MS-grade methanol, formic acid, acetic acid, and acetonitrile were purchased from Fisher Scientific (Milford, MA, United States). Ultra-pure water (18.2 MΩ) was prepared with a Milli-Q water purification system (Bedford, MA, United States). Histidine, glutamine, and creatinine were purchased from Tokyo Chemical Industry (Tokyo, Japan). Five types of fatty acids were purchased from Larodan (Stockholm Sweden). Internal standards for 11 amino acids were purchased from Cambridge Isotope Laboratories (Andover, MA, United States). Other standards were purchased from Sigma-Aldrich (St. Louis, MO, United States). Detailed information on the standards including compound names, CAS names and sources are provided (Supplementary Table S1).

Phosphatidylethanolamine (PE; 12:0/13:0), lysophosphatidylcholine (LPC; 19:0/0:0), phosphatidylcholine (PC; 19:0/19:0), sphingomyelin (SM; d18:1/12:0), and ceramide (Cer; d18:1/17:0) standards were procured from Avanti Polar Lipids (Alabaster, AL, United States). Triglyceride TAG (15:0/15:0/15:0), free fatty acid (FFA; C19:0), and d3-FFA (C16:0) were purchased from Larodan, and d3-FFA (C18:0) was purchased from Sigma-Aldrich. All other chemicals and reagents used were of analytical grade. The Human ET-1 QuantiGlo ELISA Kit (lot QET00B), Human sICAM-1/CD54 Allele-specific Quantikine ELISA Kit (lot DCD540), and Human sVCAM-1/CD106 Quantikine ELISA Kit (lot DVC00) were purchased from R&D Systems (Minneapolis, MN, United States). An NO assay kit (lot A012-1-2) and a TXB2 assay kit (lot EM0232) were purchased from Nanjing Jiancheng Bioengineering Institute (Nanjing China), and a 6-Keto-PGFlα ELISA kit (lot 515211-96S) was purchased from Cayman Chemical (Ann Arbor, MI, United States). An ATP ELISA kit (lot CEA349Ge) was purchased from Cloud-Clone (Wuhan, China).

The patients were recruited from Xiyuan Hospital at the China Academy of Chinese Medical Sciences. Sixty CHD patients with stable angina pectoris (aged 48–75 years) were selected and diagnosed by Western medical standard diagnostic methods such as angiography or CT angiography. The patients were documented with >50% of coronary stenosis in at least one major coronary artery. Then, we have established TCM diagnostic and differentiation criteria for Shi syndrome and Xu syndrome. The subjects were diagnosed by two chief doctors, enabling scaling different symptoms into scores. We determined Shi and Xu based on the TCM syndrome scores of the two syndrome types (Supplementary Table S2). The individuals in the control group were free of CHD and angina-related symptoms. The studied participants provided details regarding to their age, sex, cardiac risk factors, and prior cardiac disease and underwent haematological and biochemical examinations. The Ethics Committee of the Medical Experimental Center of the Chinese Academy of Chinese Medical Sciences approved the collection of samples, and written informed consent was obtained from each subject.

Participants in this study were diagnosed with stable angina pectoris and CHD according to the criteria defined by the International Society of Cardiology, the WHO, the clinical nomenclature standardization joint special group report on “Naming and Diagnostic Criteria of Ischemic Heart Disease” (Cardiology, 1981), and the 2013 European Society of Cardiology guidelines for the management of stable coronary artery disease (Montalescot et al., 2013).

TCM-based diagnosis of the Shi and Xu subtypes of CHD was based on the Standard Program for the Diagnosis and Treatment (2014) of Coronary Heart Disease with Angina Pectoris in Chinese Medicine formulated by the State Administration of TCM of China (China and T.S.A.o.T.o, 2014). The diagnostic criteria of both syndromes included main symptom(s) in combination with at least one minor symptom. The main symptoms of Shi are chest pain and chest tightness, and the minor symptoms are fullness chest, hypochondrium and sighing. Nonetheless, the main symptoms of Xu are chest pain and chest tightness, and the minor symptoms are dizziness, insomnia, and dreaminess. At least five points were needed to place the patients in the Shi (TCM diagnostic criteria: main symptom(s) and at least one minor symptom ≥5); at least three points of differentiation were needed to place the patients in the Xu (TCM diagnostic criteria: main symptom(s) and at least one minor symptom ≥3). Moreover, as an alternative approach for subtyping, the two TCM doctors also classified those included patients into two subtypes based on the patient's tongue manifestation, sublingual vein, tongue coating and pulse manifestation status (Supplementary Table S2). Our classification of Shi and Xu based on differentiation criteria was in accordance with the diagnoses by the doctors according to their experience in checking the condition of tongues and pulses of all patients.

