFBGTGHDL-CYear2021202220182018201220202020PopulationMostly EuropeanEuropeanEuropeanEuropeanEuropeanEuropeanEuropeanSexFemale and maleFemale and maleFemale and maleFemale and maleFemale and maleFemale and maleFemale and maleNcaseNANANA54,358NANANANcontrolNANANA408,652NANANASample size18,340461,920462,166463,01058,074441,016403,943Data sourcesMibioGenUK BiobankMRC-IEUMRC-IEUAlisa K manning et al.UK BiobankUK BiobankGWAS IDNANAukb-b-9405ukb-b-12,493ebi-a-GCST005186ieu-b-111ieu-b-109 Table 1Data source of all GWAS data used in this studyGMMetSWCHypertensionFBGTGHDL-CYear2021202220182018201220202020PopulationMostly EuropeanEuropeanEuropeanEuropeanEuropeanEuropeanEuropeanSexFemale and maleFemale and maleFemale and maleFemale and maleFemale and maleFemale and maleFemale and maleNcaseNANANA54,358NANANANcontrolNANANA408,652NANANASample size18,340461,920462,166463,01058,074441,016403,943Data sourcesMibioGenUK BiobankMRC-IEUMRC-IEUAlisa K manning et al.UK BiobankUK BiobankGWAS IDNANAukb-b-9405ukb-b-12,493ebi-a-GCST005186ieu-b-111ieu-b-109 Data source of all GWAS data used in this study For MetS, the GWAS summary statistics were obtained from Walree et al. for 461,920 Europeans [19]. Rather than running GWAS on phenotypic clustering of MetS [20], Walree et al. used structural equation modeling to model the shared genetic architectures across the five comprising trait of MetS and identified SNPs with effects on this shared dimension [19]. Total observed scale SNP heritability for MetS from Walree et al. is 0.14 (SE 0.01), which is higher that the observed scale SNP heritability of 0.09 by Lind [20]. For WC, data were obtained from the Medical Research Council Integrative Epidemiology Unit (MRC-IEU) UK Biobank GWAS pipeline, comprising a dataset of 462,166 individuals of European descent. For hypertension, data was similarly sourced from the MRCIEU UK Biobank pipeline. The dataset encompassed 54,358 cases and 408,652 controls, all of European ancestry. For HDL-C and TG, summary statistics were acquired from the UK Biobank, comprising a dataset of over 400,000 individuals of European descent [21]. For FBG, data were sourced from a database containing 58,074 individuals of European descent. The dataset was appropriately adjusted for body mass index (BMI) [22]. Detailed information about the data source of this study is shown in Table 1. Genetic variants selection criteria and data harmonization Within GWAS data of gut microbiota, 15 traits with no specific names were excluded, leaving 196 distinct bacterial traits. To ensure the authenticity and accuracy of the conclusions on the causal link between GM and MetS with related traits, we selected instrumental variables (IVs) based on the following criteria: (1) candidate SNPs must reach the genome-wide significance threshold (p < 5 × 10− 8). However, only 27 SNPs can be considered eligible IVs under this threshold, drastically reducing the statistical power, while many GM taxa are left with no IV [23–27]. While relaxing significance threshold introduces a higher risk of false positives, the nature of this study remains exploratory. Using a less stringent p-value threshold can allow us to include a great number of genetic variants, upon which sensitivity analysis is viable and statistical power can be increased. Therefore, a more lenient significance threshold (p < 1 × 10− 5) is chosen, which is frequently implemented in similar studies [23–27]; (2) to avoid linkage disequilibrium, the SNPs were only retained after clumping (r2 < 0.001 and window size = 10,000 kb), using the 1000Genomes Project Phase 3 European LD reference panel; (3) proxy SNPs with linkage disequilibrium R2 > 0.8 were found for SNPs with no match in GWAS of outcome; (4) SNPs with a minor allele frequency (MAF) less than 0.01 were removed; (5) strength of each SNP was measured by the F-statistics [28], and the variance explained was quantified using the r2 [29]; (6) To eliminate weak IVs, only those SNPs with F-statistics greater than 10 were kept. (7) palindromic SNPs were eliminated to assure the effects of SNPs on exposure corresponded to the consistent allele as the effects on outcome. Statistical analysis The main design of statistical analysis is shown in Fig. 1. We chose inverse variance weighted (IVW) method with multiple random effects for primary MR analysis as recommended by a recent guideline for MR analysis [30]. The IVW method combines the ratio estimates from each SNP with a meta-analytic approach by using the inverse of outcome variance as the weight for fitting [31]. We conducted a total of 1172 primary analysis: 196 exposures × 6 outcomes – 4 exposures /outcomes with fewer than 3 overlapping SNPs as instrument variables (IVs). We applied a Benjamini-Hochberg correction for multiple testing using a false discovery rate of 5% to reduce the propensity of false positive finding. Compared to stricter method like Holm-Bonferroni corrections, Benjamini-Hochberg correction is less