Abstract
Aims
Individual heterogeneity in the glucose-lowering response to glucagon-like peptide-1 receptor agonists (GLP-1RAs), including exenatide, limits efficient treatment selection for type 2 diabetes mellitus (T2DM). This study assessed whether baseline clinical characteristics together with serum metabolomic features could predict the glucose-lowering efficacy of exenatide.
Methods
A total of 93 Chinese adults with T2DM received exenatide treatment for 16 weeks. Treatment response was defined by glycated hemoglobin (HbA1c) change value (ΔHbA1c): responders (ΔHbA1c ≤ −0.3%, n = 70) and non-responders (ΔHbA1c > −0.3%, n = 23). Baseline serum metabolites were profiled by non-targeted liquid chromatography–mass spectrometry. Predictors of exenatide-induced glucose-lowering response were screened and modeled using univariate and multivariate logistic regression analysis and evaluated with receiver-operating characteristic (ROC) analysis.
Results
At baseline, responders presented with a higher HbA1c level and a lower HDL-C level than non-responders. Logistic regression analysis indicated that baseline HbA1c and HDL-C levels were associated with ΔHbA1c after treatment. Metabolomic comparison analysis identified 15 discriminative serum metabolites between two groups. Among these metabolites, butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) were correlated with the exenatide-induced glucose-lowering response. Baseline clinical characteristics (higher HbA1c level and lower HDL-C level) combined with the metabolomic features [lower butenylcarnitine, higher LysoPC(18:2(9Z,12Z)) and higher PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)] predicted better glucose-lowering response to exenatide, with a sensitivity, specificity and area under the ROC curve of 78.1%, 90.0% and 0.895, respectively.
Conclusions
Combination of the baseline clinical characteristics (HbA1c and HDL-C) and metabolite profiles [butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z))] can effectively predict glucose-lowering efficacy of exenatide in patients with T2DM.
Keywords: Exenatide, Non-targeted metabolomics, Glucose-lowering efficacy, Type 2 diabetes mellitus
1. Introduction
The continuing rise in diabetes prevalence has become a major global health concern. By 2024, 589 million adults were estimated to be living with diabetes worldwide, with projections reaching 853 million by 2050 [1]. Effective glycemic control can significantly reduce diabetic complications and mortality in diabetic patients [2,3]. However, achieving safe and effective glycemic control through antidiabetic drugs remains a challenging and time-consuming task. For instance, only approximately 64.1% of adult patients in China meet the targeted level of glycated hemoglobin (HbA1c) [4]. Therefore, it is necessary and urgent to optimize the hypoglycemic effect of medications and achieve an improved glycemic control.
For patients with type 2 diabetes mellitus (T2DM), glucagon-like peptide-1 agonists (GLP-1RAs) provide multi-dimensional benefits. GLP-1RAs improve glucose control by stimulating insulin secretion and suppressing glucagon in a glucose-dependent manner, and can also support weight reduction through appetite suppression and delayed gastric emptying [5,6]. Exenatide, the first marketed GLP-1RA, has been widely prescribed worldwide. GLP-1RAs generally lower HbA1c level by approximately 1.0∼1.5%, which indicates their high efficiency in lowering blood glucose level [7]. However, the treatment responses vary markedly across individuals with T2DM. To address this heterogeneity and advance precision medicine, identifying biomarkers that can predict which patients will derive the greatest benefit is crucial [8,9].
Metabolomics, systematic profiling of small-molecule metabolites, emerged after genomics, transcriptomics, and proteomics. Its application in precision medicine, particularly pharmacometabolomics, seeks to correlate baseline metabolomic features with therapeutic outcomes [10]. This is exemplified by studies on drugs such as aspirin, simvastatin and acamprosate [11]. In our previous study, a metabolomic combination was shown to predict the exenatide-associated improvement in endothelial injury [12].
In the present study, 93 patients with T2DM who were treated with exenatide were included. We aimed to evaluate whether baseline clinical and metabolomic data could predict the glucose-lowering efficacy of this drug. Our findings intended to inform diabetes treatment strategies and support more precise disease management.
2. Methods
2.1. Study design and subjects
Participants were drawn from a previously published randomized, multicenter, non-inferiority clinical trial (ChiCTR-IPR-15006558) [13]. The parent trial enrolled patients aged 20–70 years with T2DM (HbA1c 7.0–10.0%) who were on metformin and/or insulin secretagogues. Subjects were randomized to branded (Byetta®) or generic exenatide, with a standardized subcutaneous dose escalation for 16 weeks (5 μg to 10 μg, twice daily). To rule out the possible bias introduced by the branded and generic drugs, only 120 patients treated with branded exenatide were included in this post hoc analysis. Of these, 27 were excluded due to protocol violation, loss to follow-up, adverse events, missing metabolomic data, or refusal to continue. Therefore, 93 patients were included in the final analysis.
