Lipid Profiling Reveals Different Therapeutic Effects of Metformin and Glipizide in Patients With Type 2 Diabetes and Coronary Artery Disease

Cardiovascular disease (CVD) is the most common complication and the leading cause of mortality in patients with type 2 diabetes (1,2). In past decades, sulfonylureas and metformin, either alone or in combination, have been the cornerstones of drug therapy for type 2 diabetes (3). However, it remains unclear whether these two drugs have different effects on cardiovascular risk in addition to their ability to lower glucose (4–6). We recently evaluated the different effects of glipizide (a commonly used sulfonylurea) and metformin on the cardiovascular events and mortality in type 2 diabetic patients with coronary artery disease (CAD) in a multicenter interventional study (Study on the Prognosis and Effect of Anti-diabetic Drugs on Type-2 Diabetes Mellitus With Coronary Artery Disease [SPREAD-DIMCAD]) (7). The study results demonstrated that metformin therapy for 3 years substantially reduced the incidence of major cardiovascular events in a median follow-up of 5.0 years compared with glipizide, but no differences were identified between the two groups in terms of clinical risk factors, including plasma glucose, glycated hemoglobin, serum lipids, and blood pressure. Therefore, the potential benefit of metformin therapy on cardiovascular outcomes was not fully explained by changes in these conventional cardiovascular biomarkers.

However, several recent metabolomics and lipidomics studies have demonstrated that the high-throughput metabolite quantification by using emerging technologies can identify novel biomarkers associated with the risk of future diabetes and CAD (8–21). Wang et al. (9,14,15) and Würtz et al. (16) reported metabolic signatures that associated branched-chain and aromatic amino acids with future diabetes, insulin resistance, and other metabolic risk factors in humans. Floegel et al. (17) and Wang-Sattler et al. (18) found an independent association between certain metabolic alterations, including sugar metabolites, amino acids, and choline-containing phospholipids, and a higher risk of type 2 diabetes (and prediabetes). Fernandez et al. (11) described an association between eight lipid species and the incidence of CVD. Bernini et al. (13) reported that novel metabolites (e.g., 3-hydroxybutyrate, α-ketoglutarate, threonine, and dimethylglycine), in addition to the common lipid markers, might be related to the biochemistry of CVD risk. These results suggested that identifying different metabolite profiles will provide us valuable information concerning drug action on metabolism and enable us to elucidate molecular mechanisms of action. Here, we performed a comprehensive lipidomics study using liquid chromatography–quadrupole time of flight–mass spectrometry (LC-QTOF/MS) to evaluate the different lipid metabolites in serum samples from participants in the SPREAD-DIMCAD study and to elucidate the differential drug-induced effects on global lipid metabolites and their potential mechanisms of action regarding cardiovascular end points.

Synthetic lipid standards, including lyso-phosphocholines [LPC (12:0) and LPC (19:0)], phosphatidylethanolamines [PE (30:0) and PE (34:0)], and phosphatidylcholines [PC (34:0) and PC (38:0)], were obtained from Avanti Polar Lipids, Inc. (Alabaster, AL). Triacylglycerols [TAG (45:0) and TAG (51:0)] were purchased from Sigma-Aldrich Shanghai Trading Co. Ltd (Shanghai, China).

Distilled water was purified using a Milli-Q system (Millipore, Bedford, MA). Dichloromethane (CH₂Cl₂), acetonitrile, methanol (MeOH), and isopropanol of high-performance liquid chromatography (LC) grade were purchased from Tedia (Fairfield, OH). Analytical-grade ammonium formate and 98% formic acid were purchased from Sigma-Aldrich (St. Louis, MO).

