|Home | About | Journals | Submit | Contact Us | Français|
Type 1 diabetes was induced in Sprague–Dawley rats using streptozotocin. Rat urine samples (8 diabetic and 10 control) were analyzed by 1H nuclear magnetic resonance (NMR) spectroscopy. The derived metabolites using univariate and multivariate statistical analysis were subjected to correlative analysis. Plasma metabolites were measured by a series of bioassays. A total of 17 urinary metabolites were identified in the 1H NMR spectra and the loadings plots after principal components analysis. Diabetic rats showed significantly increased levels of glucose (P < 0.00001), alanine (P < 0.0002), lactate (P < 0.05), ethanol (P < 0.05), acetate (P < 0.05), and fumarate (P < 0.05) compared with controls. Plasma assays showed higher amounts of glucose, urea, triglycerides, and thiobarbituric acid-reacting substances in diabetic rats. Striking differences in the Pearson's correlation of the 17 NMR-detected metabolites were observed between control and diabetic rats. Detailed analysis of the altered metabolite levels and their correlations indicate a significant disturbance in the glucose metabolism and tricarboxylic acid (TCA) cycle and a contribution from gut microbial metabolism. Specific perturbed metabolic pathways include the glucose–alanine and Cori cycles, the acetate switch, and choline metabolism. Detection of the altered metabolic pathways and bacterial metabolites using this correlative and quantitative NMR-based metabolomics approach should help to further the understanding of diabetes-related mechanisms.
Type 1 diabetes is an autoimmune disease resulting from the selective destruction of insulin-producing beta cells in the pancreatic islets . It is growing in incidence and currently affects roughly 0.5% of the population in developed countries . Despite the vast understanding of its epidemiology, a definite solution for the prevention of type 1 diabetes or its deleterious effects is still not forthcoming. Causes of diabetes are believed to be associated with genetics, environmental factors, nutritional effects, or a combination of these. Many experimental studies have used simulated diabetes on animal models to understand the molecular events associated with this disease and its complications [3,4]. Streptozotocin (STZ)1 is well known to induce diabetogenic action by the selective destruction of pancreatic beta cells, and it is the most frequently used model for inducing diabetes mellitus [5–7]. The STZ-induced diabetic rat model provides a good platform to investigate diabetes using new approaches that may provide additional information on and insight into the mechanisms and effects of the disease.
The rapidly growing field of metabolomics (or metabonomics) that relates biological end points to multiple altered metabolite concentrations provides a wealth of biological information on complex systems [8–14]. It has been applied to a variety of diseases such as cancer [9,10], type 2 diabetes [4,15], and inborn errors of metabolism [16,17], and it has been used to study the effects of diet , drugs , toxins , and stress . Metabolomics combines data-rich advanced analytical techniques such as nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry with multivariate statistical analysis (MSA). NMR-based metabolomics is capable of simultaneously detecting a wide variety of extracellular (and intracellular in the case of tissue) metabolites that provide information on a large number of normal and abnormal biochemical pathways. To date, NMR spectroscopy has been applied to a number of diabetes studies [4,22–25]. NMR spectroscopy provides very high quantitative accuracy and reproducibility; is nondestructive, cost-effective, and rapid (typical requirements are a few minutes/sample); and requires limited or no sample preparation .
NMR data in combination with MSA allow the identification of potential biomarkers in biological specimens and an improved ability to identify specific metabolic pathways and networks associated with disease [1,17,18]. Widely used MSA methods such as principal component analysis (PCA) and partial least squares (PLS) discriminant analysis are applied for disease classification (through the use of scores plots) and biomarker detection (through the use of loadings plots) . The computation of the correlation  among the detected metabolites can be used to study biological networks in metabolomics ; however, such approaches remain at a preliminary stage. In principle, the application of the correlation matrix is straightforward and holds a great deal of promise for tracing specific metabolic pathways to investigate metabolic disturbances due to disease or other effects.
This study aimed to identify disturbed metabolic pathways in STZ-induced diabetes using a 1H NMR-based metabolomics approach. Absolute changes in the daily excreted quantities of detected metabolites (diabetic vs. control) and changes in the correlations among detected metabolites (for diabetic and control separately) were investigated. An integrated evaluation of metabolic changes in both quantities and correlations allowed us to trace specific disturbances in the metabolic pathways most associated with induced type 1 diabetes. We also compared our findings with current hypotheses regarding the detailed diabetes biochemistry.
