This is a cross-sectional study, where recruited patients were individuals who had been hospitalized for coronary angiography in our center, as previously described [12]. This study was approved by the local Medical Ethics Committee of Peking Union Medical College Hospital (PUMCH) (JS-1195). Prior to their participation in the study, all subjects provided written informed consent. CAD definition: $\ge$50% stenosis in at least one main coronary artery. Patients were stratified into the following subgroups: (1) MI: defined as a rise of cardiac troponin with at least one value above the 99th percentile upper reference limit and with at least one of the following: (i) ischaemia symptoms; (ii) new or presumed new significant ST-segment-T wave changes or new left bundle branch block; (iii) development of pathological Q waves in the electrocardiogram; (iv) imaging evidence of new viable myocardium loss or new regional wall motion abnormality; and (v) identification of an intracoronary thrombus by angiography or autopsy [17]; (2) Stable coronary artery disease (SCAD): based on the presence of chest pain that did not change in pattern in the preceding 2 months [18]; and (3) Control: adolescents who exhibited negative results upon coronary computed tomography angiography (CCTA) or coronary angiography examination or were identified as having no CAD-related clinical signs and symptoms. The inclusion and exclusion criteria were previously summarised in accordance with the established criteria [12]. Subjects were excluded if they were diagnosed with gastrointestinal diseases, tumors, autoimmune/infectious diseases, a history of an abdominal operation in the previous year, or had been administered antibiotics for more than three days in the previous three months.

The burden of coronary atherosclerosis was estimated using the Gensini and Syntax scores, as previously reported [12]. Blood pressure was measured with a standard mercury sphygmomanometer when the patient was in a seated position after $\ge$5 min of rest. Participants’ body mass index (BMI) was calculated as weight (kg) divided by squared height (m²). For the MI patients, peripheral blood was drawn from the radial or femoral artery. For the SCAD and controls, peripheral fasting blood was drawn in the morning the day after admission. Peripheral blood samples were centrifuged at 3000 rpm for 5 min and the supernatant was purified. All information provided during the interviews was recorded in the case report form.

We provided a stool sampler and detailed instructions for sample collection. Freshly collected stool samples were immediately frozen at –80 °C. DNA extraction was performed using the Qiagen QIAamp DNA Stool Mini Kit (51504, Qiagen, Hilden, Germany) according to the manufacturer’s instructions. DNA quantity was determined using a NanoDrop spectrophotometer, Qubit Fluorometer (with the Quant-iT™dsDNA BR Assay Kit, Q33130, Eugene, OR, USA), and gel electrophoresis.

DNA library construction was mainly performed following the manufacturer’s instruction (Illumina, San Diego, CA, USA). We constructed one paired-end (PE) library with an insert size of 350 bp, followed by high-throughput sequencing with PE reads of length 2 $\times$ 100 bp. High-quality reads were obtained by filtering low-quality reads with ambiguous “N” bases, adapter contamination, and human DNA contamination from the Illumina raw reads, and by trimming low-quality terminal bases of reads simultaneously [19].

Employing the same parameters to construct the Metagenomics of the Human Intestinal Tract (MetaHIT) gene catalogue [20], we performed gene prediction using SOAPdenovo v1.06 (Beijing Genomics Institute, Hongkong, China) [21] and GeneMark v2.7 (Georgia Institute of Technology, Atlanta, GA, USA) [22]. Genes were aligned using BLAT (BLAST-like alignment tool, Kent Informatics, Inc, Mountain View, CA, USA) and genes with over 90% of their length aligned to another with more than 95% identity were removed as redundancies.

We performed taxonomic assignment using an in-house pipeline, we collected the microbial reference genomes from the Integrated Microbial Genomes (IMG) database to make alignment. We used 85% identity as the threshold for genus assignment, and threshold of 80% for the alignment coverage based on sequence similarity across phylogenetic ranks by MetaHIT [23]. For genes, the highest scoring hit(s) above these two thresholds were chosen for the genus assignment.

We calculated $\alpha$-diversity using the Shannon index based on genus level in order to estimate the microbiome richness. A high $\alpha$-diversity normally indicates abundant richness of genera within the sample. Principal component analysis (PCA) was analysed on the genus level, and distance-based redundancy analysis (dbRDA) was calculated using Bray–Curtis distance as previously described [24].

Differentially enriched KEGG (KEGG database release 59.0, genes from animals and plants removed) pathways were identified according to their reporter score as follows. Wilcoxon rank-sum tests were performed on all KOs of our samples and adjusted for multiple testing using the Benjamin–Hochberg procedure. An absolute reporter score value higher than 1.6 (95% confidence on either tail, according to normal distribution) was used as threshold.

