1H-NMR-Based Metabolomics Study of Cerebral Infarction
Background and Purpose—Stroke is one of the leading causes of adult disability and death in developing countries. However, early diagnosis is difficult and no reliable biomarker is currently available. Thus, we applied a 1H-NMR metabolomics approach to investigate the altered metabolic pattern in plasma and urine from patients with cerebral infarctions and sought to identify metabolic biomarkers associated with stroke.
Methods—Metabolic profiles of plasma and urine from patients with cerebral infarctions, especially small vessel occlusion, were investigated using 1H-NMR spectroscopy coupled with multivariate statistical analysis, such as principal components analysis and orthogonal partial least-squares discriminant analysis.
Results—Multivariate statistical analysis showed a significant separation between patients and healthy individuals. The plasma of stroke patients was characterized by the increased excretion of lactate, pyruvate, glycolate, and formate, and by the decreased excretion of glutamine and methanol; the urine of stroke patients was characterized by decreased levels of citrate, hippurate, and glycine. These metabolites detected from plasma and urine of patients with cerebral infarctions were associated with anaerobic glycolysis, folic acid deficiency, and hyperhomocysteinemia. Furthermore, the presence of cerebral infarction in the external validation model was predicted with high accuracy.
Conclusions—These data demonstrate that a metabolomics approach may be useful for the effective diagnosis of cerebral infarction and for the further understanding of stroke pathogenesis.
Early diagnosis or prognosis of stroke is not straightforward. Although CT, MRI, and transcranial Doppler can be used for stroke diagnosis, they are costly, time-consuming, complex, and not universally available. Worse, small vessel occlusion (SVO), a subtype of cerebral infarctions that includes lacunar infarction, can be diagnosed with a normal CT/MRI examination only if a relevant brain stem or subcortical hemispheric lesion has a diameter of <1.5 cm and no potential cardiac sources of embolism exist. Moreover, no more than 50% greater stenosis of the large extracranial arteries and ipsilateral artery can occur,1 so diagnosis via traditional methods is difficult. Thus, the study of plasma biomarkers for the diagnosis and prognosis of stroke has been the focus of extensive research.2,–,6 However, at present, neither a biomarker nor a biomarker profile is generally accepted in clinical practice.
Additionally, stroke occurs because of a variety of complex factors, including cardiovascular disease and diabetes. For this reason, a new approach, such as “omics” technology for stroke, is required. In particular, metabolomics, an emerging field of omics technologies, can investigate the perturbed metabolic pattern in a complete set of metabolites in body fluids or tissue for early diagnosis and therapy monitoring and can clarify the pathogenesis of many diseases.7 To date, however, no metabolomics study for stroke has been reported. In this study, we investigated the perturbed metabolic pattern in plasma and urine from patients with cerebral infarction and identified metabolic markers associated with stroke using nuclear magnetic resonance (NMR) spectroscopy coupled with multivariate statistical analysis.
Subjects and Methods
Stroke patients with SVO1 by the TOAST classification and healthy volunteers were recruited from Daejeon Korean Oriental Hospital and Wonkwang Korean Oriental Hospital. Samples in this study were collected from patients with sudden onset of neurological symptoms who were admitted for a suspected ischemic event within 72 hours of onset.
The healthy individuals were free of medically manifest disease. Individuals with a stroke history or showing any sign of stroke, based on CT/MRI evaluation, such as TIA or silent infarction, were excluded from the healthy control group to reduce the confounding effects of additional risk factors; individuals with diabetes mellitus or other serious vascular diseases were also excluded from both patient and healthy control groups (information of subjects is in Supplemental Tables I and II, http://stroke.ahajornals.org).
Informed consent was obtained from all participants. The study protocol was approved by the Institutional Review Board of the Korea Institute of Oriental Medicine and the 2 medical centers that recruited the participants.
Sample Preparation and 1H-NMR Spectroscopy
Venous blood and urine were collected from overnight-fasted participants for standardized conditions. For the plasma experiment, 200 μL of plasma was mixed with 450 μL of saline solution (10% D2O for locking signal, 0.9% NaN3). Then, 600-μL aliquots of the supernatant were transferred into 5-mm NMR tubes for analysis. For the urine, 400-μL aliquots of supernatant were mixed with 230 μL of sodium phosphate buffer (0.2 mol/L, pH 7.0, 0.018% NaN3 dissolved) and adjusted to pH 7.0±0.1. Then, 540 μL of urine and 60 μL of D2O (5 mmol/L DSS dissolved) were transferred to 5-mm NMR tubes for analysis. 1H-NMR spectra of plasma and urine were acquired on a VNMRS-600MHz NMR spectrometer (Varian) at 25°C using a triple-resonance 5-mm HCN salt-tolerant cold probe. Details about NMR sequences and data handling of plasma and urine are in the Supplemental information (http://stroke.ahajornals.org).
