Baseline Clinical Characteristics of Study Cohort
To explore the role of the microbiome composition in mediating clinical outcomes of SCAD+DM patients, we used a prospective cohort of coronary artery disease from PUMCH (Peking Union Medical College Hospital), China (9). A total of 164 subjects were divided into the following groups based on guideline (21): SCAD+DM group (N = 38), SCAD group (N = 71), and HC group (N = 55). The traditional cardiovascular risk factors of the 164 subjects are summarized in Table 1. There were significant differences in cardiovascular risk factors between the three study groups, except for triglyceride (TG) (P = 0.125), and high-sensitivity C-reactive protein (hs-CRP) (P = 0.113). Coronary atherosclerotic burden and severity were assessed using the Gensini score (22), indicating no significant difference between patients with or without T2DM (P = 0.298). We focused on the comparison of SCAD+DM versus SCAD. We observed a significant increase in body mass index (BMI) (P < 0.001), fasting blood glucose (FBG) (P < 0.001), and hemoglobin A1c (HbA1c) (P < 0.001) levels in the SCAD+DM group compared with the SCAD group. The medication usage rate of oral hypoglycemic agents (P < 0.001) and insulin (P < 0.001) were also significantly higher in the SCAD+DM group compared to the SCAD group. HOMA-IR (11) was used to assess insulin resistance, The SCAD+DM group showed a higher level than the SCAD group (P < 0.001). While HOMA-β, which reflected pancreatic β-cell function (11), was significantly lower in the SCAD+DM group compared to the SCAD group (P = 0.005). As for the expression of myocardial enzyme, there were significant differences between the two groups, such as creatine kinase (CK) (P = 0.028), creatine kinase-MB (CK-MB) (P = 0.015), and cardiac troponin-I (cTnI; P = 0.003). Although the results of cardiac catheterization did not reveal differences in the severity of atherosclerosis between SCAD+DM and SCAD groups, we suppose that glucose metabolism and myocardial damage were disturbed between the two groups.
Gut Microbial Diversity and Enterotype in SCAD and SCAD+DM
In order to investigate the gut microbiome of all study subjects, we performed metagenomic shotgun sequencing on a total of 164 fecal samples. After removing low-quality reads and human DNA reads, on average 58.6 million high-quality sequencing reads per sample were aligned to a comprehensive reference gut microbiome gene catalog comprising 9.9 million genes (23), which allowed, on average, 72.89 ± 2.42% of the reads in each sample to be mapped (Table S1), consistent with saturation of the gene-coding regions. The Shannon index and Simpson index were calculated to estimate the alpha diversity. The Shannon index at the genus level was much lower in SCAD and SCAD+DM groups (P = 0.04, HC vs. S; P = 0.005, HC vs. S+D; Wilcoxon rank sum test, Figure 1a). Consistently, the Simpson index was significantly decreased in both SCAD and SCAD+DM groups as compared to the controls (P = 0.016, HC vs. S; P = 0.006, HC vs. S+D; Wilcoxon rank sum test, Figure 1b). To assess the overall structure of the gut microbiota, the score plot of dbRDA (distance-based redundancy analysis) based on Bray-Curtis distances was constructed. The results indicated that the structure and composition of the microbiota differed significantly between different groups, even in SCAD vs. SCAD+DM comparison (P = 0.0017, H vs. S; P = 0.0091, H vs. S+D; P = 0.0063, S vs. S+D; Permanova test, 9999 permutations; Figure 1c). The reduced richness of genera in the gut microbiota of SCAD-combined DM was consistent with previous findings (24), suggesting a possible deficiency of healthy microbiome in atherosclerotic cardiovascular disease.
Enterotypes were identified based on the abundance of genera, in order to explore the differences between microbial communities between different groups. The principal coordinate analysis (PCoA) using Jensen-Shannon distance was performed to cluster the 164 samples into two distinct enterotypes (Figure 1d). Bacteroides was the most enriched genus in enterotype 1, while Prevotella was the most enriched genus in enterotype 2 (P < 0.0001 and P < 0.0001, respectively; Wilcoxon rank sum test, Figure 1e). Both contributors in the two enterotypes have been reported in European and Chinese populations (12). A higher percentage of HC and SCAD+DM patients were distributed in enterotype 1 (72.7% for HC, and 76.3% for S+D), while SCAD group showed a higher level of enterotype 2 (45.1% for S, Figure 1f). These findings suggest that enterotype 1 may represent bacterial structure associated with SCAD+DM, while enterotype 2 may be associated with SCAD.