The detailed design of the SPREAD-DIMCAD study has previously been published (7). Briefly, the study was a multicenter, randomized, double-blind, and placebo-controlled clinical trial with a total enrollment of 304 patients. Participants were randomly assigned to receive either glipizide (30 mg daily) or metformin (1.5 g daily) for 3 years. The study was initiated in 1 June 2004 with a median follow-up period of 5.0 years (range 3.7–5.7). The primary end points were time to the composite of recurrent cardiovascular events, including death from a cardiovascular cause, death from any cause, nonfatal myocardial infarction, nonfatal stroke, or arterial revascularization. The follow-up for the primary end points began at randomization and continued until the end of the study. The original protocol intended to collect serum samples from all the patients for metabolomic analysis during the 3 years of study drug administration. At study closure, serum samples at all four time points—0 (baseline), 1, 2, and 3 years on the study drug administration—had been obtained from 21 glipizide-treated patients and 23 metformin-treated patients; these samples were used for the final lipidomics analysis (Supplementary Fig. 1). The detailed characteristics of these 44 patients at baseline and at the end of follow-up are summarized in Table 1. The main reasons for excluding the original participants from the present analysis were the following: 1) premature termination of study drug administration (31 in the glipizide group and 32 in the metformin group), 2) withdrawing consent for serum sample collection at any point during follow-up each year (71 in the glipizide group and 79 in the metformin group), 3) death before the end of 3 years on study drug administration (8 in the glipizide group and 4 in the metformin group), or 4) other reasons (17 in the glipizide group and 18 in the metformin group).

All the serum samples were collected in the morning after an overnight fasting for 10–12 h and no smoking. The samples were frozen immediately and stored at −80°C until assayed. The study was approved by the institutional review board of Ruijin Hospital, and written informed consent was obtained from each patient. The study was conducted in accordance with the principles of the Declaration of Helsinki.

Serum lipids were extracted by a modified Bligh/Dyer extraction procedure as previously described (22,23). Briefly, 30 μL internal standard mixture (see Supplementary Table 1 for details) was added to 30 μL serum followed by the addition of 540 μL CH₂Cl₂/MeOH (2:1, v/v). After thoroughly vortexing the resulting mixture, 120 μL water was added to form a two-phase system. The lipids were located in the bottom organic phase. After centrifugation at 6,000g for 10 min at 10°C, 100 μL of each lipid extract was transferred for further LC–mass spectrometry (MS) analysis. Before injection, the lipid extracts were diluted fivefold with acetonitrile/isopropanol/water (65:30:5, v/v/v), and 10 μL of each diluted lipid extract was loaded for the lipid profiling analysis.

The LC-MS lipidomics analysis was performed on a Q-TOF/MS (Agilent) coupled to an ultra-fast LC system (Agilent). An Ascentis Express C8 column (2.7-μm particle size, 2.1 × 150 mm; Sigma-Aldrich, Munich, Germany) was used for the LC separation. The LC separation conditions were identical to those described previously (22). The autosampler plates were maintained at 12°C. Serum lipid profiling was performed on an Agilent Q-TOF/MS equipped with a dual-electrospray ion source. Full MS scans were acquired in the positive-ion mode. The ion-source temperature was set at 300°C. The voltage of the capillary was set at 4 kV. The voltages of the fragmentor and skimmer were set at 230 V and 65 V, respectively. The flow rate of the drying gas was 11 L ⋅ min⁻¹. The analytes were acquired using a mixture of 10 μmol/L purine (m/z 121.0508) and 2 μmol/L hexakis phosphazine (m/z 922.0097) as lock masses to ensure mass accuracy and reproducibility. The data were collected at a mass range of m/z 400–1100 with an ion scan duration of 20 ms using LC-MS solution software (Agilent). All the study samples were randomly analyzed. Quality-control samples generated by pooling all of the serum samples were regularly inserted in the sequence to monitor the response of the LC-MS system and to assess the lipid profiling platform.

