Circulating Metabolomic Proling Detects Novel Markers of Myocardial Infarction among Cardiac-chest Pain Cases

Background: A unique human plasma metabolic pattern of myocardial infarction (MI) has been recognized for over a decade. The discriminant metabolites are potential diagnosis and warning markers, and have connections with traditional cardiovascular risk factors and circulating transcriptomics. Methods: A total of 146 chest pain (85 MIs and 61 non-MI chest pain cases) and 84 control individuals were recruited. GC/MS and LC/MS were applied to identify and quantify the plasma metabolites. The metabolites were analyzed by multivariate statistical analysis to describe different patterns of CADs with chest pain. Receiver operating characteristic (cid:0) ROC (cid:0) and odds ratio (OR) were calculated to identify potential markers of MI cases, and the markers were further evaluated based on human circulating transcriptomics from the GEO database. The correlation of metabolites to clinical indices was assessed. Results: MI can be clearly distinguished from non-MI chest pain cases by the metabolites we identied, better than by medical history combined with routine biochemical parameters. Pathway analysis highlighted an upregulated methionine metabolism and downregulated arginine biosynthesis that were deeply involved in MI cases. Among all the metabolites, deoxyuridine (dU) and methionine scored highly in ROC (>0.91) and are related to LDH and AST (p<0.05). OR values (logOR>2) suggested, with or without adjusting by lipid levels, age, smoking, diabetes and hypertension, that dU and methionine are risk factors for alerting about MI occurrence. Transcriptomics data showed elevations of plasma CDA and UPP1, and a decrease of cardiac MARS, which are consistent with the increased levels of plasma dU and methionine, respectively. Conclusions: For the rst time, dU and methionine were suggested as markers for MI diagnosis and as high-risk markers to alert about an onset of MI. Plasma metabolite changes are not only directly related to cardiac damage but also could be related to circulating cells.


Background
A strangling feeling in the chest is a typical manifestation of coronary artery disease (CAD). In most cases, CAD develops as plaque builds up on the artery walls. When it progresses to myocardial infarction (MI), coronary heart disease will be life-threatening and extremely dangerous in the ensuing days or weeks due to its various fatal complications. However, chest pain can be caused by other cardiovascular events (e.g., unstable angina) or other, non-cardiac cases (e.g., gastroesophageal re ux disease). Thus, noticing early warning signs to alert patients about the onset of MI, taking precautions and recognizing MI are among the primary focuses of the management of atherosclerotic coronary artery disease, which is the main cause of coronary heart diseases.
Diagnostic and predictive biomarkers of MI are of crucial importance. Among all cardiac damage biomarkers, such as CK-MB, AST and cardiac troponins (cTnt), cTnt is the preferred MI biomarker clinically. However, some studies showed that an unexpected elevation of this cardiomyocyte protein was observed without obvious connection to cardiac injury. Hence, clinicians need to carefully determine whether troponin elevation is direct proof of myocardial necrosis, especially when the underlying reason for its release is unclear. [1][2][3] More potential biomarkers are urgently needed to facilitate the diagnosis of, as well as for preferably alerting about and predicting MI onset. CAD is a lifestyle disease, and the abnormal abundance of circulating metabolites surely re ects the metabolic state of CAD patients. Medical history records and laboratory tests, e.g., sex, age, smoking history, function of the primary organs, and biochemical assays, are important factors worthy of consideration in identifying diagnostic markers and clarifying the underlying mechanism involved. Since ischemic heart diseases are characterized by profound metabolic shifts at both the circulatory and local levels [4], metabolomics has been applied to study the metabolic pattern changes detected in the plasma or serum of CAD patients. Early in 2002, a pioneering work was published showing that NMR-based metabonomics had the potential to rapidly and noninvasively diagnose the presence and severity of coronary heart disease [5]. In 2005, Marc Sabatine and his colleagues identi ed metabolic biomarkers of myocardial ischemia associated with physical exercise [6]. Later studies focused on identifying biomarkers and metabolic pathways and exploring the underlying mechanisms associated with cardiovascular diseases [7][8][9][10][11][12][13]. Excitingly, a panel of potential markers has been suggested for coronary heart diseases, such as arginine and homocysteine, and the underlying mechanisms of their action have been explored [14][15][16]. Unfortunately, few publications have focused on the diagnosis and prediction of MI onset among CAD patients with chest pains, although the occurrence of MI is deeply in uenced by whether an individual has traditional risk factors, such as male sex, diabetes, tobacco smoking and hypertension.
In this study, a metabolomic platform with GC/MS and LC/MS instrumentation was employed to pro le plasma metabolites of hospitalized CAD patients with chest pain (including MI and non-MI chest pain cases) and their controls. Metabolic patterns were evaluated, and metabolic markers were screened and described based on semi-quantitative data, AUC, odds ratios (OR) [17] and their connections with well-recognized cardiovascular disease risk factors.
We aimed to identify metabolic markers to facilitate the diagnosis and early prediction of MI among chest pain patients. The mechanisms behind the biomarker changes are also brie y discussed.

