All patients were recruited from the Second Affiliated Hospital of the Harbin Medical University. This study was approved by the Ethics Committee of the Harbin Medical University (HMUDQ2019082601). Cerebral infarction was diagnosed in 2016 according to the Chinese guidelines for cerebrovascular disease prevention and treatment. The inclusion and exclusion criteria are as follows: 1) sudden onset; 2) focal nerve defects (weakness or numbness on one face or limb, language disorders, etc.), a few are comprehensive neurological defects;3) the duration of symptoms or signs has no limit when imaging shows clear ischemic lesions or has continued for more than 24 h when imaging indicated lesions are lacking; (4) excluded the non-vascular etiology; and 5) Brain CT/MRI excluded cerebral hemorrhage. According to the time from onset, patients within 3 days to 7 days away were randomly divided into five groups. Specifically, time within 3 days of cerebral infarction was defined as group A (31 CIs), time after 3 days but within 5 days of cerebral infarction was defined as group B (17 CIs), time after 5 days but within 7 days of cerebral infarction was defined as group C (13 CIs), time after 7 days of cerebral infarction was defined as group D (19 CIs), and cerebral infarction patients with glycosuria were defined as group T (20 CIs). According to the Second Affiliated Hospital’s routine clinical laboratory procedures of the Harbin Medical University, serum was kept in cryotubes and stored at −80°C in freezers after biochemical tests. The first batch of serum (test group) for non-targeted metabolomic analysis was collected from 100 CI patients and 41 healthy controls from November 2016 to January 2017. The second batch of serum (validation group) was collected from 48 CI patients and 19 healthy controls for validation analysis from January to February 2017. Two batches of serum were needed because of the limited blood collected from each person and the expansion of the population size. There was no significant difference in age and gender distribution across groups. The sample information is summarized in Table 1 and Supplementary Table S1.

We extracted metabolites from 200 µl of serum with four volumes of acetonitrile containing 12 internal standards (carnitine C 2:0-d3, carnitine C 16:0-d3, lysophosphatidylcholine (LPC) 12:0, LPC 19:0, tryptophan-d5, FFA 16:0-d3, FFA 18:0-d3, chenodeoxycholic acid-d4, tryptophan-d5, triglyceride (TG) 45:0, sphingomyelin (SM) 12:0/30:1, and phosphatidylethanolamine (PE) 30:0) (Zhao et al., 2014). Internal standards were added to each sample in order to check data from previous reports. After 15 min of centrifugation at 13,000 g, the supernatant was distributed into two aliquots and dried using a vacuum centrifuge at 4°C before the positive ion and negative ion mode tests. Aliquots were reconstituted in 100 μl of acetonitrile/water (2:8) and analyzed using a Waters ACQUITY UPLC (Waters Corp, Milford, United States), coupled with an AB SCIEX TripleTOF 5,600 System (ABSCIEX, Framingham, United States) (Zhao et al., 2014). The injection volume was 6 μl for each run. Quality control (QC) samples were carried out after every six serum samples, which were obtained by mixing 10 μl from each sample.

We used a 2.1 × 100 mm ACQUITYTM 1.7 μm C8 BEH column (Waters, Ireland) as a stationary phase in positive ion mode. Both water (phase A) and acetonitrile (phase B), including 0.1% formic acid, were the mobile phases. The UPLC was subjected to gradient elution: 95% A for 1 min, changed linearly to 100% B within 24 min, and then held for 4 min. A 2.1 × 100 mm ACQUITYTM 1.8 μm T3 HSS column (Waters, Ireland) was used for LC separation in negative ion mode (Zhao et al., 2014). Both water (phase C) and 95% methanol/water (phase D), containing 6.5 mM NH₄HCO₃, were the mobile phases. The first gradient elution was 2% D for 1 min, which then changed linearly to 100% D within 18 min, and finally maintained for 4 min. The flow was 0.35 ml/min, and the column temperature was 50°C (Zhao et al., 2014).