3-(Trimethylsilyl)propionic acid–d4 sodium salt (TSP), α-tocopherol (all-rac), STZ, 2-thiobarbituric, d-(+)-glucose, ethanol, sodium pyruvate, sodium fumarate dibasic, allantoin, and sodium azide were obtained from Sigma (St. Louis, MO, USA) and were used without further purification.
Female rats (Sprague–Dawley strain from the animal quarters located at San Pablo–CEU University), all similar in development stage (12 ± 2 weeks), were studied. The diabetic group (n = 8) consisted of animals that received an intraperitoneal dose (50 mg/kg) of STZ and showed glucose levels in the blood of more than 200 mg/dl after 4 days. The control group (n = 10) consisted of age- and gender-matched rats. Throughout the study, both diabetic and control animals were kept under appropriate conditions (22 ± 2 °C temperature and 55 ± 10% relative humidity). Air was replaced with adequate frequency, and the rats were maintained under a 12-h light–darkness cycle. Animals were housed in collective cages (1820 cm3, not more than 6 animals per cage) with a bed of poplar shaving (29/12, Souralit, Spain), and all had free access to tap water and food (Harlan Global Diet 2014, Harlan Interfauna Ibérica, Madrid, Spain). Seven days after STZ administration, the rats were housed in metabolic cages and urine was collected and pooled every 8 h. Urine collection tubes contained hydrogen chloride (0.1 M) to avoid bacterial contamination. The total volume of urine for each rat excreted in 24 h was measured and stored at –80 °C. Urine samples were transported to Purdue University by air carrier over dry ice for metabolomics studies. On the day of sacrifice, animals were anesthetized with ketamine (75–100 mg/kg)/azepromacine (2.5 mg/ml) and blood was obtained by cardiac puncture in ethylenediaminetetraacetic acid (EDTA). The study was approved by the ethics committee of the San Pablo–CEU University (RD 223/1988) and is in agreement with Amsterdam Treaty and Spanish legislation.
Plasma glucose, triglycerides, cholesterol, and urea were analyzed using commercial kits from Spinreact (Gerona, Spain). Thiobarbituric acid-reacting substances (TBARS) were measured according to Viana and coworkers .
Frozen urine samples were thawed, and 500 μl was mixed with phosphate buffer (75 μl, 0.5 M, pH 7.0), D2O (75 μl), TSP (5 μl in D2O, 0.11 μmol), and sodium azide (5 μl in D2O, 12.3 nmol). Resulting solutions were centrifuged to remove particulate matter, if any, and 500 μl of the supernatants was transferred to 5-mm NMR tubes. All 1H NMR experiments were carried out at 25 °C on a Bruker DRX 500-MHz spectrometer equipped with an HCN 1H inverse detection probe with triple axis magnetic field gradients. 1H NMR spectra were acquired using the standard one-dimensional NOESY pulse sequence with water presaturation during the recycle delay of 3 s and a mixing time of 100 ms. Each dataset was averaged over 32 transients using 32,000 time domain points. The data were Fourier transformed after multiplying by an exponential window function with a line broadening of 0.3 Hz, and the spectra were phase and baseline corrected using Bruker XWinNMR software (version 3.5).
A total of 17 metabolites were identified in the 1H NMR spectra based on their characteristic chemical shifts and multiplicities. Individual metabolite quantities were then calculated by spectral integration of the full resolution data, taking into account the NMR signal intensity and number of protons that contributed to the measured NMR signals for both metabolites and the reference, TSP, and corrected using the total volume of urine excreted in 24 h.
Each NMR data set was binned to 4096 points (bin size 0.003 ppm) to minimize the effects of pH and ionic concentration differences. The data were aligned with reference to the TSP signal, and the regions containing TSP, urea, and residual water (4.6–5.3 ppm) signals were removed. Each NMR data set was then appropriately multiplied using the total 24-h urine volume. The data were mean-centered and divided by the square root of the standard deviation (Pareto scaling) to emphasize both large- and small-concentration metabolite signals. Subsequently, the data were subjected to unsupervised statistical analysis, PCA, using Pirouette software (version 3.11, Infometrix) to classify the samples (control or diabetic). P values for the individual metabolites were determined on the actual metabolite quantities using the un-paired t test available in the R software package (http://www.rproject.org). A Benjamini–Hochberg correction was applied to adjust the P values by accounting for the 17 metabolites used in the analysis. In addition, the Pearson's correlations among all of the metabolites (for diabetic and control samples separately) were also calculated using R. The statistical significance of the correlation coefficients was tested using a t test by giving a null hypothesis (H0) of r = 0, where r represents the correlation coefficient. If H0 holds, approximately follows the t distribution with degrees of freedom equal to n–2, where n represents the sample size. A low P value (<0.05) for this test means that there is evidence to reject the null hypothesis in favor of the alternative hypothesis or that there is a statistically significant relationship between the two variables (metabolites). Correspondingly, an absolute value of r larger than 0.6 (calculated for the control group) or 0.7 (calculated for the diabetic group) is considered as a statistically significant relationship between the two metabolites.
