General characteristics of study participants during a follow-up of 3 years

The study cohort in our previous work [24] consisting of 49 healthy controls, 50 stable CHDs and 100 STEMI patients were recruited for the present study. Stable CHD was defined as previous myocardial infarction (12 weeks earlier), unstable angina (6 weeks earlier), evidence of coronaryartery disease by arteriography, or a revascularization procedure [25].

All STEMI patients were followed-up for 3 years after STEMI onset. The occurrence of a first major adverse cardiovascular event (MACE) was regarded as the follow-up endpoint, which included cardiovascular death, non-fatal ischemic stroke, recurrent MI, need for emergency or repeat revascularization, and re-hospitalization for heart failure, as defined previously [24]. Thirty-three patients suffered from MACEs were recorded, including seven cardiovascular deaths, three non-fatal ischemic strokes, one recurrent MI, 12 emergency or elective repeat revascularization, and ten readmissions for heart failure (Table 1). When comparing STEMI patients suffered MACEs (MACE[+]) to those with event-free survival (MACE[−]), MACE (+) patients showed lower LVEF and higher SYNTAX score (Table 1). All STEMI patients presented a higher increase of LPS and d-lactate on day 2, and a significant difference was observed between MACE (+) and MACE (−) patients in plasma d-lactate on the day 2 of STEMI onset.

Gut-associated bacteria are enriched in systemic microbiome

As the product for gut bacterial translocation (LPS and d-lactate) in STEMI patients was significantly increased and peaked on day 2 of symptom onset, the blood samples drew on day 2 was used for microbiota analysis (Fig. 1 and Fig. 2a-c). The bacterial 16S rDNA in plasma samples was extracted and sequenced on the Illumina platform. A total of 10,560,437 high-quality sequences were generated, with an average of 53,336 per sample (Additional file 1: Table S1). The total number of operational taxonomic units (OTUs) was higher in STEMI patients than that of controls and CHDs (Fig. 1a). Consistently, the within-sample (α) diversity (present as Shannon index) of STEMI group was also increased (Fig. 1b). The elevated OTUs and bacterial diversity indicate more abundant bacteria in the systemic circulation of STEMI patients. Principal Coordinate Analysis (PCoA) (unweighted UniFrac distance metric based on the OTU table) showed that blood bacterial characteristics in STEMI patients were clearly separated from that in either control or CHD group (all P < 0.001 by Anosim analysis and multi-response permutation procedure (MRPP)), although CHD group was also distinct in comparison with healthy controls (Fig. 1c). To identify the taxonomic information of the blood bacteria, the RDP classifier and GreenGene database were used. At the phylum level, the blood dominant bacteria were from Proteobacteria, Cyanobacteria, Firmicutes, Bacteroidetes, Actinobacteria and Fusobacteria, Euryarchaeota, Acidobacteria, Nitrospirae, and Chloroflexi (Additional file 2: Table S2, Additional file 3: Figure S1), similar to the previous report [27]. In STEMI patients, a total of 210 genera were at significant abundance differences, and the genera abundance distribution was quite discrepant to that of controls and CHDs (Fig. 1d, Additional file 2: Table S3). The heat map (according to top 40 the most different genera) showed the specific genera distribution among each group (Fig. 1e). Based on reference genomes in HMP database from the human gut and oral, we found more than 12% of post-STEMI plasma bacteria were dominated by intestinal-sourced-flora, distinguishing from CHD and control group (Fig. 1f). The possibility was raised that intestinal bacteria might invade into blood post-MI, leading to the increased richness and distinct structure of systemic microbiome.

Products of gut bacterial translocation are associated with inflammation and LV function