Data were obtained in full scan mode from m/z 80 to 1,000 with a cycle time of 275 ms. The MS parameters were set as follows: ion spray voltage, 5500 V (positive ion) and 4500 V (negative ion); curtain gas, 35 PSI; ion source gas 1, 50 PSI; ion source gas 2, 50 PSI; and interface heater temperature, 500°C (Zhao et al., 2014; Hu et al., 2017; Chew et al., 2018; Gu et al., 2018).

A Markerview workstation (AB SCIEX, United States) was used to extract and align the non-targeted metabonomic data. Internal standards were selected based on the lowest relative standard deviation value in the QC sample. Before further data analysis, we calibrated the intensity of each peak by its suitable internal standard and integrated the positive and negative ion non-targeted data into one table. We utilized the SIMCA-P software (version 14.1; Umetrics, Umea, Sweden) for multivariate statistical analysis. After unit variance scaling, principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA) were applied to distinguish between healthy controls and patients with CI at different stages (Zhao et al., 2014; Li et al., 2015; Fan et al., 2018).

To prevent model overfitting, we conducted a permutation test. For non-targeted metabonomics, we used metabolites with variable importance in the projection values (VIP) larger than one for analysis (Kuboniwa et al., 2016; Li et al., 2020). The Wilcoxon Mann–Whitney test was used to identify significantly different metabolites. Statistical significance was defined as p< 0.05, and p< 0.10 as a criterion for the false discovery rate found in multiple comparisons. We demonstrated the relationship between different metabolites using the MeV4.5.1 version of software (Saeed et al., 2003). In Figure 3 to Figure 9, all data are the mean ±SEM. *p<0.05 vs. N group; **p<0.01 vs. N group; bt, p<0.05 B group vs. T group; dt, p<0.05 D group vs. T group; two-tailed Mann–Whitney U test. Raw data were presented in Supplementary Table S2.

To explore key metabolites from serum metabolism at different periods of cerebral infarction, serum metabolic profiles from the different groups (N, A, B, C, D, and T) were analyzed by UPLC-TOF-MS. QC data were also obtained to ensure dependability. High reproducibility of QC data indicated high quality results (Figures 1A,C). The relative standard deviation value of 73% (2301 ions) of the total ions (3140 ions) in the QC sample was less than 30% (Figure 1B). We performed OPLS-DA to further explore key metabolites from each group (Figures 1D–H). A clear differentiation was observed between groups A and N (Figure 1D), B and N (Figure 1E), C and N (Figure 1F), D and N (Figure 1G), and T and N (Figure 1H), respectively. All OPLS-DA models were reliable because none of the permutation tests had overfitting. The results obtained here were then used to explore critical metabolites and multiple metabolic pathways at different periods. VIP and p value data were presented in Supplementary Table S3.

We investigated metabolic pathways between the N group and various stages of the cerebral infarction group (Figures 2A–E). After comparison with healthy people, thirty-five metabolic pathways were identified and explored. Furthermore, we focused on the top ten important pathways that differed between the A, B, C, D, or T groups and N (Figure 2F). For instance, the top ten pathways that differed between the A and N groups were steroid hormone biosynthesis, aminoacyl-tRNA biosynthesis, taurine and hypotaurine metabolism, pyruvate metabolism, glycolysis or gluconeogenesis, primary bile acid biosynthesis, nitrogen metabolism, phenylalanine metabolism, arginine and proline metabolism, and propanoate metabolism. Importantly, our results revealed that phenylalanine metabolism was always involved in the development of cerebral infarction within 3 to 7 days. Interestingly, three pathways that differed only between the T and N groups were found. They were galactose metabolism, inositol phosphate metabolism, and ascorbate and aldarate metabolism. There were also only three pathways differentially active between the B and N groups. They are valine, leucine, and isoleucine biosynthesis (Gu et al., 2018); valine, leucine, and isoleucine degradation (Tong et al., 2020); and tryptophan metabolism, ubiquinone, and other terpenoid–quinone biosynthesis (Tong et al., 2020). Tyrosine metabolism and phenylalanine, tyrosine, and tryptophan biosynthesis differed significantly between the C and N groups. Finally, the four pathways differed between the D and N groups: glycerolipid metabolism, methane metabolism, linoleic acid metabolism, and d-glutamine and d-glutamate metabolism. These findings reveal that different pathways play different roles in the progression of CI over time, helping us to further explore the etiopathogenesis of different stages of cerebral infarction.