Selected demographic and blood parameters determined by biochemical analysis of diabetic and control rat samples are shown in Table 1. Mean values for plasma glucose, urea, α-tocopherol, TBARS, triglycerides, and liver α-tocopherol were higher in diabetes than control, whereas those for plasma protein and cholesterol were slightly lower. Diabetic rats consumed more water and excreted more than 10 times more urine (mean volume 129 ± 39 ml) than control rats (mean volume 9.6 ± 4.0 ml) in 24 h.
Typical 1H NMR spectra of urine from a diabetic rat and a control rat are shown in Fig. 1 along with the metabolite assignments. Diabetic rat urine spectra contained very high-intensity signals arising from glucose. In addition, the spectra contained numerous signals from the following small molecules: alanine, lactate, ethanol, acetate, pyruvate, 2-oxoglutarate, dimethylglycine, fumarate, succinate, citrate, formate, hippurate, urea, allantoin, dimethylamine, and creatinine. The assignments were made based on literature values , and tentative assignments made due to the overlap or slight shift in their positions (such as for glucose, ethanol, pyruvate, fumarate, and allantoin) were confirmed by rerecording 1H NMR spectra for urine before and after the addition of small quantities of the respective standard compounds obtained from commercial sources.
Table 2 shows the metabolite mean quantities and standard deviations in both diabetic and control groups as well as P values between the two groups. Glucose excretion was enormous in diabetic rats, with the quantity being nearly 7500-fold higher than in control rats (P = 0.000099). The metabolite with the second highest increase was lactate, with the amount being nearly 40-fold higher in diabetes (P = 0.045). Other significantly enhanced metabolites were alanine, ethanol, acetate, and fumarate (P < 0.05). The average quantities of pyruvate, formate, succinate, and dimethylglycine were 2- to 4-fold higher in diabetes than in controls but were not statistically significant. 2-Oxoglutarate, citrate, hippurate, allantoin, dimethylamine, and creatinine did not show any large or significant differences (Ps = 0.089–0.97).
The PCA scores plot for the quantitative NMR data shows well-separated clusters for disease and control rats (Fig. 2A) and a tight grouping of the control animals. The PC1 loadings plot (Fig. 2B) shows a large number of metabolites in addition to glucose that contributed to the separation. Supervised PLS discriminant analysis also was applied to the data after Pareto scaling and showed similar scores and loadings plots to those for PCA (Fig. S1 in Supplementary material). Although glucose signals dominate the loadings plot, other metabolites such as alanine, lactate, ethanol, acetate, pyruvate, 2-oxoglutarate, dimethylglycine, succinate, formate, and allantoin show positive loadings (similar to glucose). Citrate, creatinine, and hippurate have negative loadings (opposite to glucose). Fumarate does not clearly show up in the PC1 loadings plot, possibly because it has a relatively small variance across all samples as well as a low concentration. The additional contribution of metabolites other than glucose was further demonstrated by performing PCA again (Fig. 2C and D) after removing the glucose region, and the scores plot still shows a clear separation between the diabetic and control rats.