The data are presented as the means and SD or as the median with the interquartile range. Univariate statistical analyses were performed with SPSS for Windows, version 13.0, software (SPSS, Chicago, IL). P values <0.05 were considered to be statistically significant. Within-group comparisons were performed using paired-sample t tests to evaluate the differences from baseline in each group. Student t test (for data that were normally distributed) or the Mann-Whitney test (for data that were not normally distributed) and a one-way ANOVA with the two-sided Dunnett post hoc test for multiple comparisons were performed to investigate the differences between groups. During the lipid profiles analysis, the appropriate lipid internal standard was used to correct the signal intensities of each lipid. Multivariate statistical analyses were performed using SIMCA-P software (version 11.5; Umetrics AB, Umeá, Sweden). A Cox regression model adjusted for the duration of diabetes, the duration of CAD, age, sex, and smoking history at baseline was used to calculate the hazard ratios (HRs) of composite cardiovascular events for individual lipids in each follow-up year (using baseline values as the reference).

A modeling strategy called scaled-to-maximum, aligned, and reduced trajectories (SMART) analysis was performed to visualize the multivariate response similarity and to facilitate the interpretation of the lipidomics data (24). Briefly, the average spectrum of the baseline measurement for each drug treatment was subtracted from all of the spectra for each study (pretreatment subtraction). Then, the vector magnitudes were calculated as the square root of the sum of the squares of the obtained values for each adjusted spectrum. The average magnitude at each time point for each drug treatment group was determined. The largest magnitude for each drug treatment group was used to evaluate the overall magnitude of the therapy-related effect and to scale the treatment group data to a common magnitude. In the end, the average spectra at each time point were analyzed.

During the 3-year study drug administration, serum samples were obtained from 21 glipizide-treated patients and 23 metformin-treated patients at each time point (0 [baseline], 1, 2 and 3 years of study drug administration) and subjected to the lipidomics analyses. Detailed characteristics of the 44 patients at baseline and at the end of follow-up are summarized in Table 1 and Supplementary Tables 2 and 3. BMI, body weight, and waist circumference were significantly lower in the metformin group than in the glipizide group after treatment (Table 1); there were no significant differences in the other clinical characteristics between the glipizide and metformin groups at either baseline or the end of the 3-year treatment. The distribution and doses of the glucose-lowering agents at baseline and after treatment were generally similar, except for the use of insulin, and there were no significant differences between the two groups regarding other concomitant medications, except for statin use at baseline (Table 1).

Lipid profiling was performed on fasting serum samples obtained at baseline and at 1, 2, and 3 years of drug treatment for all the study subjects. A total of 118 serum lipid molecular species were identified and quantified using the LC-MS lipidomics approach. To compare the treatment effects of the two drugs on the serum lipidome over time, SMART was used to analyze the lipidomics data from the two parallel drug treatment groups at different time points. Figure 1 illustrates the SMART analysis of the serum LC-MS data after 0, 1, 2, and 3 years of intervention with glipizide and metformin. The two trajectories clearly point in similar directions, indicating that the lipid metabolic responses were similar. However, the metformin treatment trajectory was more open than the glipizide treatment trajectory, suggesting that metformin substantially affected serum lipid metabolism in type 2 diabetic patients with CAD, whereas the effect of glipizide on the serum lipid profile was limited compared with the baseline. SMART has been demonstrated to significantly reduce the risk of misinterpreting the results of principal component analysis (24). One-way ANOVA of the LC-MS lipidomics data for the entire cohort revealed that a total of 10 lipid species were significantly changed in the glipizide group and that 23 lipids were significantly changed in the metformin group (Table 2). Among these significantly changed lipids, both drug treatments significantly increased PC lipids and significantly decreased sphingomyelin (SM) lipid species compared with baseline.