LPS and d-lactate were recognized as products and systemic markers for increased gut permeability and bacterial translocation [28, 29]. To measure LPS and d-lactate, blood samples of STEMI patients were collected on days 1, 2, 3, 5, and 7 of symptom onset. And the dynamic changes of LPS and d-lactate in STEMI patients were obtained. In comparison to control and CHD patients, STEMI patients’ plasma LPS and d-lactate levels increased since admission, reaching the peak on day 2 (Fig. 2a–c) and maintaining the tendency to day 7. Interestingly, the changes of LPS and d-lactate were consistent with monocytosis [24]. A prominent feature for the post-infarction inflammation is the recruitment of monocyte’s subpopulations [30–33], and CD14++CD16+ monocytes represent unique pro-inflammatory properties associated with MI [34]. To investigate the possible correlation between gut bacterial translocation and monocytosis, the association of LPS, d-lactate (Δ-value of days 2–1) with monocyte counts was analyzed. Increased plasma LPS positively correlated with total monocytes, CD14++CD16– counts (Additional file 4: Figure S2a–b) and CD14++CD16+ subset’s recruitment in the total cohort (r = 0.214, P = 0.033), especially in anterior wall infarction patients (r = 0.388, P = 0.007; Fig. 2d). However, the correlation between d-lactate and monocyte subsets was not striking (Fig. 2e, Additional file 4: Figure S2c–d). Intriguingly, although d-lactate and LPS in STEMI patients have the same trend during the 7 days of observation, changes of d-lactate but not LPS were negatively correlated with LVEF by analyzing the correlation coefficients (Additional file 5: Figure S3). The association of Δ LPS and LVEF did not reach statistical significance. These findings indicated that inflammatory response after MI was partly associated with gut bacterial translocation.

Gut microbial translocation promotes cardiovascular events after STEMI

The elevated monocytosis was associated with cardiovascular events post-STEMI [24]. Therefore, we assessed the possible role of gut microbial translocation in cardiovascular events. The discriminative and prognostic capacities of Δ LPS and Δ d-lactate were confirmed in receiver operator characteristic (ROC) curve analysis (Fig. 3a, b), the performance of which was higher than other day values (Additional file 6: Figure S4). Kaplan-Meier analyses further verified that the optimal cut-off values for Δ LPS (Fig. 3d) and Δ d-lactate (Fig. 3e) were capable of identifying subjects with higher risk for post-STEMI MACEs. To incorporate the influence of the two translocation markers, we calculated the z-score values of Δ LPS and Δ d-lactate, and defined the sum of two z-score values divided by 2 as translocation z-score. Kaplan-Meier survival plot and ROC curve analysis showed that translocation z-score led to a more efficient risk stratification, increased Youden index, and improved AUC of ROC curve (P for multiple ROC comparisons, 0.080; Fig. 3c, f). By performing a forward stepwise selection, an optimal Cox proportional hazards model was constructed: among the candidate variables, only translocation z-score, hs-CRP, and CD14++CD16+ monocytes remained statistical significances in multivariate analysis. Syntax score and LVEF were excluded from the final model. To simplify their synergistic effects, we defined a simple risk score, i.e., TCM score (using the combination of the three capital letters for three markers), based on the optimal cut-off values in ROC analysis for MACEs (Additional file 1: Table S1). The multivariate Cox regression curve showed that TCM score was efficient for risk stratification by showing clearly separated survival curves as TCM score increasing from 0 to 3, even after adjustment by age, LVEF (day 2), and syntax score (Fig. 3g). ROC curve analysis further demonstrated that TCM score had an AUC of 0.822 (95% CI 0.738–0.906, P < 0.001) and was well calibrated (P = 0.398 by Hosmer-Lemeshow’s test) for MACEs (Fig. 3h).

Previous studies and our recent work have demonstrated hs-CRP, CD14++CD16+ monocytes, and LVEF are predictors for post-STEMI MACEs. We next sought to examine whether and how much microbial translocation could explain or mediate such associations. By using causal mediation analysis, we demonstrated that translocation z-score had a significant mediation effect (22.7%, P < 0.05) on the CD14++CD16+ monocytes→MACEs association (per 1 SD increase), a significant mediation effect (25.4%, P < 0.05) on the LVEF→MACEs association (LVEF < 52%), and a significant mediation effect (13.7%, P < 0.05) on the hs-CRP→MACEs association (per 1 SD increase) (Fig. 4). Thus, microbial translocation could partially explain post-STEMI inflammatory response and compromised LV function-associated adverse outcomes.

Gut permeability upon MI leads to microbial translocation and inflammation

To investigate the gut permeability contribution to microbial translocation, MI mice model was used. By fluorescein isothiocyanate (FITC)-dextran tracing assay, colon permeability increased from days 1 to 5 after MI, and recovery at day 7 (Fig. 5a). As colon, the permeability of distal small intestine and was also elevated (Fig. 5b). Accordingly, gut bacterial products (LPS and d-lactate) were increased and peaked on day 3 (All P < 0.01, Fig. 5c, d). In MI mice, the changes of the Ly6C^hi proportion monocyte subset showed the similar tendency (All P < 0.01, Fig. 5e, f), and the proportion of Ly6C^hi monocytes was positively correlated with peripheral blood concentration of LPS as well as d-lactate, consistent with STEMI patients (Fig. 5g–h). Thus, this mouse model was further used for mechanisms of gut permeability post-MI.