To investigate relationships among the metabolites, their quantities were correlated separately for diabetic and control rats using Pearson's correlation (Fig. 3). Correlation coefficients range from 1.0 (maximum positive correlation) to 1.0 (maximum anticorrelation), with a value of 0 representing no correlation. It is clear from the comparison of the two correlation plots in Fig. 3 that the changes between the diabetic and control animals were minimal for the correlation of glucose with alanine, pyruvate, 2-oxoglutarate, fumarate, citrate, allantoin, and creatinine. In diabetic rats, the correlation of glucose with urea was increased dramatically from 0.3 to 1, whereas the correlation of glucose with lactate decreased from 0.6 to 0. The correlation of citrate with alanine, lac-tate, and succinate was decreased from 0.9 to 0.4, from 0.8 to –0.3, and from 0.8 to –0.5, respectively, whereas the correlation of citrate with pyruvate, fumarate, and urea increased. Several metabolites showed a high positive correlation with one another in both diabetes and controls. For example, the metabolite pairs of acetate and ethanol, 2-oxoglutarate and citrate, 2-oxoglutarate and creatinine, and allantoin and creatinine showed high correlation (R > 0.9). Overall, for control rats, most metabolites correlate with one another with high correlation coefficients, as indicated by the predominantly red and orange regions in Fig. 3A, except that acetate, ethanol, and formate are relatively isolated from other metabolites and show blue stripes (|R| < 0.4) in the correlation network. In contrast, the increased blue regions in Fig. 3B for the diabetic animals indicate that a number of correlations are significantly decreased and show an overall decrease in metabolic correlation.
Upon STZ administration, plasma glucose showed elevated levels, with its quantity being nearly three times higher than in control rats, as anticipated. A number of other changes in blood parameters are also observed (Table 1). However, much more dramatic changes are seen in the urine, as indicated by the 1H NMR data. This is further shown using a quantitative metabolomics approach, in which a distinct classification of diabetic and normal rats is observed (Fig. 2A and B). Statistical analysis of the NMR data, even after removing the glucose signals, showed a number of metabolites contributing to the clear separation of diabetic animals from control animals (Fig. 2C and D). This is corroborated by the high metabolite mean ratios (>2) of 8 of the metabolites (Table 2). As described below, a detailed analysis of the metabolite correlation matrices (Fig. 3) in combination with their altered metabolite levels indicates a significantly altered metabolism in the glucose metabolism and tricarboxylic acid (TCA) cycles and the effect of gut microbial metabolism in rat diabetes. Fig. 4 compiles these observations based on the KEGG metabolic pathway database (http://www.genome.jp/kegg/pathway.html). Each metabolite is designated using a specific shape and is filled with a specific color to indicate correlations and altered quantities, respectively. Four types of shapes are used to indicate diverse correlations among the observed 17 metabolites, whereas the five different colors are used to indicate the changes in quantities of these metabolites. Three colored borders are used to differentiate the correlations with three key metabolites: glucose (blue), acetate (purple), and creatinine (pink). Black and purple arrows are used to differentiate mammalian pathways from bacterial pathways. The origin and implications of these relationships are discussed below in terms of the altered glucose metabolism, the TCA cycle, and symbiotic metabolism observed in this study.
Increased glucose in diabetes is due in large part to the enhanced endogenous glucose production (EGP) from noncarbohydrate precursors (gluconeogenesis) and the breakdown of glycogen (glycogenolysis). It has been shown that glycogenolysis was the major factor causing EGP in early stage insulin deficiency . At a later stage, especially for type 1 diabetes, EGP is attributed mainly to gluconeogenesis fueled by amino acids from degraded protein  or dietary ingestion. Although such findings are still somewhat in question due to the lack of reliability in the quantitative measurements [32,33], our studies support this hypothesis. In particular, significantly or moderately increased levels of alanine and urea with glucose in diabetes and their increased correlation with one another clearly suggest enhanced amino acids-fueled EGP in diabetes. The increased alanine concentration found in this study for the diabetic rats is in agreement with reports on human type 1 and type 2 diabetes [23,24]. An observed increased correlation of glucose with alanine but not with lactate in diabetic rats suggests that the EGP might be associated more closely with amino acids than with lactate even though the EGP effect from lactate is normally double that arising from alanine . The observations of severely enhanced lactate in both diabetic urine (Table 2) and plasma indicate a significantly reduced consumption of lactate compared with its production. The association of EGP with protein/amino acid metabolism in diabetes was also evident from the observed urea changes. The increased urea average content in both plasma and urine (Tables 1 and and2)2) in the diabetic animals suggests an increasing nitrogen load to the liver where urea is formed. We observed remarkable changes in the relationship between urea and glucose; there is essentially no correlation between these two metabolites (R = 0.3) in the control animals, whereas a strikingly high correlation between glucose and urea (R = 1.0) develops in diabetes (Fig. 3). The correlation agrees with a known fact that the amino-acid-fueled EGP enhances the nitrogen load on the liver and, consequently, urea synthesis is closely related to glucose synthesis . Because we observed minimal loss of body mass, the high correlation between urea and glucose in diabetic samples more likely indicates that high glucose is fueled mainly by ingested amino acids along with likely reduced protein synthesis.