SCAD and SCAD+DM Subjects Harbor Different Species of Gut Bacteria
To identify gut microbiota with potential value for SCAD and SCAD+DM diagnosis, we investigated bacterial alterations at the species level. The abundances of 13 bacterial species were observed to be significantly different in SCAD when compared with HC (Wilcoxon test P value < 0.05, absolute log2 fold change > 0.5). Six of these were SCAD-enriched while seven others were SCAD-depleted (Figure 2a, Table S2). Species were more abundant in the SCAD group, including Fusobacterium ulcerans, Odoribacter splanchnicus, and Parabacteroides merdae, while species like Coprococcus eutactus, unclassified Ruminococcus sp. SR1/5, and unclassified Ruminococcus sp. 5_1_39BFAA showed higher levels in the HC group. Compared to the HC group, the SCAD+DM subjects were characterized by 13 species, consisting of Lactobacillus amylovorus, Veillonella parvula, and Parabacteroides distasonis. Six species were decreased in the SCAD+DM group (relative to HCs) belonging to genera Ruminococcus and Roseburia inulinivoran (Figure 2b). Furthermore, in the SCAD vs. SCAD+DM comparison, 14 Species changed greatly (P < 0.05, Wilcoxon rank sum test), and five species were altered significantly (Absolute log2 fold change > 1). Bifidobacterium longum, Bifidobacterium catenulatum, and Ruminococcus torques were more abundant in patients with DM (P = 0.037, P = 0.025, P = 0.028, respectively, Wilcoxon test). While Alistipes putredinis and Roseburia inulinivorans were more abundant in the SCAD group (P = 0.029, P = 0.032, respectively, Wilcoxon test, Figure 2c). Recently, research identified that Alistipes putredinis was negatively correlated with OGTT (oral glucose tolerance test) response in gestational diabetes patients (25). We further found that Alistipes putredinis was positively correlated with AS severity (Rho=0.3, P=0.0021), while negatively correlated with HbA1c% (Rho=-0.21, P=0.027) and HOMA-IR (Rho=-0.24, P=0.02). While Bifidobacterium longum was showed positively correlated with FBG levels (Rho=0.29, P=0.0038, Figure 2d).
The divergence of gut microbiota (GM) composition in each subject was assessed to explore the correlation of 171 species with 34 clinical indicators. We found that seven atherosclerotic phenotypes, including age, systolic blood pressure (SBP), New York Heart Association (NYHA) class, statin, total cholesterol (TC), TNF-α, and cTnI, were significantly associated with perturbated species abundance (Figure S2, Supporting information). Species belonging to the Lactobacillus genera were positively correlated to cTnI and tumor necrosis factor-α (TNF-α), while negatively correlated to TC. Species such as Ruminococcus, Lachnospiraceae, and Streptococcus were negatively associated with cTnI and positively associated with TC. In summary, we show that the microbiota composition differed between SCAD and SCAD+DM, and correlated with the major phenotype indicators of AS and DM.
Metabolic Profiling in SCAD and SCAD+DM Group
Considering the aberrant function profiles of gut microbes in disease subjects, we investigated the microbe-host interactions in atherosclerotic and diabetic metabolic diseases. As certain products of fermentation from the GM could enter the bloodstream and influence host physiology, we explored the host metabolic profile in fasting serum of a subset of 75 subjects using high-throughput liquid chromatography-mass spectrometry (LC/MS) and examined the relationship between GM and serum metabolites. Metabolomic profiling yielded 5461 features after eliminating impurity peaks and duplicate identifications. Based on the orthogonal projections to latent structures discriminant analysis (OPLS-DA) models of metabolite profiling data, we found that the serum metabolites were significantly separated between all SCAD patients with or without DM and healthy controls. The compositional changes in patients involved 66 analytes that were significantly different between HC vs. SCAD, and 118 analytes between HC vs. SCAD+DM (detailed in Methods Section). There were 15 metabolites, which were obviously different, in both SCAD and SCAD+DM groups as compared to the control (Figure 3a). Notably, these metabolites exhibited statistically analogous profiles of alterations in SCAD and SCAD+DM, which agreed with our gut microbiome findings (Table S3).