Human plasma samples
A total of 146 chest pain cases highly suspected of MI, and 84 individual controls were recruited from October, 2017 to March, 2018, with exclusions criteria of fever, hepatic decompensation, renal failure, cancers, endocrine and hematological diseases. The blood samples of chest pain cases were collected within 48 hours since the symptom was rst reported and fasting for at least 6 hours. All control cases were fasting for at least 6 hours too.
After routine diagnostic procedures, including evidence from ECGs, serum infarction biomarkers, computed tomography angiography or coronary angiography, 85 were later con rmed to be NSTEMI or STEM as the MI cases (MIs) and the remaining 61 were con rmed to be non-MI chest pain cases (non-MIs). All the non-MIs include 34 unstable angina (UA) cases, and 27 other non-MI cardiac cases (non-MICs), including myocarditis, valvular heart diseases etc.
The venous blood samples were collected from fasting state volunteers in EDTA-Na anticoagulated tubes in the morning. Within 2 hours, blood samples were centrifuged at 1000 g for 5 min, and each of the supernatant plasma was transferred to another tube, frozen at -80℃ in refrigerator. Before use of the plasma samples, they were thawed by incubation at 37 ℃ bath for 15 min, vortexed and centrifuged at 650 g for 5 min.

Equipment for blood examinations
The instrument blood count and blood biochemistry, model xn-10, is made in Hyogo, Japan. Both NT-proBNP and serum infarction marker (cTns) are Roche company of Germany, model Cobas 6000, produced in Mannheim, Germany. The biochemical instrument is Beckman Coulter of the United States, model au5800, produced in Shizuoka, Japan.

GC/MS and UPLC-QTOF/MS analysis
A well-developed metabolomic platform was utilized to pro le metabolites in plasma samples of the clinic patients. For the analysis in GC/MS system((Shimadzu GCMS-QP2010 Ultra, Kyoto, Japan), the plasma samples were pretreated, extracted, and derivatised in a similar way to that described previously [18], Method S1. To minimize systematic variations, the samples were randomly analyzed with the quality control (QC) samples inserted, and the intensity of internal standard(IS), myristic-1,2-13 C 2 acid, and an external reference standard, methyl myristate was evaluated. The raw GC/MS data were processed using a post-run analysis software(Shimadzu GC/MS solution 2.0), each of the metabolite was identi ed and the quantitative data was acquired in the same way as reported [19].
For the analysis in UPLC-QTOF/MS system (AB SCIEX TripleTOF® 5600, Foster City, CA), the plasma samples were pretreated in the same way as in GC/MS analysis with a few modi cations, which used the other internal standard of 5-13 C-glutamine dissolved in methanol at 15 µg/ml. The UPLC-QTOF/MS analysis was carried out as previously reported [20]. The detail information and protocol for GC/MS and UPLC-QTOF/MS is available in the Supplementary material, method S2.

