Abstract Background Metabolomics studies have identified various metabolic markers associated with stroke risk, yet much uncertainty persists regarding heterogeneity in these associations between different stroke subtypes. We aimed to examine metabolic profiles associated with incident stroke and its subtypes in Chinese adults. Methods and Results We performed a nested case–control study within the Dongfeng‐Tongji cohort, including 1029 and 266 incident cases of ischemic stroke (IS) and hemorrhagic stroke (HS), respectively, with a mean follow‐up period of 6.1±2.3 years. Fifty‐five metabolites in fasting plasma were measured by ultra‐high‐performance liquid chromatography–mass spectrometry. We examined the associations of metabolites with the risks of total stroke, IS, and HS, with a focus on the comparison of associations of plasma metabolite with IS and HS, using conditional logistic regression. We found that increased levels of asymmetrical/symmetrical dimethylarginine and glutamate were significantly associated with elevated risk of total stroke (odds ratios and 95%, 1.20 [1.08–1.34] and 1.22 [1.09–1.36], respectively; both Benjamini‐Hochberg‐adjusted P <0.05). When examining stroke subtypes, asymmetrical/symmetrical dimethylarginine was nominally associated with both IS and HS (odds ratios [95% CIs]: 1.16 [1.03–1.31] and 1.39 [1.07–1.81], respectively), while glutamate was associated with only IS (odds ratios [95% CI]: 1.26 [1.11–1.43]). The associations of glutamate with IS risk were significantly stronger among participants with hypertension and diabetes than among those without these diseases (both P for interaction <0.05). Conclusions This study validated the positive associations of asymmetrical/symmetrical dimethylarginine and glutamate with stroke risk, mainly that of IS, in a Chinese population, and revealed a novel unanimous association of with both IS and HS. Our findings provided potential intervention targets for stroke prevention. Keywords: metabolite, stroke subtypes, total stroke Subject Categories: Cerebrovascular Disease/Stroke __________________________________________________________________ Nonstandard Abbreviations and Acronyms ADMA/SDMA asymmetrical/symmetrical dimethylarginine BH Benjamini‐Hochberg HS hemorrhagic stroke IS ischemic stroke Clinical Perspective. What Is New? * In Chinese adults, we found that increased levels of asymmetrical/symmetrical dimethylarginine and glutamate were associated with an elevated risk of total stroke. * When analyzing the ischemic and hemorrhagic stroke subtypes, we found, for the first time, that asymmetrical/symmetrical dimethylarginine was nominally associated with risks of both ischemic and hemorrhagic stroke, while glutamate was associated with only ischemic stroke. * The association of glutamate with ischemic stroke was significantly stronger among individuals with hypertension and diabetes than among those without these diseases. What Are the Clinical Implications? * Our findings provide new insight into the cause of stroke and its subtypes. * Asymmetrical/symmetrical dimethylarginine and glutamate might be intervention targets for stroke prevention. Stroke remains the second‐leading cause of death around the world.[50] ^1 In 2019, there were 12.22 million incident cases of stroke globally, with 2.87 million of them in China. This enormous burden poses an urgent need to identify promising risk biomarkers and discover potential intervention targets.[51] ^2 Metabolomics aims to capture alterations of small molecules in biological samples to provide a systematic view of metabolism perturbations and has been proven to be an efficient approach to search for metabolic markers, which may lead to uncovering novel pathophysiological mechanisms.[52] ^3 Prospective metabolomic studies have identified various novel metabolites associated with elevated risk of stroke, including amino acids, carboxylic acid, nucleosides, and various lipids such as acylcarnitines, ceramides, and lipoproteins.