conservative, which increases the likelihood of accepting false positives when a large number of tests are performed. Meanwhile, 0.05 is set as a lenient p-value threshold and any estimate with a p-value come in between these two thresholds is considered suggestive. For datasets with a p-value < 0.05 in the primary MR analysis, further sensitivity analysis was conducted. We used weighted median, weighted mode, MR-Egger regression and MR-PRESSO to test robustness of the estimate under different assumptions. Weighted mode method can estimate causality more correctly when more than 50% of the instrumental variables are invalid [32]. Likewise, weighted median method can produce robust estimate when the majority of IVs are valid [33]. MR-Egger regression makes the assumption of instrument strength independent of direct effect (InSIDE). It is often used to evaluate the existence of horizontal pleiotropy with intercept terms [34]. As a general test to the presence of outliers, we used the outlier-robust method MR-PRESSO, which is a simulation approach that can detect potential outliers in IV based on their contributions to heterogeneity. Identified potential outliers would be removed and an outlier-corrected effect size estimate would then be calculated [35]. For MR-Egger regression, if the p-value of the intercept term < 0.05, the result would be considered to suffer from horizontal pleiotropy. Furthermore, we quantified the heterogeneity among the selected SNPs with Cochran’s Q statistic and identified potential heterogeneous SNPs with the “leave-one-out” analysis omitting each instrumental SNP in turn. If the p-value of Cochran’s Q < 0.05, leave-one-out sensitivity analysis would be used to investigate the contribution of each SNP to overall heterogeneity. Additionally, we performed Steiger filtering to remove SNPs with a reverse causal relationship (from outcome to exposure). The Steiger test provides a p-value under the null hypothesis that there is no difference in variance explained [36]. All analyses were performed on R (version 4.3.3) and MR analysis was done using TwoSampleMR (version 0.5.8) [37], MRPRESSO (version 1.0) [35]. Results SNP selection After processing MibioGen GWAS summary data through genetic variants selection criteria, only 27 SNPs were found to be associated with various GM taxa at a significance level of p < 5 × 10− 8. Given the exploratory nature of this study and to increase the statistical power, we chose a more lenient threshold of p < 1 × 10− 5 for genetic instrument selection. Under p < 1 × 10− 5, we found 2129 individual SNPs associated with various gut microbiota taxa abundance. After calculating F statistics for each SNP, we found no SNP with a F statistic ≤ 10. These criteria were implemented to minimize weak instrument bias and allow the use of sensitivity analysis to assess the validity of the MR assumptions. The details of these 2129 SNPs in exposure and outcomes are shown in S1-7 Table. Primary MR analyses All identified taxa with no less than 3 IVs were retained for primary analysis. Results of all 1172 primary MR analyses conducted can be found in S8 Table. We then performed primary MR analysis using the IVW method with multiple random effects. Out of 1172 associations from primary analysis, a total of 82 associations passed the threshold of p < 0.05 (S10 Table). To reduce the likelihood of interpreting false positives as significant findings when performing over a thousand tests, Benjamini-Hochberg correction was applied to control the false discovery rate. 11 out of 1172 associations remained after multiple testing correction with a false discovery rate (FDR) < 0.05 (S9 Table). Fig. 2Forest plot of nominally significant IVW MR results (p < 0.05). ** indicates the causal estimate from IVW remained significant after adjusting for multiple comparison Forest plot of nominally significant IVW MR results (p < 0.05). ** indicates the causal estimate from IVW remained significant after adjusting for multiple comparison Exploration of promising findings and sensitivity analysis One key assumption of MR analysis is that genetic instruments of exposure doesn’t affect outcome via other mechanisms. When IVs have an effect on outcome outside of its effects on exposure, this phenomenon is called horizontal pleiotropy [35]. To avoid horizontal pleiotropy, we performed additional robust MR analysis on all 82 potentially causal relationships (primary analysis P-value <0.05). We chose three different methods with different assumptions about the underlying causal relationship: weighted median, weighted mode, MR-PRESSO. Consistency across different methods further supports association. We also used MR-Egger intercept to assess the robustness to horizontal pleiotropy, finding no significant horizontal pleiotropy except for phylum Actinobacteria on MetS (P = 0.018) (Table S11). MR estimates without validation from at least three robust analyses (weighted median P < 0.05, weighted