For instance, leucine biosynthesis-related KEGG pathway details were obtained from KEGG database ‘M00432’ (list of KO genes) under the ‘Valine, leucine and isoleucine biosynthesis’ category. Prevalence of genes higher than 5% were analyzed for any of the stages (SCAD, MI) compared to the controls. Pathways shown in our results were manually modified according to KEGG database or referring to the literature. We showed KO genes with significant differences between group comparisons (p 0.05).

Metabolomics analysis was performed on a Waters ACQUITY ultra-high-performance liquid chromatography system (Waters Corporation, Milford, MA, USA) coupled with a Waters Q-TOF Micromass system (Waters Corporation, Milford, MA, USA) as previously described [25, 26]. We used databases such as KEGG, MetaboAnalyst, Human Metabolome Database, and METLIN to identify metabolic pathways. Next, SIMCA-P 14.0 software (Umetrics AB, Umea, Sweden) was implied to acquire metabolite information and variables.

A total of 7061 serum metabolic features were yielded after pre-processing above. We conducted a “cross-comparison scheme” as follows to downsize metabolic features associated with disease status. (i) Multi-comparisons set A: various stages of CAD were compared with healthy control (HC) and to each other using the Wilcoxon test and T test, the metabolites will be retained for the next stage of analysis both statistics results satisfied the adjusted p value (q value) 0.05 meanwhile the fold change $>$1.2 or 0.83; (ii) Multi-comparisons set B: compare the metabolites between the four groups and HC vs. CAD, metabolites were reserved when any test of Jonckheere-Tespstra test or Kruskal. test satisfied p value 0.05. We then integrated the two collections, serum metabolomics profiling yielded 562 features after removing the drug metabolites.

We cluster metabolites into metabotypes using the R package WGCNA [27] as previously described. For parameters, scale-free topology threshold was set for $\beta$ = 14 and deepSplit threshold was set for 4 [28]. The serum metabolites clusters (labelled M01–M35) were collectively termed serum metabotypes.

All standards were obtained from Sigma-Aldrich (St. Louis, MO, USA), Steraloids Inc. (Newport, RI, USA) and TRC Chemicals (Toronto, ON, Canada). All the standards were accurately weighed and prepared to obtain individual solutions at a concentration of 5.0 mg/mL. Formic acid was of analytical grade obtained from Sigma-Aldrich (St. Louis, MO, USA). Methanol (Optima LC-MS, Thermo Fisher Scientific, Waltham, MA, USA), acetonitrile (Optima LC-MS), and isopropanol (Optima LC-MS, Thermo Fisher Scientific, Waltham, MA, USA) were purchased from Thermo-Fisher Scientific (FairLawn, NJ, USA). Sample preparation was performed as previously described [26]. Targeted quantitation was performed to determine the concentration of metabolites as previously described by reverse-phase UHPLC using a Prominence 20 UFLCXR system (Shimadzu, Columbia, MD, USA) with a Waters (Milford, MA, USA) BEH C18 column (2.1 $\times$ 100 mm $\times$ 1.7 µm particle size).

Germ-free (GF) ApoE -/- mice were provided by the Institute of Laboratory Animal Sciences (ILAS) at the Chinese Academy of Medical Sciences and Peking Union Medical College [research license no. SYXK (Beijing) 2015-0035], which is a member of (and accredited by) the American Association for the Accreditation of Laboratory Animal Care. All experiments were performed in accordance with the guidelines of the Institutional Animal Care and Use Committees of the ILAS. The mice were maintained under standard GF conditions with a 12:12-hour light:dark cycle. GF mice were maintained in flexible film isolators and their GF status was checked weekly by aerobic and anaerobic culture. All GF ApoE -/- mice were male. All mice were used after reaching the age of 6 weeks. All GF ApoE -/- mice received sterilized high fat diet (40 kcal% fat and 1.25% w/w cholesterol; product number D12108S) which was purchased from Changzhou SYSE Bio-Tec. Co., Ltd (Changzhou, Jiangsu, China).

Bacteroides thetaiotaomicron (ATCC 29741) and Prevotella copri (control strain, ATCC 33547) were cultured in tryptic soy agar/broth supplemented with defibrinated sheep blood (ATCC medium 260) under anaerobic conditions. Strains were harvested in logarithmic phase, aliquoted using fresh broth and concentrated to 10⁹ CFU/mL and diluted into 10% glycerol under anaerobic conditions and stored at –80 °C until use. GF ApoE-/- mice were kept in individual isolators and received a high-fat diet. In the next three weeks, the mice were gavaged twice a week, each time with 100 µL of bacterial solution containing B. thetaiotaomicron or P. copri (5 $\times$ 10⁸ CFU/mouse).