Following collection, blood samples were centrifuged for 10 min at 870×g at 4°C, and supernatants (serum) were frozen at −80°C for analysis.

Levels of the following serum biochemical parameters were analysed for all study participants: endothelin-1 (ET-1), nitric oxide (NO), 6-keto prostaglandin F1α (6-Keto-PGF 1α), soluble vascular cell adhesion molecule-1(sVCAM-1), soluble intercellular adhesion molecule-1 (sICAM-1), thromboxane B2 (TXB2), and adenosine triphosphate (ATP). ET-1 was analysed by chemiluminescence enzyme immunoassay using a LUMO Chemical Luminescence Immunity Analyzer from Auto Biological Company (Zhengzhou, China). NO was analysed by spectrophotometry using a model 721 spectrophotometer from Yili Fine Chemical Company (Shanghai, China). TXB2, sICAM-1, sVCAM-1, ATP, and 6-Keto-PGF 1α were analysed by ELISA using SpectraMAX M2 multi-mode microplate readers from Molecular Devices Company (Silicon Valley, CA, United States of America). Assays to detect the above five indicators were performed in 96-well ELISA plates. The ELISA kits included ELISA buffer, washing buffer, extraction buffer, substrate, standard enzyme conjugate, and antibody-coated plates. All assays were performed according to the manufacturers’ instructions.

Standard stock solutions at 1 mg/ml were prepared in methanol or water and stored at 4°C.

Untargeted serum metabolomics analyses were conducted by mixing 50 μL of serum with 150 μL of cold methanol containing the following standards: LPC (19:0/0:0) (37.5 ng/ml), PE (12:0/13:0) (500 ng/ml), SM (d18:1/12:0) (50 ng/ml), and phenylalanine-d5 (100 ng/ml). The mixtures of serum samples and standard solutions were vortexed for 1 min and centrifuged for 10 min at 14,000×g at 4°C. The supernatants were immediately analysed via ultra-performance liquid chromatography/ Time of Flight -mass spectrometry (UPLC/TOF-MS) (described below).

For targeted amino acid analyses, 4 µl of serum was mixed with a 991 μL of a solution of 1.7 mm ammonium formate in 85% acetonitrile containing 0.1% formic acid and 5 μL of the each following 11 internal standards: l-alanine-d4 (4.7 ng/ml), l-arginine-13C, d4 (10.8 ng/ml), l-aspartic acid-d3 (6.8 ng/ml), l-citrulline-d2 (8.9 ng/ml), glycine-13C, 15 N (19.3 ng/ml), l-leucine-d3 (6.7 ng/ml), l-methionine-d3 (7.6 ng/ml), l-ornithine-d2 (8.5 ng/ml), l-phenylalanine-13C6 (8.6 ng/ml), l-tyrosine-13C6 (9.4 ng/ml), and l-valine-d8 (6.3 ng/ml). The samples were mixed for 1 min and centrifuged for 5 min at 14,000×g at 4°C, after which the supernatants were analysed by ultra-performance liquid chromatography-mass spectrometry/mass spectrometry (UPLC/MS-MS) (described below).

For targeted lipid and fatty acid analyses, 10 μL of serum was mixed with 150 μL of cold methanol supplemented with the following standards: PC (19:0/19:0) (30 ng/ml), LPC19:0 (30 ng/ml), PE (12:0/13:0) (100 ng/ml), Cer (d18:1/17:0) (200 ng/ml), SM (d18:1/12:0) (40 ng/ml), TAG (15:0/15:0/15:0) (200 ng/ml), FFA 16:0-d3 (200 ng/ml), FFA 18:0-d3 (200 ng/ml), and FFA 19:0 (200 ng/ml). The samples were mixed for 30 s, after which 500 µl of MTBE was added, and the samples were agitated at room temperature for 20 min to fully extract the lipids. Next, 125 µl of water was added, and the samples were shaken and centrifuged for 10 min at 16,826×g at 4°C. A 100 µl volume of the sample supernatant was then dried using concentrator and resuspended in 200 µl of a solution of water:isopropanol: acetonitrile [5:30:65 (v/v/v)]. The samples were again shaken for 1 min and centrifuged as described above, and the supernatants were analysed by UPLC/MS-MS (described below).

One-millilitre volumes of each serum sample were pooled to produce a pooled quality control (QC) serum was prepared for each analysis as described above. This sample was used to validate the stability of the LC-MS system (described below).