With gut permeability changes, occludin (tight junction protein as the intestinal permeability marker) mRNA and protein expression in the colon was dramatically decreased after MI (Fig. 6a, b). In association with occludin expression downregulation in the mucous membrane of intestinal epithelium (Fig. 6c), mucosal microvillus disarrangement, structure disrupted, and even loss were detected by transmission electron microscopy after MI (Fig. 6d). In keeping with these changes, blood flow of superior mesenteric artery decreased from day 1 and reached the nadir on day 3 (P < 0.01; Additional file 7: Figure S5a), which was negatively correlated with FITC-dextran permeating in the small intestine (Additional file 7: Figure S5b), but positively correlated with LVEF (r = 0.64, P < 0.001; Additional file 7: Figure S5c). These data suggested that elevated gut permeability was due to the intestinal mucosa barrier dysfunction by ischemia stress after MI.

Inhibition of gut microbial translocation reduces infarct size

To suppress gut bacterial translocation and eliminate microbial products in systemic circulation, polymyxin B (PMB; an antibiotic with distinct cationic characteristics against LPS-induced toxicity) was used to treat MI mice. By PMB administration for 3 days, the plasma LPS level was significantly reduced (Fig. 7a), and the proportion of Ly6C^hi was coincidently decreased (Fig. 7b). At other time points post-infarction, PMB treatment also lowered the Ly6C^hi counts although LPS was no changed (Additional file 8: Figure S6).

Consistent with decreased LPS and Ly6C^hi, PMB treatment significantly lowered infract size, as determined by triphenyltetrazolium chloride stainings (Fig. 7c). Accordingly, PMB treatment also reduced local inflammation (Fig. 7d), macrophage infiltration (Fig. 7e), and myocardial fibrosis (blue area, by Masson stainings; Fig. 7f). These findings confirmed that by abrogation of gut bacterial translocation, inflammatory response and heart injury post-MI could be improved.

Study cohort and patient characteristics

The patients admitted to Pingjin Hospital Heart Center from November 2012 to May 2013 in our previous work were enrolled [24]. The diagnosis and treatment of STEMI were performed according to the guideline recently published [50]. Patients with infectious and inflammatory disorders, cancer, previous MI, and decompensated heart failure in the past 6 months were excluded. Individuals were also excluded if they had received antibiotics, probiotics, or hormone-replacement therapy within the last 8 weeks. All the STEMI patients have taken aspirin prior to primary PCI. PCI was performed with conventional techniques and coronary stents were used without restrictions. The infarct-related artery was the only target of the procedure. Finally, there were 49 healthy controls, 50 stable CHDs, and 100 STEMI patients in the study cohort. The admission blood samples were collected after initial diagnosis and prior to the first antiplatelet medication. The study was performed in accordance with the Helsinki declaration and was approved by the hospital Research Ethics Committee. Informed consent was obtained from study participants.

Blood biochemical testing

Baseline blood routine test and biochemical assays were performed at admission using an automated hematology analyzer (XE-5000, Sysmex, Kobe, Japan) and a Hitachi 7180 Clinical Analyzer (Hitachi, Tokyo, Japan). High-sensitivity C-reactive protein on day 2 of STEMI onset was assayed by a commercially available ELISA kit (CUSABIO, Wuhan, China; Cat No. CSB-E08617h) according the manufacturer’s instruction.

LPS and d-lactate measurement

Fasting blood samples were collected from subjects in control, CHD, and STEMI group via the antecubital vein in ethylenediaminetetraacetic acid anti-coagulated tubes. The whole blood samples were centrifuged at 3500 rpm for 10 min at 4 °C and separated into plasma. Plasma samples were subsequently used for LPS and d-lactate measurement. The levels of LPS were determined by a kinetic chromogenicmethod-based Tachypleusamebocyte lysate assay (China Horseshoe Crab Reagent Manufactory Inc., Xiamen, China; Cat No.KT22) according to the manufacturer’s instruction. The kit provides pretreatment reagent that eliminated the inhibition factors in plasma. The assay range of the test was 0.01–10 EU/ml, and the precision was 0.005 EU/ml. d-lactate levels were measured using a d-lactic acid (d-lactate) (Rapid) assay kit (Megazyme, Bray, Ireland) as the others did previously [51]. The absorbance of the sample was detected at 340 nm in a spectrophotometer. The assay range of the test was > 0.214 mg/L. All the samples were assayed in duplicate.