Multivariate statistical analysis
After acquirement of the data in GC/MS and UPLC-QTOF/MS,the detected peaks/identi ed metabolites in each sample were aligned, quantitated, and all the samples data constructed a data matrix, where the missing data were excluded as default during analysis. After normalization against the IS, the data was evaluated using SIMCA P14.1 software (Umetrics, Umeå, Sweden) [19], Method S3. Concisely, principal component analysis(PCA), partial least square to latent structure discriminant analysis (PLS-DA), orthogonal PLS-DA(OPLS-DA) models were calculated to show the clustering or separation of samples from different groups For PLS-DA modeling, samples from the different groups were classi ed such that all samples were divided into different groups (e.g., MIs, non-MIs, controls, etc.) as the qualitative 'dummy' variables, Y. The goodness of t for the models was evaluated using three quantitative parameters: R 2 X and R 2 Y were the explained variation in X and Y, respectively, and Q 2 Y was the predicted variation in Y. Permutation test was assessed for model validation, where the higher level of R 2 Y and Q 2 Y, and lower value of the intercept of R 2 (lower than 0.2) and Q 2 (lower than 0.0) suggested good model and prediction ability.

Discriminant metabolites and statistical analysis
After normalization against the IS, all the semi-quantitative data from both GC/MS and UPLC-QTOF/MS was logarized so that the state probabilities of the data queue tended to a normal distribution. The discriminant metabolites between groups were screened and chosen based on Variable Importance (VIP) using SIMCA-P 14.1, and the independent sample t-test of the logarized data using SPSS (version 23.0, SPSS Inc., Chicago, IL, US).
Metabolic pathway enrichment and topology analyses were performed using Metaboanalyst 3(https://www.metaboanalyst.ca/), Method S4. The KEGG ID of discriminatory compounds were uploaded and embedded in human pathway library for pathway analysis and hypergeometric tests, with the pathway analysis algorithms of Fisher's Exact Test, a pathway topology algorithms of Relative-betweeness Centrality, and KEGG pathway library version of Homo sapiens. Potential biomarkers were identi ed based on the identi ed metabolic pathways in databases such as KEGG (http://www.genome.jp/kegg/), human metabolomic data base (HMDB), (http://www.hmdb.ca/) and LipidMaps (http://www.lipidmaps.org/).
For the data inconformity with normal distribution from clinical assaying, a nonparametric test (Mann-Whitney U test, two-sided) was employed to evaluate statistical signi cance. ROC analysis and OR calculation were performed using SPSS as well. Before computing OR value, the logarized data of a metabolite(i) for each subject of ORi was normalized by subtracting mean value of ORmean within this group, and then divided by the standard deviation(SD) within the group, shown as the normalized ORs=(ORi-ORmean)/SD.

Transcriptomics data
We studied open access transcriptomics data from GEO. The human myocardial infarction plasma data are from GEO accession GSE48060. Mice myocardial infarction data are from GSM12346. Date were directly plotted without transformation. Tables 1 and 2 show the history examinations and basic laboratory tests of the volunteers. Generally, higher glucose, AST, LDH, HBDH, and CK levels and lower ALB and Ca 2+ concentrations were detected in the chest pain patients than the control group. Among the 146 chest pain inpatients, approximately 65% had taken aspirin and statin treatments before blood collection. As a result, TC, TG, LDL-c and HDL-c levels were all lower in chest pain inpatients than the controls.  Values are presented with Mean ± SE. Mann-Whitney U test was applied to produce P value between MI and non-MI chest pain cases.