A glucose homeostasis was observed in control rats from the positive correlation of glucose with both the gluconeogenesis precursors alanine, lactate, and pyruvate and the TCA cycle intermediates 2-oxoglutarate, fumarate, succinate, and citrate. This can be explained by the fact that a good control of glucose catabolism and anabolism is highly associated with the glucose–alanine and Cori (glucose–lactate) cycles between peripheral tissues and the liver, as demonstrated in Fig. 4. Both lactate and alanine are formed peripherally by glucose-derived pyruvate and are transported to the liver, where their carbon skeleton is reconverted to glucose. The maintenance of glucose homeostasis also demands a well-functioned TCA cycle capable of producing gluconeogenesis precursors and facilitating glucose oxidation. In addition to the observation that glucose correlates with the identified TCA cycle, it was found from our study that pyruvate (the glycolytic end product that enters the TCA cycle) correlates with the three detected TCA cycle intermediates: 2-oxoglurarate, fumarate, and citrate (Figs. 3A and and4).4). The metabolic profile from STZ-induced diabetes shows a dramatic perturbation of both glucose metabolism and the TCA cycle, as indicated in both Figs. 3B and and4.4. First, although glucose correlates highly with alanine, it fails to correlate with lac-tate (Fig. 3B), indicating that the Cori cycle might be perturbed. Second, the correlation of glucose with the TCA intermediates is markedly disturbed with decreasing or missing correlations with 2-oxoglutarate, fumarate, and succinate. Third, the correlations among the TCA intermediates are also significantly altered; for example, succinate no longer correlates with 2-oxoglutarate and citrate in diabetic rats.
An accumulation of acetate in STZ-induced diabetic urine suggests a defective acetate switch, that is, the molecular switch that regulates the dissimilation and assimilation of acetate . The elevation of acetate indicates an increased acetyl–coenzyme A (CoA) pool due to both exogenous (bacterial fermentation) and endogenous (mammalian) production [36,37]. Mammalian up-regulation of acetyl–CoA synthetase 2 (AceCS2) leads to an accumulation of acetate under ketogenic conditions . AceCS2 occurs in most organs but not in the liver, and its up-regulation in other organs stimulates liver excretion of acetate, which accumulates in the bloodstream [36,37]. In addition, the reduced expression of mammalian AceCS1 (which converts acetate to acetyl–CoA) might also contribute partially to acetate accumulation .
The observed correlations of acetate with ethanol, lactate, succinate, and formate suggest that the origin of these metabolites may be more microbial than mammalian. It is well known that gut bacteria produce these metabolites, and this evidence was also confirmed by an NMR study of gut fluid (jejunal aspirate) . Another in vitro 1H NMR spectroscopy study on bacteria such as Klebsiella pneumoniae showed that the bacteria produced acetate, succinate, and ethanol in the presence of a carbon source such as glycerol . Furthermore, the quantity of these bacterial metabolites increased significantly with increasing bacterial count, establishing their link with the bacteria . The high correlation of acetate with succinate in diabetes (R = 1) versus their moderate correlation in controls (R = 0.6) is in agreement with the intestinal microbial overgrowth in STZ-induced rats . A strong association between acetate and ethanol was observed in both control and STZ rats. Detection of acetate and ethanol in the rat digestive tract was reported previously [36,42], and their enhancements in diabetes [43,44] are consistent with our results. Alternatively, excess glucose can augment the carbon flux to the TCA cycle, inhibiting the expression of many TCA cycle enzymes and producing a number of partially oxidized metabolites (e.g., ethanol, lactate, formate, succinate). This situation is known as the bacterial Crabtree effect . In the current study, an explanation for the increased amounts of these bacterial metabolites in diabetic rat urine may be due to (i) an increase in bacterial metabolites produced in the gut of diabetic rats due to the microbial overgrowth, (ii) the Crab-tree effect, or (iii) both effects (i) and (ii). Significant levels of bacterial metabolites observed in diabetic rats and their possible origin, gut microbes, could have significant implications for understanding and controlling the secondary effects of these acids on diabetic patients. However, a further investigation is needed to verify this microbial hypothesis.