Follow-up

All patients were followed-up after STEMI onset as we described previously [24]. To date, all the patients have finished a follow-up for 3 years since STEMI onset. Thirty-three first major adverse cardiovascular events were detected, including seven cardiovascular deaths, three non-fatal ischemic strokes, one recurrent MI, 12 emergency or elective repeat revascularization, and ten readmissions for heart failure. The occurrence of a first MACE was regarded as the endpoint. Medical records were obtained from the treating physicians to verify all events reported by study participants (Fig. 8).

Calculation SYNTAX score

Coronary lesion severity, as assessed by SYNTAX score, was calculated using the SYNTAX score algorithm (www.syntaxscore.com).

DNA extraction and Illumina sequencing

Isolating genomic DNA from whole blood is classically done using the buffy coat. Bacterial DNA is extracted from peripheral blood leucocytes. Genomic DNA from 200 ul blood samples was isolated using the MoBioPowersoil DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA) following the manufacturer’s protocol. The concentration and quality of extracted DNA were assessed photometrically using a NanoDrop® ND-2000c UV–vis spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). The universal primer set 515F (5′-GTG CCA GCM GCC GCG GTA A-3′) and 806R (5′-GGA CTA CNN GGG TAT CTA AT-3′) was used for the amplification of the V4 region of bacterial 16S rRNA gene. After 16S rDNA library preparation and generation, the library quality was assessed on the Qubit@ 2.0 Fluorometer (Thermo Scientific) and Agilent Bioanalyzer 2100 system. Then, the libraries were sequenced on IlluminaHiSeq platform 2500 and 250 bp paired-end reads were generated at Novogene (Beijing, China).

Quality filtering, OTUs picking, and annotation

Based on samples’ unique barcode, raw reads were assigned to different samples, then the assigned paired-end reads of each sample were merged to raw tags by using FLASH (Version 1.2.7) [52]. The merged raw tags were filtered and developed into clean tags according to QIIME (Version 1.7.0) quality controlled process [53]. After the quality control, clean tags were aligned to Gold database (Release 20110519) and chimera sequence was detected by using UCHIME Algorithm (Version 7.0.1001) [54, 55], these non-chimera clean tags were defined as effective tags. The effective tags were clustered into OTUs with ≥ 97% similarity by Uparse (Version 7.0.1001) [56]. The representative sequence for each OTU was selected and the taxonomic information was annotated using RDP classifier (Version 2.2) [57] and GreenGene database (Release gg_13_8) [58] (Additional file 9: Supplementary methods).

Echocardiography

Transthoracic echocardiography was performed on a Philips iE33 system (Phillips, Andover, MA, USA) at 24 h after the onset of STEMI patients. For mice model, blood flow velocity in the superior mesenteric artery (SMA) was measured using the using the Vevo 2100 High-Resolution Ultrasound System with a 40-MHz transducer (VisualSonics, Toronto, Ontario, Canada) under anesthesia with isoflurane (0.25 to 0.50%) supplemented with 100% O₂, by an experienced sonographer. Analysis of flow velocity and sound was performed in pulse-waved Doppler mode. The angle between the Doppler beam and the mesenteric artery was < 60°. The average velocity of blood flow was determined by multiplying velocity time integral by heart rate.

Animal model

C57BL/6J male mice at 6-week-old were randomly divided into groups of MI and sham group after adaptively feeding for 1 week, with five time points (1, 3, 5, 7 days; n = 5 for each group). Surgical MI was induced by the ligation of the left anterior descending coronary artery. Sham-procedure mice underwent the same protocol, but without ligation of the coronary artery. For MI+saline (NS) and MI+PMB group, normal saline was given as vehicle control, and an equal volume of PMB (2.5 μg/g) was administered by intraperitoneal injection immediately after MI induction, and on days 2 to 4 after surgery, once daily. When sacrificed, the peripheral blood was collected for measurements of LPS, and monocyte counts, heart, and colon tissues were obtained for the further study.