mode P < 0.05, MR-PRESSO outlier-adjusted test P < 0.05) were removed. To further evaluate the robustness of individual SNPs, we calculated Cochran’s Q statistic for all 82 associations that passed the primary analysis. A total of 18 associations had a p-value of Q statistic < 0.05. Leave-one-out IVW MR analysis was performed for these 18 associations (Figure S1-16). For the relationship between genus Ruminococcus gauvreauii and triglyceride, after dropping the potentially heterogenic SNP rs2105937, the effect size estimate became nominally significant (p < 0.05), but this statistical significance disappeared upon further sensitivity analysis. All genetic instruments didn’t display reverse causality as indicated by Steiger filtering tests. Out of the 82 MR estimates, only 1 remained consistent across all robust analyses: order Bifidobacteriales (OR = 0.96, 95%CI 0.94–0.99, P = 1.06*10− 2) negatively associates with WC (Fig. 2C). Yet the primary IVW estimate of this association ceased to be significant after multiple testing correction (adjusted P = 0.55). Notably, for the relationship between order Bifidobacteriales and WC, heterogeneity of effect size estimate was significant (Q_pval = 0.001). Through leave-one-out analysis, we found out that rs182549 accounts for a large portion of heterogeneity (Figure S1). After the removal of rs182549, the effect size estimate ceased to be significant (Fig. 3), while p-value of Cochrane’s Q drastically increased (Q_pval = 0.006). We discuss this result in detail in the discussion section. Of those that remained significant through three of four sensitivity analysis, effect estimates of 3 taxa on WC were significant after adjusting for multiple comparison: class Melainabacteria (OR = 1.02, 95%CI 1.01–1.03, adjusted P = 0.01), genus Eubacterium hallii (OR = 1.03, 95%CI 1.01–1.04, adjusted P = 0.03), order Gastranaerophilales (OR = 1.02, 95%CI 1.01–1.03, adjusted P = 1.20*10− 3). Another 13 associations remained significant in at least three robust analyses, MR-Egger intercept term included (Fig. 3). None of the association between GM and hyperglycemia remained after sensitivity analysis. All effect size estimates were small (abs(β) < 0.1). Fig. 3Forest plot of the associations that were consistent across at least 3 robust MR analyses. * Indicates the effect size estimate was calculated without the pleiotropic SNP rs182549. Dichotomous traits are reported on a log(OR) scale. Continuous are reported on 1-SD scale. Dots depicts the point estimate. Horizontal bars depict 95% confidence interval (CI) Forest plot of the associations that were consistent across at least 3 robust MR analyses. * Indicates the effect size estimate was calculated without the pleiotropic SNP rs182549. Dichotomous traits are reported on a log(OR) scale. Continuous are reported on 1-SD scale. Dots depicts the point estimate. Horizontal bars depict 95% confidence interval (CI) Discussion This study assessed the impact of individual GM taxa abundance on MetS and its five comprising traits using the summary statistics of by far the largest GWAS dataset. Based on the result of two-sample MR, 16 associations were discovered after sensitivity analysis, yet only 3 associations remained after multiple test correction: class Melainabacteria, genus Gastranaerophilales and genus Eubacterium hallii all showed a positive effect on abdominal obesity. For the relationship between order Bifidobacteriales and WC, rs182549 accounts for a large portion of heterogeneity in the effect size estimate (Figure S1). SNP rs182549, located in the LCT (lactase) gene, has a strong association with the abundance of genus Bifidobacterium (p = 1.28 × 10 − 20), which belongs to order Bifidobacteriales [18]. The rs182549 variant itself predisposes individual to lactose intolerance. It was also associated with multiple dietary and metabolic phenotypes, including obesity and type 2 diabetes mellitus [38, 39]. Since including rs182549 would greatly increase the risk of biasing the causal estimate, it was excluded from the analysis and all effect estimates ceased to be significant (Fig. 3). All estimates of effect size were small (abs(b) < 0.1). Altogether, results of this study suggest that previously reported associations between GM and MetS with its comprising traits might have been due to biases such as reverse causality or confoundings. The impact of GM on these traits might be overexaggerated. Although the majority of effect estimates of GM on MetS-related phenotype investigated in this study turned out to be statistically insignificant, we still identified 3 taxa that are positively associated with obesity, represented by waist circumference: Class Melainabacteria (β = 0.02, adjusted P = 0.01) with affiliated order Gastranaerophilales (β = 0.02, adjusted P = 1.20*10− 3) and genus Eubacterium hallii (β = 0.03, adjusted P = 0.03). Eubacterium hallii plays a role in the fermentation of dietary fibers and production of