For further investigation of the lipid metabolic differences between metformin and glipizide, all the lipids were analyzed in the metformin-glipizide comparison after eliminating individual differences in lipid metabolism (i.e., the differences in the baseline lipid profiles for each treatment were eliminated). Specifically, the amount of each lipid species in the serum samples from patients at 1, 2, and 3 years of drug treatment was first normalized to the lipid concentration at baseline (i.e., serum lipid data from patients at 1, 2, and 3 years of drug treatment were divided by the baseline data). Then, the geometric ratio of each analyzed lipid concentration after normalization was calculated in the fasting serum samples from those who received metformin versus those who received glipizide at the 1-, 2-, and 3-year time points. With use of this approach, differences between the two medications and specific lipid species (i.e., potential lipid biomarkers) that may contribute to the observed clinical effects over time could be identified. The results of the metformin-glipizide comparison for all the lipid analytes are presented in Fig. 2A–C (1, 2, and 3 years of therapy, respectively). The differences in the lipid metabolites between these two treatments are directly reflected by the P values plotted on the y-axis. The data indicated that the difference between the glipizide and metformin groups was limited at 1 year, whereas the treatment differences at 2 and 3 years were larger. In detail, only PC (36:5) (metformin vs. glipizide geometric ratio = 1.91, P = 0.046) and PC (O-36:3) (metformin vs. glipizide geometric ratio = 1.29, P = 0.020) were significantly different between the groups after 1 year of treatment. However, 11 lipid species, including cholesterol ester (20:4) (metformin vs. glipizide geometric ratio = 0.62, P = 0.043), PC (34:3) (0.68, 0.030), PC (O-34:1) (0.45, 0.003), PC (O-34:2) (1.72, 0.030), PC (O-36:4) (2.67, <0.001), PC (O-38:5) (0.76, 0.038), PC (O-38:6) (0.67, 0.039), SM (d18:1-14:0) (0.17, 0.014), SM (d18:1-16:1) (0.27, <0.001), TAG (44:0) (1.92, 0.018), and TAG (44:1) (1.90, 0.024), were significantly different after 2 years of treatment. Twelve lipid species, including LPC (16:1) (metformin vs. glipizide geometric ratio = 0.69, P = 0.032), LPC (18:1) (0.67, 0.006), LPC (20:3) (0.74, 0.028), LPC (20:4) (0.59, 0.004), LPC (22:6) (0.73, 0.029), PC (34:0) (0.77, 0.035), PC (O-34:2) (2.35, <0.001), PC (O-36:4) (2.14, <0.001), PE (36:4) (0.76, 0.013), PE (38:6) (0.63, 0.034), SM (d18:1-14:0) (0.23, <0.001), and SM (d18:1-16:1) (0.34, <0.001), were significantly different after 3 years of treatment. The lipid profiling data indicated that metformin had a significantly greater effect on serum lipid metabolism than glipizide in type 2 diabetic patients with CAD in our study. When all the significantly changed lipid species in the 44 patients were analyzed together, three lipid metabolites, PC (O-34:1) (HR 28.673 [95% CI 1.373–598.997], P = 0.030), SM (d18:0-24:1) (3.797 [1.120–12.880], P = 0.032), and SM (d18:1-20:1) (73.040 [3.183–1675.890], P = 0.007), at different follow-up points (at 1 year, 1 year, and 2 year, respectively), were associated with long-term composite cardiovascular events with use of the adjusted Cox regression model.

It has been reported that TAG profiling can improve the outcome prediction for diabetic patients (8). To further explore which medication was more beneficial for the treatment of CAD and diabetes, the TAG pattern associated with the reported diabetes risk was investigated by comparing the metformin and glipizide group data for the TAG lipid analytes measured in the current study. Briefly, the geometric mean ratios of the TAG levels after normalization (the TAG levels in treated patients were normalized to those at baseline) in metformin-treated patients and glipizide-treated patients were compared with a focus on the acyl chain carbon number and the double bond content to determine the lipid effects of glipizide and metformin on serum TAGs in these patients. The results are presented in Fig. 3A–C (at 1, 2, and 3 years of therapy, respectively). Fig. 3 shows a clear trend of reduced TAG lipids with a relatively lower carbon number and of increased TAG with a relatively higher carbon number in the metformin group versus the glipizide group during the 3-year treatment period. There were minimal changes in TAG molecules with a different number of double bonds (a slight increasing pattern was only observed for 0–3 double bonds) in the metformin versus the glipizide group throughout the 3-year treatment (Fig. 3). The observed pattern changes, regarding both the number of acyl chain carbons and double bonds, were noted among TAGs but not among other lipid classes (Supplementary Fig. 2).