Clinical description of chest pain individuals by PCA and OPLS-DA plots
Based on clinical parameters (listed in Table 1 "variables"), including "biochemical items", "demographics" and "cardiac risk factors", an unsupervised PCA score plot was created. The model indicated a few outliers when the samples were either divided into 3 (controls, MIs, non-MIs) or 4 groups  (Table S1-2). Quantitative data were acquired for each metabolite in the plasma samples of the control, MI, UA and other non-MI cardiac cases.
Based on the metabolomic data derived from GC/MS and LC/MS analysis, the PCA score plot again showed a few outliers when the samples were divided into 3 or 4 groups, as indicated above. Unlike with the clinical data, unsupervised PCA analysis of metabolomic data showed that the majority of MIs deviated from the others, regardless of whether the 3 or 4 groups were de ned, although the control, non-MICs and UAs overlapped with each other to some extent (Fig. 1B1, 1B2). The supervised PLS-DA model revealed that samples from each group clustered closely and anchored away from the other groups when the samples were divided into 3 groups (Fig. 1B3). When the samples were divided into 4 groups, the majority of MIs and controls clustered separately, while the UAs and non-MICs primarily overlapped with each other, with a minority overlapping with MIs and controls (Fig. 1B4). The distant separation of MIs from the other groups suggested distinctly different metabolic patterns between MIs and the groups of UA and non-MICs, while the overlapping of the groups suggested similar plasma metabolic patterns between UA and the other non-MICs. In general, metabolomic data better characterized MIs than history examinations and laboratory tests, and the score plot of non-MI chest pain cases (including UA and non-MI cardiac cases) indicated that they had moderate metabolic perturbation relative to the MI cases because they anchored between MI and the controls (Fig. 1B3). The above data suggest that subgroups of MI can be recognized by multivariate analysis of identi ed plasma metabolites more effectively than by routine clinical parameters.

Pathway analysis of differential metabolites
OPLS-DA analysis showed different metabolomic patterns of the non-MIs from the controls (Fig. 2A1). Statistical analysis suggested 50 discriminant metabolites (p < 0.05) that differentiated non-MI chest pain inpatients from the controls (Table 3). Similarly, MI cases primarily showed different metabolomic patterns from non-MIs (Fig. 2B1). According to the statistical analysis and the VIP values, 54 discriminant metabolites were identi ed between MIs and non-MIs (Table 3). The data was not logarized, and expressed as Mean ± SE. Fold change (FC) were calculated by the original, non-logarized data directly. Statistical signi cance was evaluated by using T-test with equal-variance of two tails, after ANOVA assessment of the variance, ***, **, *: p < 0.001, 0.01, 0.05, respectively, between MI and non-MI chest pain cases; ###, ##, #:: p < 0.001, 0.01, 0.05, respectively, between non-MI chest pain cases and the controls. '/' represents the statistical signi cance of p values more than 0.05.
Of the 50 metabolites differentiating non-MIs from the controls, levels of gluconic acid and isoleucine were higher in non-MIs, while succinate, inosine, and arginine were lower, and all the above metabolites deviated further in MIs as the cardiac damage became more severe (Fig. 3C, 3D). These ndings indicate that the above metabolites are involved in the development of cardiac damage.
Although glycerol, salicylic acid, and deoxyadenosine showed signi cant differences between the non-MI and control groups, they had no signi cant difference between the MI and non-MI cardiac groups. These are thus suggested as markers of non-MI chest pain. In addition to endogenous metabolites, we found that salicylic acid, an exogenous metabolite, also characterized the group of chest pain cases. Salicylic acid is the primary metabolite of aspirin, and a review of inpatient information and clinical data revealed that a large portion of chest pain patients had taken aspirin for the management of CAD.
Moreover, deviated levels of dU, methionine, homoserine, etc. were only observed in MI cases, but not between the controls and non-MIs, indicating their association with the development of MI (Fig. 3B).
A Venn diagram was created to show the discriminant metabolites between MI and non-MIs and those between non-MIs and controls. The overlapping region (B) in the Venn diagram (Fig. 3E) lists the metabolites screened out in both of the two comparison groups, suggesting that they were mostly likely risk factors or markers of the occurrence and development of MI, re ecting homeostatic disturbance induced by myocardia hypoxia. Figure 3f shows the pathway analysis of the metabolites in the Venn A + B region (control vs non-MI), and Fig. 3G shows the pathway analysis of discriminant metabolites in the Venn B + C region (MI vs non-MI). Generally, arginine biosynthesis and pyrimidine metabolism were the most signi cantly altered metabolic pathways in non-MI chest pain patients' plasma compared to healthy individuals'. Enrichment and pathway analysis for the metabolites of the Venn C area by MetaboAnalyst showed that arginine biosynthesis (p < 0.01, FDR < 1%) was the most altered metabolic pathway (Fig. 3H). Alanine, aspartate and glutamate metabolism (p < 0.01, FDR < 1%) deserve attention in MI as well.
The potential capacity of each discriminant metabolite to diagnose MI was assessed by ROC analysis (Table 4). Notably, although pathway analysis did not draw our attention to the methionine-related metabolic module, methionine and homoserine showed their potential in distinguishing MI from non-MI cardiac cases. Homoserine (AUC = 0.94; speci city = 100%, sensitivity = 81%) was more speci c for MI diagnosis but less sensitive than methionine (AUC = 0.96; speci city = 94.6%; sensitivity = 89.4%) (Fig. 3J). dU also scored highly, with an AUC over 90% (Fig. 3F). Some other metabolites that showed good diagnostic potential were cysteine, 2-ketoglutarate, IDP, and uracil (Table 4).  Figure 3I). Additionally, the ORs of succinate, IDP and glyceric acid were much less than 1, indicating they are potential protective factors for MI (Table 4).