SCFAs like butyrate and propionate, essential in energy metabolism and lipid regulation. Although the primary analysis in this study suggests that Eubacterium hallii is associated with a minor increase in waist circumference (β = 0.03), this result needs to be interpreted with great caution, since this association didn’t remain after robust sensitivity analysis. Class Melainabacteria have been correlated with selected neurodegenerative, gastrointestinal, hepatic, metabolic, and respiratory diseases [40]. Previous studies have demonstrated that dysbiosis of GM is associated with extensive physiological changes and increased risk of MetS [41], although they mostly didn’t align with our findings. For example, studies conducted on rodents associated different species from order Bifidobacteriales with a decreased risk of obesity [42, 43]. Some human FMT studies also reported an increase of Bifidobacterium abundance in MetS patients after FMT [44–46]. Study in MetS volunteers have found increased Eubacterium hallii abundance in duodenum of MetS patients after FMT, while their insulin sensitivity significantly increased [47]. While a number of differences between the microbial composition of GM in healthy and diseased individuals have been identified, causality remains to be elucidated. FMT study is often used to assess the causality between GM and cardiometabolic traits. The largest meta-analysis to date of FMT clinical trials found that, although there were some studies in favor of using FMT to improve MetS related traits, no clinical significant alteration after FMT in such traits including triglyceride, HDL, FBS, BMI and weight [48]. Mice FMTs are also frequently used for exploratory purpose, but researchers should be cautious when trying to translate these finding to human subjects. This is partly because that a substantial proportion of species in human gut are not present in mice [49]. For example, several FMT studies from lean to obese mice showed significant improvement in MetS related traits, but human FMT failed to replicate these results. For observational GM studies in human, many factors can compound the association under study, like age, sex, diet, geography, antibiotic use [50]. In addition, the change in GM composition found in such studies can also be a result of pathological condition rather than causing the disease. Vast number of null results in this study may stem from factors related to the interplay between GM and MetS. First, GM’s effect on host metabolism likely involves complex mechanisms beyond taxa abundance alone. Gut microbiota can affect host metabolism through SCFAs, such as acetate, butyrate, and propionate, produced by taxa like Lachnospiraceae, Eubacterium, and Ruminococcaceae, with Bifidobacteria contributing significant acetate [51–53]. SCFAs show a dual role: promoting metabolic health through anti-inflammatory effects and satiety regulation [41], but potentially contributing to energy harvest and obesity under certain conditions [54, 55]. Additionally, the bile salt hydrolase (BSH) activity in taxa like Bifidobacteria influences liposoluble nutrient absorption, further complicating the GM–MetS relationship [41]. This complexity suggests that, while the abundance of specific taxa alone may not show strong effects on MetS, GM interactions through metabolite pathways may be key. Future studies should focus on these molecular mechanisms to understand GM’s multifaceted impact on metabolic health. Second, environmental and dietary influences further complicate this relationship, as diet and lifestyle significantly shape both GM composition and metabolic health [56–59]. Variability in dietary fiber intake, for instance, modulates SCFA production, which can mask direct GM–MetS associations in the analysis [60]. Non-nutritive sweeteners (NNS) are commonly consumed by those with imbalanced calorie intake, especially in the perinatal period. Accumulating evidence showed the transgenerational and mediative role of human microbiota in the development of early-life MetS [61]. Lastly, the use of weak genetic instruments may limit the study’s power to detect subtle causal effects of GM on MetS. The sparse availability of robust genetic instruments for specific taxa may lead to conservative estimates, masking potential associations. These limitations suggest that future research should prioritize exploring molecular mechanisms and more precise GM interactions with MetS-related traits to clarify GM’s role in metabolic health. Meanwhile, the use of larger GWAS encompassing broader genome sequence can greatly speed up related research using statistical approach such as MR. In summary, the large amount of null findings (GM taxa abundance may not cause MetS and its comprising traits) agrees with the recent literature showing an enormous positive publication bias in the GM literature with flawed statistical analysis approach [11]. Moreover, recent meta-analysis on FMT showed that current