[53] ^4 , [54]^5 , [55]^6 , [56]^7 , [57]^8 , [58]^9 , [59]^10 , [60]^11 , [61]^12 , [62]^13 However, most of these studies either focused on total stroke without differentiating the ischemic and hemorrhagic subtypes,[63] ^4 , [64]^5 , [65]^6 , [66]^7 which had many similarities but also notable distinctions in their risk factor profiles,[67] ^14 or examined only ischemic stroke (IS).[68] ^8 , [69]^9 , [70]^10 Only 2 prospective studies of European ancestry and 1 of the Chinese population made comparisons of the metabolic profiles associated with the 2 subtypes of stroke,[71] ^11 , [72]^12 , [73]^13 which found multiple metabolites associated with IS, but not with hemorrhagic stroke (HS). However, they measured mainly lipids and lipoproteins, with only a few other metabolites. There is still a lack of knowledge on the prospective association between circulating metabolites and stroke risk in the Chinese population, particularly heterogeneity between stroke subtypes. In this study, we conducted a nested case–control study within the Dongfeng‐Tongji cohort, which consisted of middle‐aged and elderly Chinese people. A total of 1295 incident stroke cases, including 1029 IS and 266 HS cases, and 1:1 matched controls were involved in this study. We measured baseline levels of 55 plasma metabolites, mainly amino acids and (acyl)carnitines, which are vital in amino acid and fatty acid metabolism and have been suggested to be candidate biomarkers for stroke risk.[74] ^15 We aimed to examine the associations of metabolites with risks of incident stroke and its subtypes. Methods The data and analytical methods that support the findings of this study are available from the corresponding author upon reasonable request. Study Population and Design The Dongfeng‐Tongji cohort is an ongoing prospective cohort study of retirees of the Dongfeng Motor Company located at Shiyan City in the northwest of Hubei Province, as previously described.[75] ^16 A total of 27 009 and 14 120 participants were enrolled, respectively, during the baseline survey in 2008 to 2009 and the first follow‐up survey in 2013. All participants underwent standard questionnaires and physical examinations and provided fasting blood samples during the surveys, and were all tracked for vital status and disease occurrence through the Dongfeng Motor Company health care service system thenceforth. The design of this nested case–control study has been described in detail elsewhere.[76] ^17 In brief, stroke was defined as the sudden or rapid onset of a typical neurological deficit of vascular origin that persisted for >24 hours or until death.[77] ^18 Stroke records from medical insurance documents, hospital records, death certificates, and questionnaires were referred to an adjudication board of clinical physicians for definite diagnosis and classification of stroke subtypes according to the clinical symptoms and imaging manifestations (computed tomography or magnetic resonance imaging). Incident stroke was defined as the first stroke occurrence during the period from the study baseline (ie, the date participants enrolled in the Dongfeng‐Tongji cohort) to the censor time, December 31, 2016. After excluding the participants without enough plasma for metabolite measurement, or with stroke, coronary heart disease, and cancer at baseline, or lost to follow‐up, there were 1295 incident stroke cases remaining for this study, including 1029 IS and 266 HS cases, with an average follow‐up period of 6.1±2.3 years. All cases were stroke‐free at baseline and developed stroke during follow‐up. For each incident stroke case, we matched 1 control according to age (±1 year), sex, and sampling date (±30 days). All controls were stroke‐free at least until the time of stroke onset of their matched case. All participants provided written informed consent. The study was approved by the Ethics and Human Subject Committees of the Tongji Medical College. Definitions of Covariates In this study, we considered various traditional risk factors of stroke for adjustment, including demographic characteristics, lifestyle factors, and other clinical variables. Among