Previous reports found carbohydrate and amino acid metabolites in glycuresis, angiocardiopathy, and therioma (Tabas, 2010). Therefore, we further analyzed serum carbohydrate and amino acid metabolites at various periods of CI. As shown in Figure 3, results revealed that five types of metabolites were changed significantly compared to those in the N group. Specifically, four metabolites, L-phenylalanine, phenylacetyl-L-glutamine, indoline, and L-isoleucine (except for the D group), were dramatically elevated at different periods of cerebral infarction compared to those in the N group, while Phe-Phe was dramatically decreased in the A and B group. Finally, 20 types of metabolites were identified that did not dramatically change between the different periods of cerebral infarction and the N group.

Some components of bile acids have been reported to have protective effects against stroke and cerebral ischemia injury (Tabas, 2010; Wang et al., 2018). We detected nine types of bile acids in the sera of patients with CI (Figure 4). Specifically, glycochenodeoxycholate (GCDCA) showed a marked increase in B, D, and T groups compared with the N group, while glycocholate showed an obvious increase only between D and N groups. With the exception of GCDCA and glycocholate, all the other seven substances showed no change between the CI groups and N.

Carnitine plays an important role in mitochondrial metabolism and in modulating the ratio of coenzyme A (CoA): acyl-CoA (Zhang et al., 2012a). Studies have shown that l-carnitine (LC) and acetyl-l-carnitine (ALC) were highly effective against ischemic injury in neuronal cells in vitro, but ALC also played a protective role in neurons after ischemic injury in vivo. Using UPLC-MS analysis, we detected free carnitine, acetyl carnitine, and nine carnitines with chain lengths from 3 to 18 carbon units (Figure 5). The most notable differences were observed for medium-chain carnitines, most of which were lower in CI serum than in the N group. Specifically, 2-octenoyl carnitine in B, C, D, and T groups was lower than in N; decanoyl carnitine in A, B, C, and D groups was lower than in N; and carnitine C10:3 in B and C groups was lower than in N. In contrast, long-chain carnitines were higher in CI groups than those in N. In particular, carnitine C18:1 in A and B groups is higher than in N; carnitine C18:2 in A, B, C, and D groups is higher than in N. However, short-chain acylcarnitines showed no differences. These results are highly interesting but need explanation.

Our study further revealed that stearate metabolites changed in patients over different periods of cerebral infarction. As shown in Figure 6, trans-vaccenic acid, linolenic acid, linoleic acid, all-cis-4,7,10,13,16-docosapentaenoic acid, arachidonic acid, and docosahexaenoic acid (DHA) decreased after 7 days of cerebral infarction (D group). In addition, palmitoleic acid, cis-8,11,14-eicosatrienoic acid, and palmitic acid were increased within 3 days of cerebral infarction (group A) but gradually returned to normal levels. In particular, 1-linoleoyl-rac-glycerol was only elevated after 3 days and within 5 days of cerebral infarction (B group). Interestingly, three metabolites were significantly increased in the T group compared with those in the N group: 1-linoleoyl-rac-glycerol, cis-8,11,14-eicosatrienoic acid, and all-cis-4,7,10,13,16-docosapentaenoic acid. The aforementioned results showed that stearate metabolism was closely related to cerebral infarction during different periods.

Our results demonstrated that concentrations of three metabolites, monoacylglycerol (MAG) 18:1, MAG 20:3, and phosphoric acid, increased remarkably in groups A and B, but gradually returned to normal in group C (Figure 7). However, choline decreased noticeably from group A to group C, but gradually returned to normal in group D. In addition, 1-hexadecanol in A, B, and D was significantly lower, but higher in the T group, than in the normal group.

Dehydroepiandrosterone sulfate (DHEA), a cytoplasmic steroid, was significantly decreased in all CI groups compared to the N group. Moreover, 5α-dihydrotestosterone sulfate was significantly lower in the A and B groups than in N. In contrast, di-n-butylphthalate was elevated in the A, D, and T groups compared with the N group. These changes suggest that sterol metabolism is involved in the occurrence and development of CI.