A Xevo G2-XS QTOF-MS (Waters; Micromass MS Technologies, Manchester, United Kingdom) with an electrospray ionization (ESI) source was used in positive and negative ion modes in this study. Separation was achieved with a Waters Acquity UPLC HSS T3 column (100 mm × 2.1 mm, 1.7 μm). The mobile phases used for separation were acetonitrile (A) and water (B); both contained 0.1% (v/v) formic acid. The linear gradient elution settings were as follows: 0–3.5 min, 95–30% B; 3.5–11.0 min, 30–15% B; 11.0–12.5 min, 15–0% B; 12.5–14.0 min, 0%; 14.0–14.1 min, 0–95% B; and 14.1–16.0 min, 95% B. The flow rate was maintained at 0.3 ml/min during separation. The MS settings were as follows: desolvation gas at 800 L/h and 400°C; 50 L/h and 100°C for the cone gas and source temperature, respectively; and 2,000 and 40 V for the capillary and sampling cone, respectively. Mass data were acquired in MS^E mode; ramp collision energy at 10–60 V. A LockSpray™ source was used to ensure the reproducibility and accuracy of the MS data collection. The (M + H)+ and (M − H)− ions of leucine-enkephalin were set at m/z 556.2771 and 554.2615 for the lock mass in positive ESI+ and ESI- modes, respectively. The profiling data for each sample were acquired from 50 to 1,200 Da. Waters MarkerLynx software (Waters; Micromass MS Technologies, Manchester, United Kingdom) was used for data analysis to identify potential biomarkers of CHD patient subgroups. Peak finding, filtering, and alignment were conducted using Waters Progenesis QI Applications Manager (v2.3) (Waters; Micromass MS Technologies, Manchester, United Kingdom). The data collection parameters were as follows: retention time = 0.5–12.5 min; mass = 50–1,200 Da. Automatic mode was used for peak selection, and no specific masses or adducts were excluded. The MetaboAnalyst analytical pipeline (http://www.metaboanalyst.ca/) was used for multivariate statistical analysis of the resultant data (Xia and Wishart, 2011). Partial least squares discrimination analysis (PLS-DA) was used to visualize clustering and trends within the resultant data using MetaboAnalyst analytical pipeline.

Analyses of amino acids, lipids, and fatty acids were conducted using a UPLC system (Waters Acquity) with a quaternary pump, an autosampler, a degasser, and a Xevo TQ-S mass spectrometer with an ESI ionization source. For amino acids, a Waters Acquity UPLC BEH Amide column (2.1 mm × 100 mm, 1.7 μm; flow rate of 0.3 ml/min) was used for separation. The mobile phases for this separation step were 1 mm ammonium formate in acetonitrile containing 0.2% formic acid (A) and 1 mm ammonium formate in water containing 0.1% formic acid (B). The linear elution gradient settings were as follows: 0–0.5 min, 85% A; 0.5–5.5 min, 85–80% A; 5.5–12.5 min, 80–60% A; 12.5–13.0 min 60–85% A, and 13.0–15.0 min, 85% A. The column was warmed to 40°C, and a 5-μl injection volume was used. The column was washed twice between injections with a weak 90% acetonitrile solution and strong 10% acetonitrile solution.

Lipid and fatty acid separation were achieved using a Waters Acquity UPLC BEH C8 column (2.1 × 100 mm, 1.7 μm; flow rate of 0.26 ml/min). The mobile phases for this separation step were 60% acetonitrile containing 5 mm ammonium formate (A) and 90% isopropanol in acetonitrile containing 5 mm ammonium formate (B). The linear elution gradient settings were as follows: 0–1.0 min, 100% A; 1.0–2.0 min, 100–70% A; 2.0–12.0 min, 70–30% A; 12.0–12.5 min, 30–5% A; 12.5–13.0 min, 5–0% A; 13.0–14.0 min, 0% A; 14.0–14.1 min, 0–100% A; and 14.1–16.0 min, 100% A. The column was warmed to 55°C. A 1-μl injection volume was used in positive ion mode, while a 2-μl volume was used in negative ion mode.

Multiple reaction monitoring (MRM) and positive and negative ion modes were used while operating the mass spectrometer (Xevo TQ-S). Data acquisition and peak processing were conducted using Waters TargetLynx software (Waters; Micromass MS Technologies, Manchester, United Kingdom).

We tested the use of different mobile phases for untargeted metabolomics analysis of the patient serum samples and achieved the best peak resolution using a mixture of acetonitrile (A) and water (B), both containing 0.1% (v/v) formic acid. The data were monitored in positive and negative ion modes for multivariate statistical analysis to collect additional data. The desolvation gas flow rate was set to 800 L/h, while a desolvation temperature of 400°C and flow rate of 0.3 ml/min was used to remove the solvent introduced into the mass spectrometer, and shows representative base peak intensity (BPI) chromatograms for serum collected in negative and positive ESI modes (Figures 1A–F).