Traditional CAD risk factors and cardiac function in uence the metabolic pattern
Correlation analysis showed that methionine, homoserine, homocysteine and dU were all affected by smoking history, but none was obviously perturbed by hypertension (Table S3). However, in the subgroup analyses of smoking/nonsmoking, hypertensive/normotensive, diabetic/nondiabetic, aged 45-54/55-65 and male/female, means of homoserine, methionine and dU were higher in MI cases (Table 5). All data were logarized before calculating, the result showed means of each subgroup.
As a clinical indicator of cardiac function in MI, positive NT-proBNP represents cardiac dysfunction. Methionine, homoserine and deoxyuridine were further elevated in NT-proBNP-positive cases. Pathway analysis of the discriminant metabolites (Table S4) between the NT-proBNP positive and negative groups suggested that only arginine biosynthesis was severely impaired (P < 0.001, FDR < 1%), indicating that arginine biosynthesis is closely associated with cardiac function. (Figure S3A) cTnt, CK-MB, AST, LDH and HBDH are well-recognized indicators involved in myocardial damage and infarction. Six metabolites, 2-hydroxybutyrate, 3hydroxybutyrate, homocysteine, palmitic acid, stearic acid and 1-monooleoylglycerol, were positively and signi cantly correlated with both cTNT and CK-MB ( Figure S3C). As candidate predictors of MI, methionine, dU and homoserine were signi cantly and positively correlated with LDH, HBDH and AST ( Figure S3B).

Altered plasma pyrimidine and methionine metabolism in MI cases
The pathway of pyrimidine metabolism was deranged in the MI cases, as shown by the dramatic changes in dU and uracil. Figure 4D shows metabolites and metabolic enzymes in the dU-related pathway. According to a circulating transcriptomics dataset (GEO accession: GSE48060), the dU-related enzymes cytidine deaminase (CDA) and uridine phosphorylase 1 (UPP1) are upregulated in MI patient plasma. According to The Human Protein Atlas, CDA and UPP1 are highly enriched in immune cells (mainly in neutrophils and monocytes). Figure 4G shows white blood cell (WBC) counts, neutrophil (NE) counts, monocyte (MO) and lymphocyte (LY) counts and dU abundance in MI and non-MI cases. Immune cells generally increased in MI plasma. However, statistically, the correlation analysis did not detect a relationship between dU abundance and the number of any type of immune cell (including total WBCs).
Our study on transverse aortic constriction (TAC) mouse models showed that cardiac CDA mRNA expression increased as BNP and ANF mRNA levels increased ( Figure S3 D-F,H). dU level is signi cantly higher in the proBNP-positive group than in the proBNP-negative chest pain group (Fig. 4H). We hypothesize that dU is also a potential cardiac function marker.