the considered covariates, information on age, sex, smoking status, drinking status, regular exercise, history of disease, and medication use were collected by a semistructured questionnaire. Specifically, smoking status was categorized as current (smoking at least 1 cigarette per day and lasting for more than half a year), former, and never smoking groups. Drinking status was categorized as current (drinking at least once a week and lasting for more than half a year), former, and never drinking groups. Regular exercise was defined as exercising (walking, jogging, dancing, cycling, swimming, climbing, doing tai chi, or playing ball games) at least 5 times a week, at least 30 minutes each time, and lasting for at least 6 months. The participant whose first‐degree relatives had been diagnosed with stroke was regarded as having a family history of stroke. For clinical variables, blood pressure, body weight, standing height, and ECG were measured by trained staff during the physical examination. Serum lipids and fasting glucose were analyzed by the biochemical laboratory of Dongfeng General Hospital. Body mass index (BMI) was calculated by dividing weight in kilograms by the square of standing height in meters. Hypertension was defined as blood pressure ≥140/90 mm Hg,[78] ^19 self‐reported physician‐diagnosed hypertension, or current use of antihypertensive medication. Diabetes was defined as fasting glucose ≥7.0 mmol/L,[79] ^20 self‐reported physician‐diagnosed diabetes, or current use of hypoglycemic medication or insulin. Atrial fibrillation and left ventricular hypertrophy were determined by clinical doctors according to the presence of the ECG. Metabolite Measurement The method for metabolite measurement was described in detail previously,[80] ^21 with some modifications in the present study. In brief, fasting blood samples collected by EDTA tubes were centrifuged to obtain plasma and then stored at −80 °C until pretreatment. For each participant, we extracted metabolites from 10 μL plasma using 90 μL 75:25:0.2 (vol/vol/vol) acetonitrile/methanol/formic acid, which contained stable isotope‐labeled internal standards (0.25 μmol/L valine‐d8, 0.18 μmol/L phenylalanine‐d8, and 0.18 μmol/L carnitine‐d9 (all from Cambridge Isotope Laboratories, Andover, MA). After vortexing, we centrifuged the mixture at 12 000g for 10 minutes at 4 °C and injected the supernatant onto a Cortex hydrophilic interaction liquid chromatography column (100×2.1 mm; Waters Corp., Milford, MA). Metabolites were measured using Agilent 1290 infinity series ultra‐high‐performance liquid chromatography (Agilent Technologies Inc., Palo Alto, CA) equipped with Applied Biosystems/Sciex 6500 QTRAP mass spectrometry (Applied Biosystems, Foster City, CA). Mobile phase A was an aqueous solution containing 10 mmol/L ammonium formate and 0.1% formic acid. Mobile phase B was an acetonitrile solution containing 0.1% formic acid. The gradient eluted condition was 0 to 0.5 minutes, 5% A; 0.5 to 7.5 minutes, 5% to 60% A; 7.5 to 8.5 minutes, 60% A; 8.5 to 9.5 minutes, and 60–5% A. The post time was 6 minutes for column equilibration. The mass conditions were as follows: multiple reaction monitor scan was performed in positive ion mode. The spray voltage of the electrospray ionization source was 4.5 kV and the source temperature was 350 °C. The 55 measured metabolites in this study included 26 amino acids, 12 (acyl)carnitines, 3 B vitamins, 2 cholines, 2 amines, 2 indole derivatives, 2 purine derivatives, and 6 other metabolites (cotinine, creatinine, hippurate, kynurenic acid, uridine, and α‐glycerophosphocholine). Signal peaks of metabolites, transitions, declustering potentials, and collision energies were identified and optimized using corresponding standard substances (all from Sigma‐Aldrich Corp., St. Louis, MO). All samples were measured in randomized order, and all laboratory staff were blinded to the case and control status. We inserted pooled quality control samples in the analytical queue every 20 samples. Internal standard peak areas were used to monitor the variance introduced by the pretreatment process and instrument status serving as the quality control of measurement. After automated peak integration, we manually viewed metabolite peaks to ensure the quality of integrations and to confirm identity by comparison against standard substances. Using the statTarget tool, we conducted quality control–based random forest signal correction to reduce the unwanted variance of all raw peak areas caused by the batch effect in the measurement process.