FMT was unable to bring any clinical significant improvement in cardiometabolic traits [48]. Evidence from larger GM interventional experiment with rigorous design, corroborated by statistical causal inference approach (e.g. MR), is needed to ascertain the clinical value of GM-targeted intervention [11]. The present study has several strengths. First, this study evaluates the causal relationship of GM on MetS and its five comprising traits using the largest GWAS dataset by far. Second, the result of the two-sample analysis is less susceptible to confounding factors, reverse causality than in observational study. Third, the GWAS data was mostly restricted to individuals of European ancestry to reduce the bias due to population stratification. Our study, however, has limitations. First, robust genetic instruments for GM are challenging to find because of high microbiome heterogeneity and high interindividual variability. Given this nature flaw of microbiome GWAS, positive findings from MR should be further corroborated by microbiome transplant experiments. Second, the microbiome GWAS used in this study targeted only V1-V4 region rather than the entire 16 S RNA gene, which greatly reduced its ability to achieve a sufficient taxonomic resolution to identify potential therapeutic targets. Future microbiome GWAS should focus on extending the sequencing scope with hope of more advanced sequencing technology. Third, given the exploratory nature of this study, we employed a loose GWAS significance threshold of P < 1 × 10 − 5 to select SNPs, which in turn increased the risk of type 1 error (finding false positive results). Given the subject of this study, instrumental variables for GM were all derived from the host. Such instruments have a high chance of being pleiotropic. In general, pleiotropy is more likely to bias away from the null [62]. Therefore, our null results are generally more robust to horizontal pleiotropy. Finally, relationships identified in this study are limited in European population and cannot be extended to other populations, since GWAS data used in this study derived mostly from European cohorts. Therefore, conducting large-cohort microbiome GWAS outside European population is of great significance. Conclusion Using two-sample MR, we found little evidence supporting a large causal effect, i.e. absolute effect size > 0.1, of gut microbial taxa abundance on metabolic syndrome and its comprising traits. While our results did not support a substantial association, they should be interpreted with caution due to the inherent uncertainties in the MR approach and the limited robustness of available genetic instruments. In the future, performing larger microbiome GWAS study that sequences the full 16 S RNA gene is warranted to fully elucidate the association of human gut microbiota with metabolic syndrome. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1: S1 File. STROBE-MR checklist of recommended items to address in reports of Mendelian randomization studies. Supplementary Material 1: S1 File. STROBE-MR checklist of recommended items to address in reports of Mendelian randomization studies. Supplementary Material 2: S1-16 Figure. Leave-one-out MR analysis results of associations with Cochrane’s Q pval < 0.05. Supplementary Material 2: S1-16 Figure. Leave-one-out MR analysis results of associations with Cochrane’s Q pval < 0.05. Supplementary Material 3: S1 Table. Details of selected SNPs for GM. S2 Table. Details of SNPs used in GM and fasting glucose. S3 Table. Details of SNPs used in GM and HDL cholesterol. S4 Table. Details of SNPs used in GM and hypertension. S5 Table. Details of SNPs used in GM and metabolic syndrome. S6 Table. Details of SNPs used in GM and waist circumference. S7 Table. Details of SNPs used in GM and triglyceride. S8 Table. Result of all MR analyses. S9 Table. Primary MR analysis results with a false discovery rate (FDR) less than 0.05. S10 Table. Primary MR analysis results with p value less than 0.05. S11 Table. Result of MR-Egger intercept significance test. S12 Table. Result of Cochran’s Q statistic. S13 Table. Result of MR analysis after excluding potential heterogeneous SNP. Supplementary Material 3: S1 Table. Details of selected SNPs for GM. S2 Table. Details of SNPs used in GM and fasting glucose. S3 Table. Details of SNPs used in GM and HDL cholesterol. S4 Table. Details of SNPs used in GM and hypertension. S5 Table. Details of SNPs used in GM and metabolic syndrome. S6 Table. Details of SNPs used in GM and waist circumference. S7 Table. Details of SNPs used in GM and triglyceride. S8 Table. Result of all MR analyses. S9 Table. Primary MR analysis results with a false discovery rate (FDR) less than 0.05. S10 Table. Primary MR analysis results with p value less than 0.05. S11 Table. Result of MR-Egger intercept significance test. S12 Table. Result of Cochran’s Q statistic. S13 Table. Result of MR analysis after excluding potential heterogeneous SNP.