For lipid and fatty acid analyses, so we tested a range of injection volumes and extraction times of serum samples under the same conditions to ensure the reliability of the results while minimizing peak overload. After comprehensive consideration, We determined that a 10 μL volume and 20 min extraction period were ideal for achieving the goals of this study. Moreover, before the statistical analysis, we checked the extract ion chromatograms of each compound of each sample to ensure the absence of peak overload. The results of these comparative analyses are presented (Supplementary Figure S1). MRM mode was used for detection. The precursor-to-product ion pair, cone voltage (CV), and collision energy (CE) for each analyte are given (Supplementary Table S3). The details of targeted amino acid analyses are discussed in our previously published study (Guo et al., 2019).

We tested pooled QC samples in positive and negative ion modes to ensure LC-MS system stability throughout the duration of our analyses. These QC samples were processed for analysis in the sample manner as individual samples. The QC samples were injected every 10 injections and analyzed 9 times between samples to verify the stability of the LC-MS system.

Data were analysed using our in-house data analytics platform. This data platform contains information from publicly available data sources, omics study data, compound data, and algorithms to mine and connect different factors such as, compounds, text, ontologies, and pathway networks.

ANOVA was used to statistically assess differences in the levels of all tested metabolites obtained by targeted and untargeted analyses, and data were plotted using a histogram with means, standard errors, and 95% confidence intervals. We observed a normal distribution for these analytes.

In this study, we could subdivide CHD patients diagnosed by conventional Western medicine technologies into two metabolically distinct subgroups using the diagnostic principles of TCM. A total of 90 participants were enrolled in this study: 30 controls, and 60 patients diagnosed with CHD and angina pectoris. The 60 CHD patients were further separated into two equally sized (n = 30) subtype groups containing patients with either Xu or Shi subtypes based on TCM, which were diagnosed by two chief Chinese medicine physicians based on main and minor symptoms. Meanwhile, patients with diabetes, poor blood pressure control, severe chronic heart failure, severe arrhythmia, implanted pacemakers, or a history of myocardial infarction and those with liver, renal, haemorrhagic, autoimmune, haematological or mental disorders were excluded from the study group. The study participants provided details regarding their age, sex, cardiac risk factors, and prior cardiac disease and underwent haematological and biochemical examinations. The participant characteristics according to western clinical patrameters were shown in Table 1. No significant differences in age or sex of participants were found among the three groups. The urinary metabolites between male and female participants showed no significant difference (Supplementary Figure S2). On the other hand, clear difference between three groups (control, Xu and Shi subtypes) were observed according to TCM diagnose.

Validation of the LC-MS methods is shown by the overlapping base peak intensity (BPI) chromatograms for these QC samples (Supplementary Figures S3A–SB). The repeatability of the spiked internal standards was used to assess the overall technical variation (OTA) in our untargeted metabolomics analyses (Supplementary Table S4). The low relative standard deviations of retention time and peak areas indicated that the OTA was low and the reproducibility of these measurements over the experimental duration was acceptable. For targeted lipid, fatty acid and amino acid analyses, the majority of the coefficient of variation values corresponding to the internal standard peak area were <30% (Supplementary Tables S5,S6).

To establish untargeted and targeted metabolomics approaches to compare the serum metabolic profiles of the three groups, we used PLS-DA to analyse the data obtained in both ion modes (Figures 2A,B). We obtained a satisfactory classification indicating that CHD patients with stable angina pectoris including Shi and Xu subtypes exhibited metabolic states that were significantly different from those of the control patients. In addition, we observed separation between CHD patients with the Xu and Shi subtypes, indicating that the serum metabolic profiles of patients with these two TCM-based subtypes of CHD with stable angina pectoris differed. To validate the model, a permutation test was performed (n = 200), and the values of R2 and Q2 (Supplementary Figures S4A,B) indicate that the model showed great fitness and prediction efficacy. Together, these results show that good separation was achieved between these three groups. This observation indicated that CHD patients with the Xu and Shi subtypes disturbed the metabolism of endogenous substances and differently altered the serum metabolic fingerprint in CHD with stable angina pectoris compared to that under control conditions.