We identified a collection of differentially produced compounds belonging to both host-derived and bacterial-derived metabolites. Sixty metabolites of this collection were found significantly associated with glycometabolic indicators, including FBG, HbA1c, HOMA-IR, HOMA-β, and atherosclerosis associated factors, such as number of stenosed vessels and cTnI levels (Figure 3b, Table S4). These metabolites were separated into groups that were either positively or negatively correlated with insulin resistance (hereafter represented by IR+ or IR-, respectively) or atherosclerosis severity (AS+ or AS-, respectively). Among these, IR+ metabotypes including carboxylic acid, glucuronic acid, benzoic acid, and amino acids, were positively correlated with FBG, HbA1c, and HOMA-IR levels, while negatively correlated with HOMA-β. Surprisingly, these IR+ metabotypes were even associated with the severity of atherosclerosis. While other metabolites, such as leucine, glutarylglycine, and terpene lactone, were negatively correlated with insulin resistance and AS severity. Moreover, the IR+ and AS+ metabotypes were positively correlated metformin and other oral antidiabetic drugs, while negatively associated with interleukin1β (IL-1β) levels, which is a classic proinflammatory cytokine linked to atherosclerosis (26). AS+ metabotypes were found to be positively correlated to traditional risk factors, such as SBP, smoke, hyperlipidemia, IL-6, TNF-α. The IR- and AS- metabotypes were generally negatively related to these phenotypes (Figure S3). Taken together, these results suggested that the SCAD+DM patients had significantly different metabolite profiles compared with SCAD, and closely related to diabetes and atherosclerotic indicators.
Functional alteration in GM Linked to Metabolic Metabotypes of SCAD and SCAD+DM
Using the Kyoto Encyclopedia of Genes and Genomes (KEGG) (27) database, we evaluated gut microbial functions across groups in our study cohort. All genes were aligned to the KEGG database and assigned to the KEGG orthology (KO), after cross-comparison, we identified 356 KO in total (Table S5, Supporting information). Principal component analysis (PCA), based on KO, revealed striking differences in microbial functions at the first principal component (PC1) between HC and SCAD (P < 0.0001, Wilcoxon rank sum test) and SCAD vs. SCAD+DM (P < 0.001, Wilcoxon rank sum test, Figure S4a, Supporting information). Thereafter, we investigated the metabolic potential of the gut microbiome in relation to these metabolites using KEGG functional modules consisting of KOs (Figure S1e, Supporting information). The KOs collection set were clustered into 67 microbiome functional modules, and 21 of the 67 microbiome functional modules were significantly associated (Wilcoxon rank sum test, FDR < 0.05) with one or more of the IR and AS phenotypes. All 21 functional modules were furthermore associated with the IR and AS associated metabotypes (Figure 4), with a majority also differing in abundance in the expected direction in the cohort (Figure S4b, Supporting information). By abundance comparison, we found that the AS- metabotypes, consisting of L-leucine and cyclic alcohol derivatives, were generally highly abundant in healthy subjects. While AS+ metabotypes were more abundant in SCAD and SCAD+DM groups (Figure 3c).
The cross-domain associations between the IR and AS phenotypes, the serum metabolome, and the gut microbiome described above may suggest functional relationships. Notably, the functional modules that were negatively associated with AS + metabotypes involved carbohydrate metabolism, including nucleotide sugar biosynthesis, galactose degradation, amino sugar metabolism, two-component system, and bacterial secretion systems. In contrast, the microbial functional modules positively associated with AS + metabotypes contained fatty acid synthesis pathways and aromatic amino acid metabolism. In addition, microbial modules positively correlated with IR, primarily including the metabolism of cysteine and methionine, cofactors and vitamins, and amino sugar and nucleotide sugar systems. Study have showed that some of these microbial pathways have elevated expression in gut microbiomes when transplanted into mice from obese donors (28).