Measurement of acute thrombus formation were performed as described earlier [29]. Some mice did not survive meaning there were ultimately 8 mice in the B.t group and 7 mice in the Prevotella copri (P.c) group. After the mice were anesthetized, the left jugular vein and right carotid artery were carefully dissected. The mice were injected intravenously with Rhodamin B to label the platelets. The ferric chloride injury model was performed by placing 7.5% FeCl₃ solution for 1 min laterally to the common carotid artery. Then, the resulting thrombus formation and occlusion of the artery was recorded using a high-speed wide-field Leica M205 FCA fluorescence stereo microscope with Leica Application Suite X (LAS X) software (Leica Microsystems GmbH, Wetzlar, Germany). We recorded the occulusion time. The experiment was terminated if the carotid artery did not occlude within 30 minutes.

Mouse platelet-rich plasma (PRP) was prepared as described previously with minor modifications [30, 31, 32]. Mouse whole blood ($\approx$900 µL) was taken by orbital and collected into a sodium citrate blood collection tube and gently shaken upside down 8 times. PRP ($\approx$600–700 µL) was separated by centrifuging at 200 g for 15 min. The pellet was resuspended in 500 µL HEPES Tyrode’s buffer (pH 7.4) and centrifugation was repeated (to fully remove other blood cells) at 200 g for 5 min (room temperature), then the supernatant was aspirated.

The final platelet suspensions (100 µL PRP dilute to 700 µL with HEPES Tyrode’s buffer (pH 7.4); 2 $\times$ 10⁷ platelets/mL) were recovered for 30 min at room temperature prior to the experiments. Washed platelets for experiments were used within 4 h after isolation. Agonists of collagen were added separately at different concentrations (10 µg/mL, 15 µg/mL and 20 µg/mL) and a blank was kept as a control. The platelets were incubated with agonists for 10 min at room temperature. APC-conjugated anti-P-selectin (CD62P, 148304, Biolegend, San Diego, CA, USA) was added to each tube and incubated in the dark for 15 min. The platelet suspensions were then fixed with 100 µL of 4% paraformaldehyde. Data was acquired on a flow cytometer (Accuri™ C6 plus, BD Biosciences). Twenty thousand (20,000) events were acquired. The data was analyzed using FlowJo v10.4 software (BD Biosciences, San Jose, CA, USA).

For the platelet spreading functional study, we largely followed procedures as previously described [31]. Microfluidic shear flow experiments were performed using the rectangular flow chamber assay (31-010, Glycotech, Rockville, MD, USA) equipped with single channel syringe pump NE-1000 and flexcell VP750 vacuum pump. Briefly, cover glasses were degreased (24 by 60 mm) by incubating overnight in undiluted chromosulfuric acid. Then the glass coverslips were coated with collagen (100 µg/mL; C7661-10MG, Sigma, St. Louis, MO, USA) and fibrinogen (100 µg/mL; F3879, Sigma). The coated coverslips were blocked with Tyrode’s HEPES buffer containing 1% bovine serum albumin (BSA). The rested and washed platelets were stained by FITC labeled anti-CD41Mab (1:100; 133903, Biolegend) in the dark at room temperature for 30 min. Unbound antibodies were washed away and prostacyclin was added at a final concentration of 10 ng/mL to prevent platelet activation. The platelet pellet was reconstituted in HEPES Tyrode buffer to achieve a platelet concentration of 150,000 platelets/µL. After the incubation, platelet samples were perfused over chips coated with or without agonist at a physiological shear rate (7.5 µL/min) using a microfluidic device for 20 min. Immediately after performing the flow experiments, samples were fixed for 15 minutes in 4% paraformaldehyde in PBS at room temperature in the dark. Microscopic images were acquired using a Nikon (Tokyo, Japan) confocal microscope with a 60$\times$ oil immersion objective. Image analysis was performed using ImageJ software (1.53a, Wayne Rasband, USA) predominantly as described previously [33]. In short, 15 consecutive images along the length of the channel and covering the proximal, middle, and distal regions of the flow channel were acquired, converted into 8-bit grayscale images, and underwent adjustment for brightness and contrast. Images of stained platelets were quantified using ImageJ software. Intensity threshold was chosen to select for specific staining and quantified for integrated optical density.