Discussions
The onset of myocardial infarction is always urgent, sudden and sometimes associated with a dim prognosis. Early warning and recognition of myocardial infarction is a great challenge for clinicians and is of crucial importance to treat suspected MI patients. The early detection and intervention of MI require growing knowledge of the biology data of MI patients. In this study, we demonstrated that metabolomic data well discriminate MI cases from other non-MI chest pain cases. Some metabolites have the potential ability to differentiate and diagnose MI patients among CAD patients with the symptom of chest pain and suggest potential markers to predict the onset of MI.

Plasma metabolic features and metabolite biomarkers for the diagnosis of MI
This study identi ed a panel of discriminant metabolites that were also suggested as potential markers of MI in previous reports, such as taurine, methionine, leucine, isoleucine, valine, ornithine, tryptophan, citrate, and 2-ketoglutarate. [21] To our surprise, some of the metabolites showed great potential in the diagnosis of MI. Both elevated and decreased levels of metabolites were observed in the MI group. Among the differential metabolites between MI cases and non-MI cases, 10 of them in the MI group had twice the abundance as in the non-MI group (MIs/non-MIs, FC > 2); 17 metabolites had less than half the abundance as in the non-MI cases (MIs/non-MIs, FC < 0.5).

Elevated metabolites in MI cases
Based on metabolites that were elevated in the MI group, we found that methionine and cysteine metabolisms are involved in MI. In this metabolic module, methionine is a precursor of homocysteine, homoserine is utilized in the biosynthesis of methionine, and homoserine was also reported as a serum marker for cardiac disease in atherosclerosis patients with stent restenosis [22].
Although the causes of plasma metabolite alterations could be many (e.g., gut microbial metabolites), we evaluated plasma transcriptomics data because circulating substances in uence plasma metabolites most directly. Transcriptomics data showed that plasma methionine-related mRNA tends to utilize methionine and homocysteine to produce more cystathionine. Yet there was not a signi cant change of the key genes involved in the pathway of methionine transsulfuration. Plasma Cystathionine-β-synthase (CBS) and cystathionine-γ-lyase (CTH) a similar rna level between control and MIs, indicating blood cells methionine transsulfuration pathway in mRNA level cannot explain the elevation of methionine. One explanation is hinted at by murine MI models showing decreased levels of MARS mRNA in the heart, which suggests that the damaged heart has di culty binding methionine to tRNA. As a result, free methionine is released into the blood and detected in the circulatory system of MI patients.
The pyrimidine metabolism pathway was also deranged in the MI cases. The elevation of dU has been identi ed as a potential adverse factor for nucleotide pool balance and mitochondrial function in the case of mitochondrial neurogastrointestinal encephalomyopathy [23,24]. As a metabolite in pyridine metabolism, dU has never been suggested to play a role in cardiovascular diseases before. Unlike methionine, the elevation of dU is more likely to be related to blood cells. A previous study [25] (Dateset: GSE103182) showed that STEMI (ST-elevation myocardial infarction) patients express more neutrophil-speci c CDA mRNA in plasma than NSTEMI (non-ST-elevation myocardial infarction) patients. Consistent with this nding, our data showed that in the STEMI group, both dU and neutrophil counts were higher than those in the NSTEMI group ( Figure S3G). It is possible that granulocyte-speci c CDA is enriched in one subset of neutrophils. The subset of neutrophils varies signi cantly among non-MI, STEMI and NSTEMI cases. Alternatively, the elevation of CDA could be due to the heightened necrosis of granulocytes.