[81] ^22 Finally, the peak area of each metabolite in each sample was normalized to the nearest quality control sample. The experiment was performed in the central laboratory at the School of Public Health, Tongji Medical College, Huazhong University of Science and Technology. Statistical Analysis Baseline characteristics between cases and controls were compared by 1‐way ANOVA and χ^2 test for continuous and categorical variables, respectively. The metabolite data used for analysis were naturally logarithmically transformed and scaled to z scores. We assessed the pairwise correlations of metabolites in controls by Pearson product–moment correlation with adjustment for age and sex. Conditional logistic regression was used to analyze the associations between metabolites and risks of total stroke and its subtypes (IS and HS), and the risk estimates were presented as odds ratios and 95% CIs for per SD increment of metabolite levels. Adjusted covariates included age, BMI, smoking status, drinking status, regular exercise, hypertension, atrial fibrillation, left ventricular hypertrophy, fasting glucose, serum total cholesterol, high‐density lipoprotein cholesterol, and triglycerides, family history of stroke, and storage time of samples. We used the Benjamini‐Hochberg (BH) method to control false discovery rates in multiple tests. A 2‐sided P <0.05 was regarded as nominally significant, and a BH‐adjusted P <0.05 was considered statistically significant. To explore biological patterns underlying the stroke‐associated metabolites, we further conducted metabolic pathway enrichment analysis based on the Kyoto Encyclopedia of Genes and Genomes database with Metaboanalyst 6.0. Metabolites that at least reached nominal significance in the aforementioned association tests were used in this analysis. A metabolite pathway was regarded as nominally associated with stroke risk if at least 1 stroke‐associated metabolite can be found in this metabolite pathway at a 95% probability. A pathway with BH‐adjusted P <0.05 was considered statistically significant. To assess the improvement in risk prediction by the stroke‐associated metabolites for IS and HS, we constructed 2 risk‐prediction models for IS and HS, respectively: a reference model including traditional risk factors (age, sex, BMI, smoking status, drinking status, regular exercise, hypertension, atrial fibrillation, left ventricular hypertrophy, fasting glucose, serum total cholesterol, high‐density lipoprotein cholesterol, and triglycerides, family history of stroke, and storage time of samples) and another model additionally including the metabolites associated with stroke risk. We then calculated and compared the c‐statistics of the above models by a bootstrap validation method with 1000 resampling iterations. R package pROC was used. We examined the associations of metabolites with IS and HS risk among participants with different risk profiles through subgroup analyses with stratification by traditional stroke risk factors including age (<65 or ≥65 years), sex (male or female), BMI (<24 or ≥24 kg/m^2), smoking status (current or noncurrent smoking), drinking status (current or noncurrent drinking), regular exercise (yes or no), hypertension (yes or no), and diabetes (yes or no). Statistical analyses were performed using SAS 9.4 (SAS Institute) and R version 3.6.3 (R Core Team). Results The baseline characteristics of the participants are presented in Table [82]1. The mean age of stroke cases was 66.61 ± 7.65 years old at enrollment, with men accounting for 62.2% of