2.2. Blood collection and measurement
Fasting blood samples were obtained at baseline and at week 16 after treatment. Serum was separated by centrifugation at 4000 rpm for 10 min at 4 °C and stored at −80 °C until testing. The serum was subsequently subjected to comprehensive laboratory testing, including HbA1c levels, glucose profiles, lipid profiles, and hepatic and renal function parameters.
2.3. Non-targeted metabolomic analysis
The detailed method of non-targeted metabolomic analysis has been previously reported [12]. Briefly, a DIONEX Ultimate 3000 ultrahigh-performance liquid chromatography (UPLC) system (Dionex Corporation, Sunnyvale, CA, USA) connected to an ESI‒Quadrupole time-of-flight tandem mass spectrometry (QTOF/MS) mass spectrometer (Thermo Fischer Scientific, Waltham, MA, USA) was employed for metabolomic analysis. The samples were processed according to standard procedures and molecular features were recorded with TraceFinder software (Thermo Fischer Scientific).
2.4. Metabolomic data processing
Response classification used HbA1c change value (ΔHbA1c) with a cutpoint of −0.3%, a widely recognized threshold of effectiveness for glycemic-lowering agents [[14], [15], [16]]. Baseline metabolite differences were first explored using orthogonal partial least-squares discriminant analysis (OPLS-DA) in SIMCA-P (version 13.0; Umetrics, Umeå, Sweden). Data quality was monitored by randomly inserting quality control samples in each analytical batch. Candidate predictive biomarkers were subsequently selected by integrating multiple criteria: a variable importance in projection (VIP) value > 1 from the OPLS-DA S-plot, a P-value <0.05 in the Wilcoxon rank-sum test, and significance in logistic regression analysis. Significant metabolites were also confirmed with an false discovery rate (FDR) threshold of 0.05, estimated using Storey's q-value approach. Discrimination was quantified via the area under the curve (AUC) from receiver-operating characteristic (ROC) analyses, and statistical analyses were performed in R (version 2.9.1; Boston, MA, USA).
2.5. Statistical analyses
Continuous variables were checked for normality with the Shapiro–Wilk test and presented as mean ± SD or median (interquartile range) depending on distribution. Group comparisons employed Student's t-test or Mann–Whitney U test, and categorical variables were compared using Pearson's chi-square or Fisher's exact test. To identify baseline predictors of ΔHbA1c, we performed univariate logistic regression followed by multivariate modeling. Variables with a variance inflation factor (VIF) > 10, indicating significant collinearity, were excluded from multivariate analyses. All analyses were conducted using SPSS version 20.0 (SPSS Japan Inc., Tokyo, Japan), with a two-sided P-value <0.05 considered statistically significant.
3. Results
3.1. Clinical characteristics of patients with T2DM at baseline and after treatment with exenatide
This study included 93 patients with T2DM treated with branded exenatide (Byetta®) for 16 weeks whose clinical characteristics before and after treatment have been published elsewhere [12].
Participants were stratified by glucose-lowering response: responders (ΔHbA1c ≤ −0.3%) and non-responders (ΔHbA1c > −0.3%). As shown in Table 1, body weight, body mass index, fasting blood glucose, postprandial 2 h blood glucose and HbA1c levels decreased significantly, whereas alanine aminotransferase increased slightly but significantly after treatment in responders (all P < 0.05). In non-responders, body weight, body mass index and high-density lipoprotein cholesterol (HDL-C) level decreased remarkably, whereas HbA1c, triglyceride and uric acid levels increased significantly after treatment (all P < 0.05).
Table 1.
Clinical characteristics before and after treatment with exenatide for 16 weeks in responders and non-responders.