Data Processing of the NMR Spectra and Multivariate Statistical Analysis
Each 1H-NMR spectrum from plasma and urine was segmented into equal width (0.005 ppm for plasma and 0.003 ppm for urine) corresponding to regions 0.75 to 8.6 ppm and 0.8 to 8.6 ppm, respectively. Spectral data were normalized to the total spectral area. Data files were imported into MATLAB (R2008a; Mathworks) and all spectra were aligned using the correlation-optimized warping method.8
The resulting data sets were then imported into SIMCA-P version 12.0 (Umetrics), and all data were then Pareto-scaled9 for multivariate statistical analysis. First, principal components analysis (PCA), an unsupervised pattern recognition method, was performed. Next, an orthogonal partial least-squares discriminant analysis (OPLS-DA) was used to maximize covariance between the measured data (peak intensities in NMR spectra) and the response variable (predictive classifications).
Analysis of NMR spectral data was accomplished using targeted profiling through the use of Chenomx NMR Suite 6.0 (Chenomx). OPLS-DA was applied to the concentration data of assigned metabolites. The regression coefficients plot was then used, which shows all metabolites that contribute to the separation between the studied groups.
Statistical Analysis and Validation
A 2-sided t test, Mann-Whitney U test (non-Gaussian distribution) for concentrations of metabolites, and Fisher exact tests for categorical data were performed using GraphPad Prism (version 5 for Windows; GraphPad Software). The default method of 7-fold internal cross-validation of SIMCA-P software was applied. External validations with acute phase stroke patients (within 24 hours from onset) with 80% of the data as the training set and the remaining 20% as the prediction set were performed. Detail about multivariate analysis and validation are in the Supplemental information.
Metabolomics Analysis of Plasma and Urine From Healthy Individuals and Stroke Patients
Figure 1 shows representative 1-dimensional 1H-NMR spectra of plasma and urine from healthy individuals and stroke patients. Spectra of plasma and urine were dominated by numerous metabolites, as shown by the metabolite key in Figure 1. Initially, principal components analysis was applied to examine the intrinsic variation in the plasma and urine data set. The score plots (Supplemental Figure I, http://stroke.ahajornals.org) showed slight separations between the 2 groups from plasma (R2X=0.757; Q2=0.514) and urine (R2X=0.613; Q2=0.245) models.
Next, OPLS-DA was performed to minimize the possible contribution of intergroup variability and to further improve the separation between the 2 groups. The score plots (Figure 2A, B) from OPLS-DA models for plasma (R2Y=0.914; Q2=0.778) and urine (R2Y=0.928;, Q2=0.627) show a clear differentiation between healthy individuals and stroke patients. The model parameters for the explained variation, R2, and the predictive capability, Q2, were significantly high (R2, Q2 > 0.5) in plasma and urine, indicating excellent models.
OPLS-DA loading and S-plots were generated to identify the metabolites responsible for the differentiation in the score plots (Figure 2C, D; S-plots not shown). Plasma of the stroke patients were characterized by higher levels of lactate, alanine, lipid CH2CH2CO, pyruvate, glycolate, and formate, and lower levels of very-low-density lipoprotein/low-density lipoprotein CH3, leucine, isoleucine, valine, very-low-density lipoprotein/low-density lipoprotein (CH2)n, lipid CH2CH2C=C, glutamine, methanol, 4-hydroxyphenylacetate, and τ-methylhistidine compared to those of the healthy individuals. In OPLS-DA loading plots, urine of stroke patients was characterized by decreased levels of citrate, dimethylamine, creatinine, taurine, glycine, and hippurate, and by increased levels of O-acetylcarnitine, trimethylamine-N-oxide, betaine, and carnitine.
To further investigate the significance of these metabolites, a combination of statistical approaches was necessary.10 Thus, we selected biomarkers using the t test, as well as jack-knifing of coefficients, and variable importance of projection values.