Choline is degraded via (i) mammalian pathways in which cho-line is converted to creatinine via dimethylglycine and (ii) symxenobiotic pathways in which choline is converted to methylamine by gut microbiota . A remarkable switch between dimethylglycine and dimethylamine was observed from our study. In controls, dimethylamine correlates highly with glucose, alanine, fumarate, and creatinine, whereas dimethylglycine fails to correlate with these metabolites. In diabetes, on the contrary, dimethyglycine correlates highly with glucose, alanine, fumarate, and creatinine, whereas dimethylamine fails to correlate with these metabolites. The reasons causing such a correlative switch are still unclear at this stage. The significant enhancement of dimethylglycine observed in STZ–diabetic rats may suggest an increased mammalian uptake of dietary choline. Alternatively, an enhanced conversion of choline to dimethylglycine may occur as compared with its conversions to dimethylamine (no observed difference between control and diabetes) or phosphatidylcholine (Fig. 4). The altered choline metabolism could be accounted for by the elevation of plasma triglycerides and liver weight in diabetic rats, as observed from our study (Table 1). High blood glucose levels in the presence of even low insulin levels stimulate the hepatic fatty acid and triglyceride synthesis . Insufficient conversion of choline to phosphatidylcholine exacerbates fatty liver and results in accumulation of triglycerides in plasma . It is possible that the conversion of choline to phosphatidylcholine was reduced due to its potentially enhanced conversion to dimethyglycine in diabetic rats. More studies are required to understand in detail the specific reasons for such a switch in the correlations among the metabolites of the choline pathway.
Elevation of lactate and pyruvate from the STZ–diabetic rats indicates an increased glycolysis. Because muscle cells in these STZ rats will be glucose starved and would likely oxidize all of the carbon from glucose through oxidative phosphorylation, the high levels of pyruvate and lactate might instead stem from glycolic production of lactate at the level of the red blood cells (RBCs) , which express a variant of the enzyme lactate dehydrogenase. It is also possible that other non-insulin-sensitive tissues are metabolizing glucose to lactate, which can be interconverted to pyruvate. This pathway may provide a good compensatory mechanism given that the monocarboxylase transporters, which carry lactate into and out of muscle cells, are not insulin sensitive. Thus, lactate would enter the cell and be converted to pyruvate. However, this scenario does not rule out the possibility of a gut microbial production contribution to the enhanced lactate and pyruvate concentrations. Further investigations need to be carried out to discern the origin of accumulated lactate and pyruvate.
The TBARS concentration in plasma has been used as an oxidative stress biomarker associated with the presence and amount of final fat degradation products induced by free radicals, and it has been shown to be increased in diabetes and to decrease after treatment with antioxidants [49–52]. This parameter has been highly criticized because of its lack of specificity , and although it should not be used alone to reach conclusions, it may add information if used in combination with other oxidative stress biomarkers.
In conclusion, this NMR-based metabolomics approach is useful for the identification and quantitation of high-concentration but important metabolites, and it provides significant information for improving the understanding of disease through the identification of altered metabolic pathways and networks. The urinary metabolites show dramatic changes in this rat diabetic model when compared with blood-based assays. Correlations among the detected metabolites show a number of striking differences between the diabetic and control rats. Metabolomics-based approaches that identify altered mammalian and symbiotic pathways observed in induced diabetes in rats will be useful for providing additional detailed knowledge concerning metabolic pathways associated with diabetes. Application of these methods to human diabetic patients and a comparison with the derived metabolite and pathway correlations may provide information for identifying additional factors associated with molecular genesis of diabetes and oxidative stress.
This work was supported by the National Institutes of Health (NIH) Roadmap Initiative on Metabolomics Technology (Grant NIH/NIDDK 3 R21/R33 DK070290-01), a Collaborative Biomedical Research Grant from Purdue University/Discovery Park, and Spanish MEC project AGL2005-06726-C04-03/ALI. The authors thank Ian R. Lanza for his highly insightful comments.
1Abbreviations used: STZ, streptozotocin; NMR, nuclear magnetic resonance; MSA, multivariate statistical analysis; PCA, principal components analysis; PLS, partial least squares; TSP, 3-(trimethylsilyl)propionic acid–d4 sodium salt; EDTA, ethylenediaminetetraacetic acid; TBARS, thiobarbituric acid-reacting substances; TCA, tricarboxylic acid; EGP, endogenous glucose production; CoA, coenzyme A; AceCS, acetyl–CoA synthetase; RBC, red blood cell.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.ab.2008.07.041.