The present lipidomic analyses were based on a previously reported long-term interventional study, which is likely the first report on cardiovascular outcomes comparing the long-term effects of glipizide and metformin on the major cardiovascular events in type 2 diabetic patients with a history of CAD (7,25,26). The results demonstrated a beneficial effect of metformin on cardiovascular outcomes compared with glipizide. However, because no differences were identified between the two groups after treatment regarding traditional clinical risk factors, such as plasma glucose, glycated hemoglobin, serum lipids, and blood pressures, the potential effects of metformin on CVD could not be explained by the clinical findings.

Because dyslipidemia is strongly associated with cardiovascular events, especially in the diabetic population (27), lipid management in patients with type 2 diabetes has the potential to reduce the risk of concomitant CAD. Moreover, lipids are the fundamental components of cellular membranes and are essential in lipidomics because they represent the biochemical activity signature during lipid metabolism and are therefore closely related to observable phenotypes. With this knowledge, lipidomics can be viewed as the process of defining multivariate lipid metabolic trajectories that represent the systemic response (i.e., holistic lipid metabolic changes) of a living system to pharmaceutical interventions over time.

In the current study, the results of the SMART analysis (Fig. 1) indicated that metformin had a substantially greater effect on serum lipid metabolism than glipizide. Such a metabolic trajectory analysis is essential for depicting the time-dependent metabolic behaviors resulting from a specific intervention and is useful in clinical diagnosis. Furthermore, by using LC-QTOF/MS lipid profiling to analyze 118 molecular lipid species across 7 lipid classes and 10 subclasses, we identified a panel of individual lipid species that were significantly altered in response to drug treatment and that were significantly different between the two treatment groups. The altered individual lipid species mainly belonged to the LPC, PC, PC-O, TAG, SM, and PE lipid classes. The role of these lipid classes in the pathogenesis and progression of diabetes and atherosclerosis has recently been recognized (28–37). Moreover, PC (O-34:1), SM (d18:0-24:1), and SM (d18:1-20:1) were associated with an increased risk of long-term composite cardiovascular events in our study. Although further studies are needed to elucidate the biological mechanism accounting for the link between these specific lipid molecular species and the risk of cardiovascular events, these findings indicated that the systematic analysis of serum lipid species, rather than lipid classes as a whole, may reveal the differential effects of antidiabetes agents on future CVD risk beyond the improvement of clinical biomarkers.

By viewing acyl chains in their natural context across distinct macromolecular species, we demonstrated an increasing trend in TAG acyl chain carbon number and a slight increase in TAGs with 0–3 double bonds in the metformin group compared with the glipizide group. Although the risk pattern for the progression of CVD indicated by acyl chain carbon number and double bond content remains unclear, several recent studies have demonstrated a relationship between plasma TAGs and future diabetic risk or insulin action (8,38). In the Framingham Heart Study cohort (FHS), Rhee et al. (8) found that TAGs of lower carbon number and double bond content were associated with an increased risk of type 2 diabetes, whereas TAGs of higher carbon number and double bond content were associated with a decreased risk of type 2 diabetes. Furthermore, they found that TAGs of lower carbon number and double bond content were elevated in the setting of insulin resistance and that TAGs of higher carbon number and double bond content had the weakest correlation with insulin resistance. Based on these observations, our findings of increased TAG lipid species with higher carbon number in metformin-treated versus glipizide-treated patients during the 3-year treatment period indicate that long-term metformin treatment would be more beneficial for patients with both type 2 diabetes and CAD. Schwab et al. (38) also determined that a sustained increase in insulin sensitivity was associated with a similar pattern in TAG changes. All of these results combined with our current findings might indicate a beneficial effect of certain TAG species in diabetic patients with future cardiovascular risk.