Among the higher levels of metabolites in the plasma of stroke patients compared with healthy individuals, lactate, glycolate, pyruvate, and formate showed statistically significant differences between the 2 groups. Among the lower levels of metabolites in stroke patients, very-low-density lipoprotein/low-density lipoprotein CH3, valine, lipid CH2CH2C=C, methanol, and glutamine showed statistically significant differences. In urinary metabolites, only citrate, dimethylamine, creatinine, glycine, and hippurate showed statistically significant differences (Table 1).
Additionally, OPLS-DA was applied to the targeted profile that we observed on the OPLS-DA loading plot. Based on the results of the OPLS-DA regression coefficients from plasma, stroke patients primarily excreted higher levels (variable importance of projection >1) of glycolate, formate, pyruvate, and lactate, whereas healthy individuals primarily excreted higher levels of methanol, glutamine, and lipid CH2CH2C=C. Stroke patients excreted lower levels (variable importance of projection >1) of citrate, creatinine, glycine, hippurate, and dimethylamine in urine (Supplemental Figure I).
To confirm the stroke pathway related with observed metabolites, homocysteine concentrations in urine model subjects were measured (Table 2). The geometric mean value of the total homocysteine was 2.05 μmol/L higher in stroke patients. Generally, the normal plasma concentration of total homocysteine concentrations after adjustment for gender and age is estimated to be up to 10 μmol/L.11 In our study, the proportion of patients with >10.0 μmol/L homocysteine was ≈2-fold higher than in control subjects.
Prediction of Class Membership
External validation was performed to test the reliability of the OPLS-DA model. In the plasma model, the corresponding training and prediction sets were randomly selected, and validation was repeated 3 times with equally numbered sets of different samples. The range for R2 was 0.885 to 0.915, and Q2 ranged from 0.723 to 0.738. The average classification rate was 96% for healthy subjects (9.6 of 10) and 100% for stroke patients (10 of 10; Figure 3A). In addition, acute phase patients (within 24 hours from onset and no overnight fasting) also had correctly predicted classes (16 of 16; Figure 3B). Furthermore, a false-negative diffusion-weighted imaging study result is not uncommon during the first 24 hours of ischemic stroke;12 therefore, the plasma of 4 false-negative diffusion-weighted imaging results were collected within 24 hours of onset, and 3 of 4 were correctly classified (data not shown).
Among 3000 SVO patients investigated in our study, 29.1% and 4.1% of SVO patients had diabetes and vascular disease, respectively. Therefore, ≈30% SVO patients with diabetes or cardiovascular disease might not fit the metabolic profiles suggested in this study, because patients with diabetes and vascular diseases were excluded from our study.
Nevertheless, potential biomarkers, such as glycolate, formate, pyruvate, lactate, methanol, glutamine, lipid CH2CH2C=C, creatinine, glycine, hippurate, and dimethylamine, observed from this study can explain the pathogenesis of stroke in a systemic review.
Among these metabolites, Graham et al2 suggested that lactate arises from a shift toward anaerobic glycolysis in potentially viable cells that continue to metabolize glucose under local hypoxic conditions. Alternatively, it may be inflammatory and phagocytic cell infiltration in the brain parenchyma after a stroke. Additionally, we observed significantly elevated levels of lactate and pyruvate from plasma, and decreased levels of citrate from urine. These are metabolites associated with energy metabolism, especially anaerobic glycolysis. Because limited oxygen exists after infarction, lactate and pyruvate increase from anaerobic glycolysis of serum glucose. Furthermore, because little pyruvate can enter the Krebs cycle under these conditions, citrate also decreases (Figure 4A).
We observed increased levels of formate and glycolate, and decreased levels of dimethylamine, glycine, hippurate, and methanol in stroke patients. These are strongly related to folic acid deficiency and hyperhomocysteinemia, which has been suggested to be an independent risk factor for stroke5,6 (Figure 4B).
As one of the folic acid activities, formate is oxidized via the folate pathway by combining with tetrahydrofolate through the mediation of formyltetrahydrofolate synthetase to form 10-formyl-tetrahydrofolate. Glycine is biosynthesized in the body from the amino acid serine via tetrahydrofolate. Thus, folic acid deficiency is related to increased formate and decreased glycine. Furthermore, these elevated formate levels may accelerate metabolic acidosis after stroke.13 Additionally, glycolate is formed from glyoxylate, which is a product of glycine catabolism. Therefore, as glycine decreases, glycolate increases. Moreover, it results in lower concentrations of hippurate, which is synthesized via glycine conjugation with benzoate in the liver.14
Among the effects of folic acid deficiency, hyperhomocysteinemia has been suggested as a major risk factor of stroke.5,6 Elevated total homocysteine induces oxidative injury to vascular endothelial cells and impairs the production of nitric oxide, a strong vascular relaxing factor, in the endothelium.15,16 Hyperhomocysteinemia also enhances platelet adhesion to endothelial cells and promotes the growth of vascular smooth muscle cells.16,17 Moreover, Iso et al18 reported that ischemic stroke, especially lacunar infarction (SVO type) had a greater association with homocysteine than did large-artery occlusion stroke or hemorrhagic stroke. These results are consistent with our data, which characterized an SVO subtype of ischemic stroke.