Flow cytometry analysis of circulating monocyte subsets

Flow cytometry (FCM) analysis was performed using a Cytomics FC500 cytometer (Beckman-Coulter, Miami, FL) according to previous work [59]. Fifty microliters of EDTA-anticoagulated whole blood was stained with antibody mix containing 10 μL PerCP-Cy5.5 (clone 1A8), 10 μL FITC-Ly6C (clone HK1.4), and 10 μL PE-CD11b antibodies, incubated for 15 min at room temperature in the dark. The following isotype controls were used: PerCP/Cy5.5 Rat IgG2a, FITC Rat IgG2c, and PE Rat IgG2b. All antibodies were obtained from BioLegend (San Diego, CA, USA). Then, 1 mL red blood cell lysis buffer (Biolegend Red Blood Cell Lysis Buffer) was added and incubated for 10 min. Followed by centrifuge at 350g for 5 min, the supernatant was carefully aspirated without disturbing the cell pellet and the pellet was re-suspended using BioLegend Cell Staining Buffer. Unstained, single stained, and Fluorescence Minus One (FMO) controls were used for setting compensation and gating boundaries. The data analysis was performed using FlowJo software (Treestar, Ashland, OR, USA).

Fluorescein isothiocyanate-dextran assessment

At 4 h before euthanasia, the mice received fluorescein isothiocyanate (FITC)-dextran diluted in water (60 mg/100 g body weight) by oral administration. An intestinal loop model was used to assess the intestinal permeability of each gastrointestinal segment [60]. Briefly, after a midline laparotomy incision, the segments of the gastrointestinal tract, including the intestine, cecum, and colon, were created with two vascular hemoclips. The length of intestine between the two clips was injected with 50 μl FITC-dextran. After 1 h, plasma was collected immediately after sacrifice, and FITC-dextran measurement was performed on an LS55 fluorescence spectrophotometer (PekinElmer Life Sciences, Cambridge, UK), with excitation wavelength at 480 nm and emission at 520 nm.

Histology

To avoid a weaker scar and greater propensity for rupture early after MI, heart tissues of mice in the MI, sham MI+NS, and MI+PMB groups were harvested at 14 days of MI. Triphenyltetrazolium chloride staining was performed, and the entire heart based on apex sections was all assessed for infarct size. Heart tissues fixed with 4% paraformaldehyde were embedded in paraffin and sectioned. Tissue sections were stained with hematoxylin-eosin to determine the severity of myocardial inflammation. As described previously [61], a myocardial inflammation score at 0 means no myocardial fibrosis, 1 means very minimal focal subepicardial interstitial fibrosis just infiltrating beneath epicardial fat, 2 represents mild subepicardial interstitial fibrosis infiltrating deeper into the subepicardial myocardium, 3 represents multifocal subepicardial interstitial fibrosis, and 4 means replacement fibrosis. For immunohistochemical staining of F4/80-positive cells, tissue sections processed through deparaffinage, rehydration and antigen plerosised, and endogenous peroxidase activity blockade, were incubated with F4/80 antibodies at 4 °C overnight. Following secondary antibodies incubation, the sections were stained with avidin-biotin complex and counterstained with hematoxylin. Collagen volume fraction at the free wall of infarct area was assessed through Masson staining. Fixed colon tissues were embedded in OCT, stained with Occludin incubated with FITC-coupled secondary antibodies, and counterstained with DAPI. The stainings were examined under a fluorescence microscope (Nikon). For electron microscopic study, the superior mesenteric arteries were fixed in 2% glutaraldehyde/osmium tetroxide, after dehydration in ethanol and embedded in spurr resin. Thin sections were obtained from a microtome and viewed with a Hitachi 7500 transmission electron microscope (Hitachi Limited, Tokyo, Japan).

RNA isolation and real-time quantitative RT-PCR

Total RNA was extracted from colon tissues using Trizol reagent according to the manufacturer’s instructions. The concentration of RNA was quantified by a NanoDrop® ND-2000c UV–vis spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). Total RNA from each sample was reverse transcribed with the Superscript First-Stand cDNA Synthesis Kit (Invitrogen CA, USA). The quantitative RT-PCR was performed with SYBR Green I on an ABI Prism 7300 sequence detection system (Applied Biosystem, Foster City, CA, USA). β-actin was used as the internal control. The specific primer sequences were as follows: Occludin: sense: 5′-CCACCCCCATCTGACTATGC-3′, antisense: 5′-TCGCTTGCCATTCACTTTGC-3′, length 78 bp and β-actin: sense: 5′-CTAAGGCCAACCGTGAAAAG-3′, antisense: 5′-ACCAGAGGCATACAGGGACA-3′, length 104 bp. The relative mRNA level of each sample was analyzed and calculated using the 2^-∆∆Ct method.