incident cases. Compared with controls, stroke cases had higher mean BMI and fasting glucose, and were more likely to have hypertension, atrial fibrillation, and left ventricular hypertrophy, but less likely to have family history of stroke. For stroke subtypes, the mean ages were 66.86 ±7.61 years for IS cases and 65.63 ±7.73 years for HS cases, with men accounting for 63.0% and 59.0% of them, respectively. Compared with matched controls, IS cases had higher mean BMI, fasting glucose, and total triglycerides, were more likely to have hypertension and atrial fibrillation, but were less likely to have family history of stroke. HS cases were more likely to have hypertension and had a higher mean fasting glucose than matched controls. Table 1. Baseline Characteristics of Stroke Cases and Controls Variable Total stroke IS HS Cases Controls P Cases Controls P value Cases Controls P value (n=1295) (n=1295) (n=1029) (n=1029) (n=266) (n=266) Age (y) 66.61 ±7.65 66.53 ±7.56 0.81 66.86 ±7.61 66.77 ±7.51 0.79 65.63 ±7.73 65.63 ±7.70 0.99 Male (%) 62.2 62.2 – 63.0 63.0 – 59.0 59.0 – BMI (kg/m^2) 24.76 ±3.56 24.37 ±3.11 0.004 24.79 ±3.39 24.41 ±3.09 0.008 24.63 ±4.15 24.24 ±3.23 0.24 Current smoker (%) 27.6 24.2 0.14 28.6 24.4 0.09 23.5 23.8 0.59 Current drinker (%) 29.4 26.0 0.16 29.9 26.8 0.23 27.2 22.9 0.40 Regular exercise (%) 71.4 73.7 0.17 70.9 73.5 0.20 72.9 74.8 0.62 Hypertension (%) 71.0 55.6 <0.001 70.0 54.9 <0.001 74.8 58.3 <0.001 Atrial fibrillation (%) 1.6 0.5 0.004 1.8 0.5 0.006 1.1 0.4 0.32 Left ventricular hypertrophy (%) 1.2 0.5 0.049 1.0 0.3 0.05 1.9 1.1 0.48 Fasting glucose (mmol/L) 6.49 ±2.38 6.02 ±1.72 <0.001 6.55 ±2.40 6.05 ±1.76 <0.001 6.30 ±2.30 5.91 ±1.53 0.03 Total cholesterol (mmol/L) 5.15 ±0.99 5.11 ±1.03 0.32 5.19 ±0.98 5.10 ±1.02 0.07 5.00 ±1.04 5.13 ±1.05 0.19 High‐density lipoprotein cholesterol (mmol/L) 1.45 ±0.47 1.43 ±0.44 0.43 1.44 ±0.47 1.43 ±0.45 0.62 1.47 ±0.48 1.44 ±0.41 0.47 Total triglycerides (mmol/L) 1.52 ±1.12 1.46 ±1.25 0.21 1.54 ±1.06 1.41 ±0.88 0.004 1.46 ±1.35 1.67 ±2.17 0.20 Family history of stroke (%) 2.8 4.6 0.02 2.8 4.5 0.046 2.6 4.9 0.17 [83]Open in a new tab BMI indicates body mass index; HS, hemorrhagic stroke; and IS, ischemic stroke. The mean correlations within amino acids and (acyl)carnitines were 0.25 and 0.39, respectively (Figure [84]S1). Among the 55 metabolites, asymmetrical/symmetrical dimethylarginine (ADMA/SDMA) and glutamate were significantly associated with an elevated risk of total stroke (Table [85]2), with ORs (95% CIs) for per SD increment being 1.20 (95% CI, 1.08–1.34; P=8.53E−04; BH‐adjusted P=0.02) and 1.22 (95% CI, 1.09–1.36; P=7.11E−04; BH‐adjusted P=0.02), respectively (Table [86]2). The inter‐ and intra‐assay coefficients of variation of ADMA/SDMA and glutamate were <10% (Table [87]S1). There were an additional 9 metabolites that achieved nominally significant association with total stroke risk, among which phenylalanine, cotinine, kynurenic acid, and α‐glycerophosphocholine were associated with elevated risk, while asparagine, betaine, glutamine, methionine, and acetylcholine were associated with decreased risk (all P <0.05) (Table [88]S2). Table 2. Associations of ADMA/SDMA and Glutamate With Risks of Total Stroke, IS, and HS Total stroke IS HS OR (95% CI) P value BH‐adjusted P value OR (95% CI) P value BH‐adjusted P value OR (95% CI) P value BH‐adjusted P value ADMA/SDMA 1.20 (1.08–1.34) 8.53E−04 0.02 1.16 (1.03–1.31) 0.01 0.18 1.39 (1.07–1.81) 0.01 0.15 Glutamate 1.22 (1.09–1.36) 7.11E−04 0.02 1.26 (1.11–1.43) 4.73E−04 0.03 1.01 (0.78–1.31) 0.96 0.98 [89]Open in a new tab ADMA/SDMA indicates asymmetrical/symmetrical dimethyl arginine; BH, Benjamini‐ Hochberg; HS, hemorrhagic stroke; IS, ischemic stroke; and OR, odds ratio. When examining stroke subtypes, we found that ADMA/SDMA was nominally associated with both IS and HS risks, with ORs (95% CIs) of 1.16 (1.03–1.31; P=0.01) and 1.39 (1.07–1.81; P=0.01), respectively. However, glutamate was associated with only IS risk, with OR (95% CI) of 1.26 (1.11–1.43; P=4.73E−04; BH‐adjusted P=0.03) (Table [90]2). There were an additional 8 metabolites that achieved nominally significant associations with IS risk, among which phenylalanine, hypoxanthine, cotinine, and kynurenic acid were associated with increased risk, and asparagine, glutamine, methionine, and octanoylcarnitine were associated with decreased risk (all P <0.05) (Table [91]S2). For HS risk, there were 9 metabolites that achieved nominally significant associations, among which citrulline and dimethylglycine were associated with increased risk, and hydroxyproline, (iso)leucine, lysine, (iso)butyrylcarnitine, niacinamide, hypoxanthine, and inosine were associated with decreased risk (all P <0.05) (Table [92]S2). The metabolic pathway enrichment analyses in this study were exploratory using the metabolites associated with IS and HS at least at nominal significance (Table [93]S3), since the number of statistically significant metabolites (2) was too small for enrichment. For IS, we observed 3 pathways reaching significance levels, including alanine, aspartate, and glutamate metabolism (enrichment ratio: 27.52, BH‐adjusted P=7.55E‐03), nitrogen metabolism (enrichment ratio: 85.47, BH‐adjusted P=7.55E‐03), and arginine biosynthesis (enrichment ratio: 36.63, BH‐adjusted P=0.03). There were 2 pathways enriched for HS at nominal significance levels (valine, leucine and isoleucine biosynthesis, as well as purine metabolism, both P <0.05). C‐statistics for IS risk prediction were 0.629 (0.594–0.664) and 0.635 (0.601– 0.670) in the reference model and in the model additionally including ADMA/SDMA and glutamate, respectively. For HS risk prediction, the c‐statistics were 0.642 (0.592–0.692) and 0.647 (0.598–0.697), respectively. We did not observe any significant difference between the 2 models for IS or HS risk prediction (both P for difference >0.05) (Table [94]S4). In subgroup analyses, the associations of ADMA/SDMA with IS and HS risks were generally similar across subgroups of age, sex, BMI, smoking status, drinking status, regular exercise, hypertension, and diabetes (all P for interaction >0.05; Table [95]3). However, the association of glutamate with IS was more pronounced among participants with hypertension and those with diabetes (P for interaction=0.04 and 0.03, respectively; Table [96]4). Table 3. Associations of ADMA/SDMA With Incident Risks of IS and HS in Subgroups Subgroups IS HS OR (95% CI) P value P for interaction OR (95% CI) P value P for interaction Age 0.48 0.96 <65 y 1.12 (0.96– 1.30) 0.16 1.10 (0.88– 1.39) 0.38 ≥65 y 1.08 (0.95– 1.23) 0.23 1.22 (0.92– 1.60) 0.17 Sex 0.93 0.40 Male 1.08 (0.95– 1.22) 0.24 1.21 (0.96– 1.53) 0.11 Female 1.16 (0.98– 1.37) 0.08 1.04 (0.80– 1.35) 0.76 BMI 0.22 0.79 <24 kg/m^2 1.20 (1.02– 1.41) 0.03 1.00 (0.79– 1.27) 1.00 ≥24 kg/m^2 1.06 (0.94– 1.21) 0.33 1.19 (0.91– 1.57) 0.20 Smoking status 0.83 0.75 Current smoking 1.08 (0.89– 1.32) 0.43 1.17 (0.82– 1.66) 0.38 Noncurrent smoking 1.11 (0.99– 1.25) 0.07 1.16 (0.94– 1.43) 0.16 Drinking status 0.81 0.34 Current drinking 1.07 (0.88– 1.30) 0.52 1.12 (0.74– 1.69) 0.59 Noncurrent drinking 1.11 (0.99– 1.25) 0.07 1.21 (0.99– 1.47) 0.06 Regular exercise 0.59 0.33 No 1.03 (0.87– 1.22) 0.74 1.33 (0.83– 2.13) 0.24 Yes 1.14 (1.01– 1.28) 0.04 1.12 (0.92– 1.36) 0.26 Hypertension 0.62 0.18 No 1.05 (0.90– 1.24) 0.52 1.32 (0.95– 1.84) 0.10 Yes 1.13 (0.99– 1.28) 0.06 1.10 (0.89– 1.35) 0.38 Diabetes 0.99 0.89 No 1.09 (0.97– 1.21) 0.16 1.10 (0.91– 1.34) 0.30 Yes 1.06 (0.87– 1.29) 0.59 1.20 (0.76– 1.88) 0.43 [97]Open in a new tab ADMA/SDMA indicates asymmetrical/symmetrical dimethyl arginine; BMI, body mass index; HS, hemorrhagic stroke; IS, ischemic stroke; and OR, odds ratio. Table 4. Associations of Glutamate With Incident Risk of IS in Subgroups Subgroups OR (95% CI) P value P for interaction Age 0.16 <65 y 1.05 (0.91–1.22) 0.48 ≥65 y 1.22 (1.08–1.38) 1.92E−03 Sex 0.46 Male 1.11 (0.99–1.25) 0.09 Female 1.21 (1.03–1.41) 0.02 BMI 0.90 <24 kg/m^2 1.16 (1.00–1.34) 0.048 ≥24 kg/m^2 1.15 (1.01–1.31) 0.03 Smoking status 0.68 Current smoking 1.14 (0.94–1.38) 0.18 Noncurrent