Decreased metabolites in MI cases
Among all the discriminant metabolites that declined in MI cases, arginine was the most noteworthy (Fig. 3C). ROC analysis showed arginine AUC equals to 0.8865 (MIs vs non-MIs). Our study showed that as cardiac damage and function worsened, these plasma arginine biosynthesis-related metabolites dropped more (Table 3). However, arginine's role as a CAD risk factor may be underestimated because arginine bioavailability, not just arginine abundance, was more signi cantly changed in MI patients. Another study reported that diminished global arginine bioavailability is predictive of increased CAD risk [16].
Uracil, a pyrimidine found in RNA, was also signi cantly decreased in MI plasma. Figure 4D [27]. In this study, diabetes and smoking also ranked the marked risk factors for MI occurrence, but the history of hypertension was not (OR < 1, p > 0.05). As for the key substances in lipid metabolism, higher levels of TC, TG and LDLc were con rmed as risk factors for MI groups (MIs vs non-MIs, Table S5). However, higher level of LDLc failed to demonstrate itself as a "risk" to develop non-MI cardiac chest pain.
(non-MIs vs the control, Table S5). Considering that routine therapy with anti-lipidemic and antihypertensive drugs lowered LDLc and blood pressure levels, medications could be responsible for some of the controversial ndings.

Metabolite risk factors
The panel of discriminant metabolites did not only offer metabolic markers facilitating the diagnosis of MI, but also contributed potential markers suggesting the conceivable occurrence of MI for prediction. Previous studies reported that methionine and dU can be promotors for CAD development because their accumulation may lead to more adverse effects on humans, e.g., high levels of methionine have been identi ed as atherogenic [28] and metabotoxic [29] for a long time, while dU acts as a conventional antimetabolite [30]. For the rst time, our data suggest that methionine and dU are highrisk markers for predicting patients about the occurrence of MI. Moreover, consistent with a previous report suggesting the relationship of dU with insulin resistance [31], our correlation analysis also suggested that dU is partially affected by diabetes (correlation analysis vs diabetes, p < 0.05, Pearson's r = 0.23), in addition to tobacco use and HDLc level.
As a precursor of homocysteine, methionine elevation is supposed to be a promoter for the development of CAD [32]. Consistent with that idea, a previous study showed that methionine in CAD cases is signi cantly higher than in healthy controls and is a risk factor for CAD occurrence in an unadjusted model [14]. However, after adjusting for race, sex, age, diabetes, smoking and hypertension, methionine is not as dangerous as it is in an unadjusted model [33]. Moreover, our data showed that methionine was an independent biomarker for MI. Notably, we found that the diagnostic performance of biomarker candidates for MI varied with individual characteristics, such as diabetis. When we studied individuals with a history of diabetes, methionine achieved an AUC score as high as 100% (sensitivity: 100.00%, speci city: 100.00%). For nondiabetic inpatients, the AUC score of methionine was only 73%, which is much lower than that of diabetic inpatients. Generally, this study suggests some metabolites that have the potential to facilitate the diagnosis and prediction of MI. We believe that the traditional marker (such as Tnt) and the metabolic markers (such as dU and methionine) re ect the damage of heart at different angles that are involved in confounding factors contributing to the onset of MI. In line with most studies, the odds ratio values consistently suggested that LDL cholesterol levels is the traditional common risk factor for cardiovascular disease [34]. Unexpectedly, in this study, a high level of LDLc (whether or not in an abnormal range) is not a risk for cardiac chest pain patients. Similarly, hypertension is a risk for developing CAD-related chest pain by healthy people, but not for the development of MI by chest pain patients. It greatly challenges our further research of the underlying mechanism.

Conclusion
Plasma deoxyuridine, homoserine, methionine, as well as a combination of homoserine, IDP and 2-ketoglutarate were suggested as markers for MI diagnosis. The changes of some metabolites abundance in diseased samples can be explained by plasma transcriptomics and could be highly associated with the blood cells. Hopefully, the growing plasma single-cell RNA sequencing evidence [35,36] and omics data (e.g., cfDNA methylome [37]) offer a clue to reveal the mechanism behind the changes soon.