Lipids, or fats, are incredibly important to human health, and are one of the most difficult biomolecules to study. Body fat deposition and excessive FFA metabolism can lead to adverse health effects, including hyperlipidemia and obesity (Ebbert and Jensen, 2013). Four LPCs, eight PCs, and nine FFAs were measured in our study (Figure 8). LPC 18:1, LPC 18:2, FFA 18:2, FFA 18:3, FFA 20:4, FFA 22:6, PC 36:5, and PC 38:4 were lower in one or more CI groups than those in the N group. These results demonstrate three things—first, LPC 18:0 and LPC 18:1 had significant difference in the A, B, C and T groups compared with the normal group; LPC 18:0 was increased and LPC 18:1was decreased. Second, FFA 18:1, FFA 18:2, FFA 18:3, FFA 20:4, FFA 22:6, PC 34:1, PC 36:3, and PC 38:4 were markedly different between the D and N groups. Third, PC 36:5 and LPC 18:2 were all decreased in the B and C groups compared to the N group. In contrast, PC 32:1, PC 34:2, PC 34:3, FFA 16:1, FFA 18:0, and FFA 22:4 were higher in the A and/or D group than in the N group. Furthermore, FFA 18:0 and FFA 22:4 in the A, B, C, and T groups were significantly higher than those in the normal group. This is an important finding to understand lipid metabolism at various periods of the pathological process in cerebral infarction.

To better understand changes in substances in different periods of cerebral infarction, we analyzed the biochemical results of all the samples. Biochemical tests were classified into six categories: ions and elements, blood glucose, myocardial enzymes, liver function, blood fatty acids, and renal function. The detailed results are shown in Figure 9. Ca, K, Cl, TG, high density lipoprotein (HDL), albumin (ALB), globulin (GLO), ALB/GLO, indirect bilirubin (IDBIL), direct bilirubin (DBIL), alanine aminotransferase (ALT), glutamete transpeptidase (GGT), alkaline phosphatase (ALP), polymerase acidic protein (PA), and creatinine (Crea) were found to be significant in distinguishing cerebral infarction in different periods.

The test of the second batch of blood samples confirmed that the significant changes in 13 metabolites were consistent with the first batch. They were 5α-dihydrotestosterone sulfate, choline, di-n-butylphthalate, FFA 16:0, FFA 18:0, FFA 18:1, FFA 18:2, FFA 20:4, L-phenylalanine, PC 16:0–18:2 (PC 34:2), PC 34:3, phenylacetyl-L-glutamine, and phosphoric acid. The specific information is shown in Tables 1, 2 and Supplementary Table S1.

We identified the top ten pathways that differed significantly between the A, B, C, D, and T groups and N. (Figure 2). Phenylalanine (Phe) metabolism was involved in the development of cerebral infarction within 3 to 7 days. Phe, along with other physiologically active aromatic amino acids, is one of the essential amino acids that cannot be synthesized by humans and animals. Recently, it was shown that phenylalanine derivatives had protective effects on blood vessels due to their fibrinolytic effects (Koizumi et al., 2016). It is well known that blood clots in vessels can cause severe ischemic diseases such as cerebral infarction and myocardial infarction (Irie et al., 2014).The lower concentrations of Phe-Phe were found in groups A and B (Figure 3). Large amounts of Phe-Phe were consumed or their source reduction in the early periods of cerebral infarction merited further research.

There were three pathways that were differentially active only between the diabetic (T) and N groups: galactose metabolism, inositol phosphate metabolism, and ascorbate and aldarate metabolism. Our data are in good agreement with the data previously reported in literature. One study reported that increased atherosclerotic volume and plaques were correlated with IgE sensitization to α-galactose (Wu et al., 2017). Galactose, which increases satiety and mobilizes fat, can protect against type 2 diabetes (Charrière et al., 2017). It has been reported that inositol phosphate accumulation can cause accelerated platelet functions in diabetes mellitus (Ishii et al., 1991), and ascorbic acid (ascorbate) could reverse high glucose-induced increases in endothelial barrier permeability (Meredith et al., 2014).