smoking 1.17 (1.05–1.31) 4.78E−03 Drinking status 0.33 Current drinking 1.22 (1.02–1.47) 0.03 Noncurrent drinking 1.12 (1.00–1.25) 0.048 Regular exercise 0.43 No 1.20 (1.00–1.44) 0.05 Yes 1.12 (1.00–1.25) 0.047 Hypertension 0.04 No 1.04 (0.90–1.21) 0.60 Yes 1.23 (1.09–1.39) 7.64E−04 Diabetes 0.03 No 1.09 (0.98–1.20) 0.12 Yes 1.46 (1.17–1.82) 9.40E−04 [98]Open in a new tab ORs were derived from unconditional logistic regression models with adjustment for age, sex, BMI, smoking status, drinking status, regular exercise, hypertension, atrial fibrillation, left ventricular hypertrophy, fasting glucose, serum total cholesterol, high‐density lipoprotein cholesterol, triglycerides, family history of stroke, and storage time of samples. In each subgroup analysis, the stratified variable was removed from the above list. BMI indicates body mass index; IS, ischemic stroke; and OR, odds ratio. Discussion In this nested case–control study based on middle‐aged and elderly Chinese people, we examined the associations of 55 plasma metabolites with risks of incident stroke and its ischemic and hemorrhagic subtypes. We validated the associations of ADMA/SDMA and glutamate with an elevated risk of total stroke for the first time in a Chinese population. Interestingly, we found both similarity and difference between IS and HS, reflected in the novel observation that ADMA/SDMA was nominally associated with both IS and HS, while glutamate was associated with only IS. We found that ADMA/SDMA was associated with an increased risk of stroke, which is consistent with previous studies. As an endogenous and nonproteinogenic amino acid, ADMA and SDMA have been extensively examined for their associations with cardiovascular risk for their well‐known function to inhibit the production of nitric oxide,[99] ^23 an endogenous gas that can relax the vascular smooth muscle cell and attenuate platelet adherence.[100] ^24 A meta‐analysis reported the association of increased ADMA concentration with a higher risk of total stroke derived from data from 7 prospective cohort studies, which were based on participants with European ancestry (n=8016) and had low heterogeneity (I^2 value=0% [95% CI, 0–71%]).[101] ^25 This study also observed a positive, though nonsignificant, association of SDMA with total stroke risk.[102] ^25 Our study extended the evidence a step further by providing validation of the association between circulating ADMA/SDMA and stroke risk in a Chinese population for the first time and further showed that the association of ADMA/SDMA was consistent for both IS and HS. Our finding highlighted that mechanisms involving disruption of endogenous nitric oxide synthesis may warrant further investigations as a shared mechanism underlying the development of both IS and HS. Another novel finding in our study was that glutamate was associated with an increased risk of IS but not HS. To date, only a case–cohort study based on 980 Spanish participants found that increased glutamate concentration in plasma was associated with an elevated risk of total stroke.[103] ^6 There was also epidemiological evidence that increased blood glutamate level was correlated with disruption in a wide spectrum of metabolic profiles, including elevated blood pressure and triglycerides, decreased high‐density lipoprotein, abnormal glucose metabolism, and insulin resistance.[104] ^26 , [105]^27 Similarly, we found that the association between glutamate and IS was significantly stronger in people with hypertension and diabetes than in those without these diseases. However, the association between glutamate and HS risk was not significant. This is probably because metabolic disorders have less impact on the risk of HS than IS.