Microbial and Metabolites Features Predict Major Adverse Cardiac Event
After identifying differences in intestinal microbial composition between SCAD and SCAD+DM in our cohort, we examined whether microbial and metabolic signature would enable prediction of major adverse cardiac event (MACE) based on baseline status of the cohort. Recent study showed that connection of longitudinal profiling of glucose metabolism with multi-omics profiling facilitating the precision medicine goal of defining diseases on the basis of molecular mechanisms and pathophysiology (29). The definitions of primary and secondary outcomes were detailed summarized in Method. During a median follow-up period of 18.6 months (interquartile range: 18–22.7 months), we effectively followed up 109 patients totally. No significant difference was found in the endpoint event comparison between the two groups (Table 2). We built a classification model based on random forest using the explanatory variables of species, metabolites and KEGG modules to exploit the prediction efficacy of MACE outcomes with a tenfold cross-validation. The model performance was evaluated with an area under the curve (AUC) of receiver operating characteristic (ROC). Using the initial relative abundance of differentially abundant species or metabolites solely, the performance was the lowest (AUC = 0.63, 95% CI 0.43-0.81; AUC=0.6, 95% CI 0.35-0.82, respectively) (Figure 5a). The prediction performance was significantly improved by using the metabolites combined with KEGG modules (AUC = 0.70, 95% CI 0.44-0.91). However, the model incorporating data on both species and metabolites showed the best performance (AUC = 0.71, 95% CI 0.4-0.83), indicating the power of shotgun metagenomics for predicting host phenotypes. The features with predictive value were metabolites including carboxylic acid, norfuraneol, 3b-Hydroxy-5-cholenoic acid while species containing Ruminococcus torques, Edwardsiella ictaluri-tarda and Bifidobacterium longum (Figure 5b).
We then stratiﬁed subjects into high versus low categories based on their median relative abundance of these features selected. Signiﬁcantly better outcomes were predicted for MACE with higher abundance of Bifidobacterium.longum (Hazard Ratio [HR] = 2.652, 95% conﬁdence interval [CI] 1.17–4.92), Ruminococcus.torques (HR = 2.363, 95% CI 1.08–4.56), and 3,8-Dihydroxy-1-methylanthraquinone-2-carboxylic acid (HR = 4.53, 95% CI 1.43–11.79) (Figure 5c). The high accuracy of our prediction models indicates that the initial condition of the gut microbiota could be a potential predictive tool for cardiovascular prognosis outcomes. Furthermore, the performance comparisons of our models suggest that combining the features of both microbiome and metabolomics improves the prediction accuracy.
Metformin Alters Gut Microbiome Signatures in SCAD+DM Patients
Metformin is the most prescribed pharmacotherapy for individuals with type 2 diabetes, accumulating evidence indicates that microbial changes might contribute to the antidiabetic effect of metformin (30). We further divided the SCAD+DM population into two subgroups, namely metformin treated (Metformin+, n=18) and metformin untreated (Metformin-, n = 20). Multivariate analysis showed significant (PERMANOVA = 0.045) differences in gut taxonomic composition between metformin untreated and metformin treated groups, consistent with a broad-range of dysbiosis in T2DM (Figure 6a). To further interpret the therapeutic effects of metformin on gut microbiota shifts, we observed that gene richness significantly increased in the T2DM metformin+ microbiome, but was reduced in T2DM metformin- microbiomes (P = 0.0186, Wilcoxon rank sum test, Figure 6b). We then compared T2D metformin+ and T2D metformin- subjects to characterize the treatment effect in more detail. Univariate tests of the effects of metformin treatment showed a significant increase of unclassified Clostridum spp. and a reduced abundance of Prevotella bryantii., Citrobacter koseri, and Acidaminococcus fermentans (Figure 6c).
We further explored gut microbiome alterations in metformin-untreated compared with metformin-treated subjects using univariate tests of microbial taxonomic and functional differences, with significant trends shown in Figure 6d. Citrobacter koseri was more abundant in Metformin- subgroup while Escherichia coli and Shewanella frigidimarina exhibited high levels in the Metformin+ subgroup. Functionally, we found enrichment of pyrimidine metabolism and modules for nitrogen metabolism varied significantly. Citrobacter koseri exhibited a strong negative correlation with nitrification and complete nitrification, implicating the interaction of metformin with gut bacteria.