The major strength of our findings is the LC-MS–based lipidomic analysis and the breadth of the lipids that were analyzed. Because the lipidomic profiling was based on our previous multicenter, randomized clinical trial, the results provide new evidence regarding the underlying mechanisms of disease progression and the effects of different drug therapies on cardiovascular outcomes in these high-risk patients. However, our study had several limitations. First, serum samples at all four time points were only available for 44 study participants out of a total of 304 patients owing to various reasons described above. However, the baseline characteristics of the 44 participants were similar to those of the total participant population (Supplementary Table 4) and of the remaining participants who were not involved in the present analysis (data not shown). Regarding the lipidomic profiling, there was no significant difference in the relative concentration of individual lipid metabolites at baseline between the 44 participants involved in the lipid analysis and the general population for whom baseline lipidomics data were available (n = 116) (data not shown). Furthermore, we performed a validation analysis using the participants with available serum sample at baseline and at 1 year of study drug administration to confirm our current findings (n = 51 in the glipizide group, and n = 58 in the metformin group). As expected (Supplementary Fig. 3), the results were similar to those obtained in the current analysis (i.e., TAGs with a higher carbon number were increased in the metformin group vs. the glipizide group after 1 year of treatment) (Fig. 3). Therefore, although the sample size was small, it was representative of the general patient population. Second, statin use at baseline was different between the two groups (90.5% in the glipizide group vs. 47.8% in the metformin group, P = 0.004). However, comparing statin use in the two groups revealed that from 1 year on, there was no difference between the two groups at each follow-up year (P = 0.062, 0.146, and 0.095 at 1, 2, and 3 years of study drug administration, respectively) and that the percentage of patients using statins in both groups decreased during the intervention. Moreover, the concentrations of total cholesterol, cholesterol ester, and lipoproteins were not different between the groups at baseline and after 3 years of treatment, suggesting a minor influence of the different treatment with statins at baseline. Thus, although we could not exclude the impact of statins on the lipid results, the confounding effects of statins were not supported by our data. Nevertheless, caution shall be exercised in interpreting the current findings owing to the relatively small sample size.

In conclusion, by applying LC-MS–based lipidomics and measuring biochemical parameters, we revealed the differential therapeutic effects of metformin and glipizide on comprehensive lipidomics, which were comparable with their different long-term effects on cardiovascular outcomes. Metformin more substantially changed serum lipid species compared with glipizide. Among the altered lipid species, three lipid metabolites were linked to long-term composite cardiovascular events. Furthermore, after treatment, TAGs of higher carbon number increased in the metformin group compared with the glipizide group. Based on these findings, we speculated that a lipidomics approach may be useful in elucidating the complex mechanism of action of particular drugs and presents a useful tool for probing the mechanisms of progression of long-term CVDs. Future studies will be required to precisely evaluate the predictive findings in additional cohorts and to determine whether we have identified an early marker and/or an effector of cardiovascular disease and its associated therapeutics.

Funding. This study was supported by the Ministry of Science, Technology and Innovation Fund and projects (No. 2011YQ030114), the National Basic Research Program (No. 2012CB517506) from the State Ministry of Science and Technology of China, the grants (No. 81170784, No. 21175132) and the creative research group project (No. 21321064) from the National Natural Science Foundation of China, the Program for Innovative Research Team of Shanghai Municipal Education Commission, the Sector Funds of Ministry of Health (No. 201002002), the National Key New Drug Creation and Manufacturing Program of Ministry of Science and Technology (No. 2012ZX09303006-001), the Shanghai Committee on Science and Technology (10dz1920802), and the Fund of Shanghai Municipal Health Bureau (No. 2012-244).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Y.Z. and J.H. contributed to the study design, data analysis, and primary drafting of the manuscript. C.H. contributed to the lipid profiling, statistical analysis, and primary drafting of the manuscript. J.Ze. contributed to the sample processing and lipid profiling. S.L. reviewed and edited the manuscript. A.L., D.Zh., W.W., and W.S. contributed to the design and implementation of the protocol and to the discussion. Q.S., Y.D., Z.Z., W.T., J.Zh., L.C., D.Zo., D.W., H.L., C.L., G.W., and J.S. contributed to the implementation of the protocol and coordinated the discussion. G.N. designed and implemented the protocol and was the lead author. G.X. contributed to the study design, data analysis, and review of the manuscript. G.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.