To find the potential biomarkers, we imported the LC-MS raw data into Progenesis QI software, more than ten thousands of features were detected in the untargeted metabolomics study. We next identified the metabolites in our analysed samples. Initially, MassLynx software was used to determine the elemental composition of the identified compounds, after which databases including the Kyoto Encyclopaedia of Genes and Genomes (KEGG) (http://www.genome.jp) and HMDB (http://www.hmdb.ca) databases were searched to identify the structures of these potential biomarkers based upon high-resolution MS and MS/MS spectra data. The identities of these metabolites were further confirmed after comparing them to defined standards on the basis of retention time and MS/MS fragmentation patterns (Supplementary Table S7).

To gain further insights into the hydrophilic and hydrophobic metabolite profiles after non-targeted analyses, we conducted targeted analyses of serum amino acids, lipids, and fatty acids. While in the targeted lipid, fatty acid, and amino acid metabolomics experiments, there were 141 lipids, 23 fatty acids, and 22 amino acids detected, respectively. To identify the metabolic potential biomarkers specific for the two CHD subtypes, we used independent samples t-tests to compare the data, using p < 0.05 as the significance threshold. Significant differences in the abundance of these metabolites were further confirmed in Table 2. As shown in Table 2, when comparing the control and the Shi CHD subtype, levels of the long-chain unsaturated lipids ceramides (Cers) and sphingomyelins were significantly lower in the Xu CHD subtype, while levels of glycocholic acid, taurocholic acid, arachidonic acid, histidine, uric acid and hypoxanthine were higher in the Xu CHD subtype. Shi-subtype CHD patients show increased levels of LPC (20:2) and downregulated levels of citric acid, linolenic acid, cysteine and glutamine, compared with the control and the Xu CHD subtype. We organized the putative metabolic potential biomarkers into a heat map to show their distribution (Figure 3). We next explored the potential pathways involved in regulating the differentially abundant metabolites detected in the above analysis using MetaboAnalyst, with those pathways that had low p-values and high pathway impacts (Figures 4A,B). Metabolites in Shi-subtype CHD patients were primarily associated with the following metabolic pathways (Figure 4A): Aminoacyl-tRNA biosynthesis, Purine metabolism, Alanine/aspartate and glutamate metabolism, Glyoxylate and dicarboxylate metabolism, Cysteine and methionine metabolism. In addition, 21 metabolic pathways were identified (Supplementary Table S8). Metabolites in Xu CHD subtype patients were primarily associated with the following metabolic pathways (Figure 4B): Taurine and hypotaurine metabolism, Aminoacyl-tRNA biosynthesis, Sphingolipid metabolism, Purine metabolism, Alanine/aspartate and glutamate metabolism, Glyoxylate and dicarboxylate metabolism. In addition, 22 metabolic pathways were identified when comparing control subjects. For full details of affected pathways (Supplementary Table S9). Detailed metabolite differences between Xu and Shi subtype CHD patients were shown (Supplementary Table S10).

The metabolite-related protein interaction network was constructed through the STRING database (minimum required interaction score >0.4). A regulatory network containing these indentified pathways, proteins, and significant molecular metabolites (Figure 5) was constructed using Cytoscape 3.8.2. Significant interactions between 6-Keto-PGFlα, NO, ET-1, sICAM-1, sVCAM-1, TXB2, ATP, and identified significant molecular metabolites, metabolic pathways were examined. To investigate how vascular endothelial function and energy metabolism were affected in the different CHD subtypes, a set of endothelial markers such as 6-Keto-PGFlα, NO, ET, sICAM-1, sVCAM-1, TXB2, and energy marker, ATP were examined. We observed significant reductions in 6-Keto-PGFIα and ATP levels in CHD patients relative to controls (Table 3); ET-1 levels were significantly higher in CHD patients as compared to those in control. Although sICAM-1 and sVCAM-1 levels were not significantly different from control levels in patients with Shi-type CHD, there was a significant increase in these sICAM-1, sVCAM-1, and ET-1 levels in Xu-type CHD patients relative to those in patients with Shi-type disease. In addition, ATP and 6-Keto-PGFIα levels were significantly lower in patients with Shi-type disease as compared with patients with Xu-type CHD. These results thus suggest that vascular endothelial function is impaired in CHD patients with stable angina and that these impairments are more severe in patients with Xu-type CHD as compared with patients with Shi-type CHD. All the differences between XU and SHI subtypes were mapped. We identified alterations in three metabolic differences that were characteristic for the Xu subtype: the long-chain unsaturated lipids Cers and sphingomyelins metabolism, bile acid metabolism and vascular endothelial dysfunction. For the Shi subtype, we also could identify a Shi type specific CHD fingerprint, Shi patients showed inflammation, increased bioactive phospholipids and disrupted energy metabolism pathways (Figure 6).