Western blot

Proteins from colon tissues were extracted with cold lysis buffer. Following ultrasonication and centrifugation, the supernatant was harvested. Protein concentration was determined by the Bio-Rad protein assay kit (Bio-Rad Laboratories, Inc., Berkeley, CA, USA). Equal amounts of protein were fractionated on SDS-PAGE and transferred to nitrocellulose membranes (Millipore, USA). The membranes were blocked and incubated with primary antibodies against Occludin, and GAPDH. After binding with secondary antibodies, the protein bands were detected using a ChemiDoc™ XRS gel documentation system (Bio-Rad, Hercules, CA, USA).

Statistical analysis

Categorical data were compared with Fisher’s exact test. For comparisons between two independent groups, an unpaired Student’s t test or a Mann-Whitney U test was used. For comparison across three groups or more, one-way analysis of variance with Tukey’s post hoc analysis or a Kruskal-Wallis test followed by a Dunn’s test were performed. Correlation analyses were performed based on Spearman’s correlation. Kaplan-Meier survival analysis followed by the log-rank test was employed to estimate cumulative adverse event-free rates. To construct a new variable that integrate the impact of Δ LPS and d-lactate, we calculated their z-scores (standard z-transformation), which was calculated as: z = (x-μ)/σ, where μ is the mean value and σ is the standard deviation. The sum of two z-scores divided by 2 was defined as translocation z-score (mean = 0, standard deviation = 1). Receiver operating characteristic (ROC) curve was plotted to assess the accuracy and the optimal cut-off value (the best Youden Index: sensitivity + specificity − 1) for each parameter to discriminate between MACE (+) and MACE (−) patients. Parameters with an area under the curve (AUC) of P value ≤ 0.1 were then used for Kaplan-Meier survival analyses and Cox proportional hazards analyses. For Kaplan-Meier analysis, STEMI patients were stratified by Youden Index-derived optimal cut-off values. To construct optimal Cox proportional hazards model for MACEs, potential variables were first transformed into binary variables by using the optimal cut-off values, and then a stepwise forward selection was performed to select variables (P < 0.05) into multivariate model. Model calibration was performed by Hosmer-Lemeshow’s chi-squared test.

Mediation analysis models were constructed as previously prescribed [62] to assess whether and how much microbial translocation could explain the associations between inflammatory markers compromised LVEF and adverse outcomes. Linear and logistic models were used to estimate the associations between predictors and mediator and the associations between predictors, mediator, and outcome. The predictor variables were CD14++CD16+ subset measured on day 2 (continuous variable, per 1 SD increase), hs-CRP (continuous variable, day 2, per 1 SD increase), and reduced LVEF (binary variable defined by Youden Index, day 2); the mediator variable was translocation z-score (continuous variable); the outcome variable was MACEs (binary variable). We estimated the total effect (TE) and average direct effect (ADE) between the predictors and outcome, the average causal mediation effect (ACME) between predictors and outcome via mediator, and the percent of mediation effect, where TE equals to the sum of ACME and ADE. We used user-written command “medeff” in STATA to perform mediation analysis [63], and the results were calculated using bootstrapping with1000 iterations. The above statistical analyses were performed using STATA version 14.1 (STATA Corp., College Station, TX, USA) and GraphPad Prism version 5 (GraphPad Prism Software Inc., San Diego, CA, USA). A two-tailed P value < 0.05 was considered statistically significant.

QIIME software package (Version 1.7.0) was used to analyze alpha diversity and beta diversity. For alpha diversity, the Shannon index was calculated based on the genera profile of control, CHD, and STEMI patients. Then, for beta diversity, OTU table was used to generate unweighted UniFrac distance matrix, and PCoA was performed and displayed by WGCNA package, extra font package and ggplot2 package in R software (Version 2.15.3) (Additional file 9: Supplementary methods). Differential abundance of genera was tested by Wilcoxon rank sum test, and P values were corrected for multiple testing with Benjamin and Hochberg method.

From HMP database, the reference genomes isolated from human gut and oral were used to tracking the source of the STEMI-enriched bacteria. Based on NCBI taxonomy database, we can get the reference genomes’ taxonomic and phylogenic information. Based on the phylogenic information and the genera profile, the source of the genera was defined and the proportion of different groups was calculated (Additional file 9: Supplementary methods). For differential analysis, the P values between different groups were calculated by Wilcoxon rank sum test.