[106] ^14 We must acknowledge that, with a relatively limited number of HS cases, the observation of the null association of glutamate with HS risk needs further replication in other populations. We conducted exploratory analyses of metabolic pathways using the metabolites nominally associated with stroke. The difference in metabolic pathways between IS and HS was striking. For IS, it was significantly enriched in alanine, aspartate and glutamate metabolism, nitrogen metabolism, as well as arginine biosynthesis pathways, whereas for HS, only valine, leucine, and isoleucine biosynthesis as well as degradation were nominally significant. Differences between IS and HS have also been observed in other metabolomic studies.[107] ^11 , [108]^12 , [109]^13 They mainly focused on lipid metabolism and found multiple lipids associated with IS but not HS. These differences in associated metabolic pathways observed in this study might shed new light on the specific pathogenesis of stroke subtypes. In subgroup analyses, we found that the associations of ADMA/SDMA with IS and HS were similar across participants with different baseline risk profiles. Consistently, a prospective cohort based on 880 women did not find any significant interaction between ADMA and smoking status, diabetes, and obesity on the risk of stroke death.[110] ^28 Our finding supported this observation and suggested that the adverse effect of ADMA/SDMA in the development of IS and HS might be independent of other risk factors. At the same time, we observed synergistic interactions of glutamate with hypertension and diabetes. As already mentioned, increased glutamate concentrations had been reported to be associated with elevated blood pressure and abnormal glucose metabolism,[111] ^26 , [112]^27 which suggested a complex interaction of glutamate with other metabolic biomarkers. Our findings might provide promising metabolic markers of IS risk for at‐risk participants, particularly those with hypertension and diabetes. Our study has several strengths. First, our study is prospective in design and has a large sample size. Second, all stroke cases in our study were confirmed with computed tomography or magnetic resonance imaging, which ensured the accuracy of diagnosis. Third, we analyzed the associations of metabolites with risks of incident IS and HS in the same population, which facilitated comparison and could increase understanding of the similarity and difference between IS and HS. There are also some limitations in the present study. First, we measured only 55 known metabolites, which did not contribute to discovering new risk biomarkers for stroke. However, through the investigation of stroke subtypes, we identified novel associations of metabolites with both or certain stroke subtypes, which had been largely overlooked in previous studies. Second, we measured plasma metabolites only at baseline. However, it has been reported that metabolites in fasting plasma remained at relatively stable levels within 2 years,[113] ^29 reflecting the reliability of epidemiological studies relying on a single measurement of the metabolome. Nevertheless, investigations on the longitudinal changes in metabolites before stroke occurrence are also needed, because they may add new information to the understanding of the role of metabolites in stroke pathogenesis. Third, this study was based on a middle‐aged and elderly population; therefore, caution is advisable when extrapolating the results to other populations. Studies based on the young population, which has a rapidly growing disease burden of stroke, should also be carried out. Conclusions In this study, we validated the positive associations of ADMA/SDMA and glutamate with stroke risk, mainly that of IS, in a Chinese population. Interestingly, we found a novel unanimous association of ADMA/SDMA with both ischemic and hemorrhagic subtypes. Our findings provided potential intervention targets for stroke prevention. Further studies are needed to explore the underlying mechanism and potential intervention strategies. Sources of Funding This study was supported by grants from the National Natural Science Foundation of China (82 073 657, 82 192 902, 81 803 311, and 82 021 005) and the Fundamental Research Funds for the Central Universities (2019kfyXJJS034). Disclosures None. Supporting information Data S1 [114]JAH3-13-e033201-s001.pdf^ (396.8KB, pdf) Acknowledgments