Synthesis of methanol in the body is attributable to S-adenosylmethionine.19,20 Thus, decreased methanol after cerebral infarction also may be related to hyperhomocysteinemia, which would cause a lack of methionine and S-adenosylmethionine.
Also, accumulated asymmetrical dimethylarginine, which is an endogenous inhibitor of nitric oxide synthase, has been proposed to be an important mediator of vascular dysfunction during hyperhomocysteinemia.21 Because hyperhomocysteinemia impairs dimethylarginine dimethylaminohydrolase activity, which inhibits accumulated asymmetrical dimethylarginine conversion to l-citrulline and dimethylamine,22 the level of dimethylamine decreases.
Recently, glial fibrillary acidic protein was found to be a sensitive marker of brain damage in patients with smaller lacunar lesions or minor strokes.23 Pekny et al24 also demonstrated that the genetic removal of glial fibrillary acidic protein is associated with an increase in intracellular glutamine concentration. Thus, postischemic upregulation of glial fibrillary acidic protein expression may be correlated with decreased glutamine concentrations.
We performed a 1H-NMR-based metabolomic approach in patients with cerebral infarction to identify potential biomarkers of stroke. Using OPLS-DA analysis, we detected a clear separation in the metabolic profiling, and metabolic pathways of cerebral infarction were characterized by anaerobic glycolysis, folic acid deficiency, and hyperhomocysteinemia.
In conclusion, this study suggested that the noninvasive approach of metabolic profiling with NMR spectroscopy can be used not only as a novel diagnostic technique but also as a tool for understanding pathogenesis of cerebral infarction.
Sources of Funding
Supported by National Research Foundation grant (No. R13-2008-028-01000-0 and No. 2010-0019394); funded by the Ministry of Education, Science and Technology of Korea (MEST); a grant from the Korea Institute of Oriental Medicine (KIOM, K10130); a grant from the Korea Basic Science Institute (T31409).
The online-only Data Supplement is available at http://stroke.ahajournals.org/cgi/content/full/STROKEAHA.110.598789/DC1.
- Received August 2, 2010.
- Accepted December 13, 2010.
- © 2011 American Heart Association, Inc.
- Adams HP Jr.,
- Bendixen BH,
- Kappelle LJ,
- Biller J,
- Love BB,
- Gordon DL,
- Marsh EE III.
- Graham GD,
- Blamire AM,
- Howseman AM,
- Rothman DL,
- Fayad PB,
- Brass LM,
- Petroff OA,
- Shulman RG,
- Prichard JW
- Graham GD,
- Hwang JH,
- Rothman DL,
- Prichard JW
- Dávalos A,
- Castillo J,
- Serena J,
- Noya M
- Howard VJ,
- Sides EG,
- Newman GC,
- Cohen SN,
- Howard G,
- Malinow MR,
- Toole JF,
- Stability of Plasma Homocyst(e)ine in Acute Stroke Patients (SHASP) Study Investigators
- Oppenheim C,
- Stanescu R,
- Dormont D,
- Crozier S,
- Marro B,
- Samson Y,
- Rancurel G,
- Marsault C
- Gatley SJ,
- Sherratt HS
- Tsai JC,
- Perrella MA,
- Yoshizumi M,
- Hsieh CM,
- Haber E,
- Schlegel R,
- Lee ME
- Iso H,
- Moriyama Y,
- Sato S,
- Kitamura A,
- Tanigawa T,
- Yamagishi K,
- Imano H,
- Ohira T,
- Okamura T,
- Naito Y,
- Shimamoto T
- Axelrod J,
- Daly J
- Stühlinger MC,
- Tsao PS,
- Her JH,
- Kimoto M,
- Balint RF,
- Cooke JP
- Herrmann M,
- Vos P,
- Wunderlich MT,
- de Bruijn CH,
- Lamers KJ