| Parameters | Responders (n = 73) |
t/Z/χ2 value | P-valuea | Non-responders (n = 20) |
t/Z/χ2 value | P-valuea | P-valueb | ||
|---|---|---|---|---|---|---|---|---|---|
| Baseline | After treatment | Baseline | After treatment | ||||||
| Age, years | 48.8 ± 9.78 | 51.9 ± 8.23 | 0.161 | ||||||
| Sex−male | 46 (63.0) | 9 (45.0) | 0.147 | ||||||
| Duration of diabetes, years | 6.12 ± 4.89 | 5.15 ± 3.51 | 0.323 | ||||||
| Background therapies | 0.067 | ||||||||
| Metformin | 38 (52.1) | 8 (40.0) | |||||||
| Insulin secretagogues | 5 (6.8) | 5 (25.0) | |||||||
| Both | 30 (41.1) | 7 (35.0) | |||||||
| Body weight, kg | 80.6 ± 15.1 | 78.5 ± 14.3 | 4.082 | <0.001 | 74.6 ± 19.3 | 73.0 ± 19.6 | 2.355 | 0.029 | 0.214 |
| Body mass index, kg/m2 | 28.3 ± 4.08 | 27.6 ± 3.81 | 4.421 | <0.001 | 27.4 ± 4.31 | 26.8 ± 4.44 | 2.326 | 0.031 | 0.388 |
| Systolic blood pressure, mmHg | 124.8 ± 11.5 | 124.3 ± 13.0 | 0.333 | 0.740 | 123.6 ± 9.26 | 124.7 ± 9.70 | −0.454 | 0.655 | 0.627 |
| Diastolic blood pressure, mmHg | 77.3 ± 8.51 | 75.9 ± 7.57 | 1.474 | 0.145 | 79.3 ± 7.15 | 76.3 ± 6.33 | 1.702 | 0.105 | 0.318 |
| Fasting blood glucose, mmol/L | 9.43 ± 1.88 | 7.64 ± 1.76 | 7.598 | <0.001 | 9.53 ± 1.90 | 9.50 ± 2.27 | 0.048 | 0.962 | 0.849 |
| Postprandial 2 h blood glucose, mmol/L | 16.6 ± 3.48 | 12.8 ± 3.91 | 7.343 | <0.001 | 15.2 ± 3.89 | 15.4 ± 3.74 | −0.241 | 0.812 | 0.155 |
| HbA1c, % | 8.32 ± 0.92 | 6.78 ± 0.84 | 15.560 | <0.001 | 7.89 ± 0.58b | 8.17 ± 0.68 | −3.065 | 0.006 | 0.048 |
| HbA1c, mmol/mol | 67.5 ± 10.0 | 50.6 ± 9.20 | 15.560 | <0.001 | 62.7 ± 6.32b | 65.8 ± 7.42 | −3.065 | 0.006 | 0.048 |
| Total cholesterol, mmol/L | 4.95 ± 1.00 | 4.78 ± 0.88 | 1.265 | 0.210 | 5.31 ± 0.99 | 5.37 ± 0.97 | −0.375 | 0.712 | 0.162 |
| LDL-C, mmol/L | 3.05 ± 0.88 | 3.01 ± 0.79 | 0.057 | 0.955 | 3.30 ± 0.75 | 3.23 ± 0.75 | 0.351 | 0.729 | 0.218 |
| HDL-C, mmol/L | 1.22 ± 0.34 | 1.21 ± 0.34 | 0.261 | 0.795 | 1.49 ± 0.32b | 1.39 ± 0.30 | 3.536 | 0.002 | 0.002 |
| Triglycerides, mmol/L | 1.68 (1.29, 2.89) | 1.54 (1.23, 2.37) | −1.406 | 0.160 | 1.49 (0.90, 1.90) | 1.94 (1.22, 2.28) | −2.166 | 0.030 | 0.073 |
| Uric acid, μmol/L | 306.9 ± 76.3 | 322.2 ± 76.6 | −1.641 | 0.105 | 311.0 ± 113.4 | 309.1 ± 90.7 | 0.116 | 0.909 | 0.879 |
| Alanine aminotransferase, U/L | 24.0 (15.5, 42.7) | 24.7 (16.3, 38.8) | −2.229 | 0.026 | 22.7 (15.7, 35.0) | 25.7 (15.0, 47.8) | −2.372 | 0.018 | 0.369 |
| Aspartate aminotransferase, U/L | 25.0 (18.1, 32.5) | 24.1 (17.6, 29.0) | −1.600 | 0.110 | 24.4 (20.3, 32.0) | 25.0 (22.0, 38.5) | −0.523 | 0.601 | 0.826 |
| Alkaline phosphatase, U/L | 68.0 (57.0, 84.0) | 64.5 (54.6, 80.5) | −1.835 | 0.067 | 72.5 (55.3, 82.3) | 75.0 (63.3, 86.1) | −0.430 | 0.667 | 0.736 |
| Creatinine, μmol/L | 65.6 ± 15.6 | 66.4 ± 16.4 | −0.928 | 0.356 | 63.9 ± 17.3 | 63.1 ± 17.1 | 0.310 | 0.760 | 0.683 |
| Blood urea nitrogen, mmol/L | 4.74 ± 1.31 | 4.61 ± 1.26 | 0.817 | 0.417 | 5.26 ± 1.18 | 5.28 ± 1.21 | −0.075 | 0.941 | 0.096 |
Note: The group of responders represents patients with ΔHbA1c ≤ −0.3%, and the group of non-responders represents patients with ΔHbA1c > −0.3%. Data are presented as mean ± SD, number (%) or median (interquartile range), as appropriate. aP-value, versus baseline in the same group. bP-value, non-responders versus responders at baseline.
Abbreviations: HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol.
Notably, the baseline HbA1c level was significantly higher in responders than in non-responders (8.32 ± 0.92% versus 7.89 ± 0.58%, P = 0.013), while the baseline HDL-C level was significantly lower in responders than in non-responders (1.22 ± 0.34 mmol/L versus 1.49 ± 0.32 mmol/L, P = 0.002). None of the other baseline clinical characteristics, including age, sex, body weight, and renal and hepatic function, revealed a significant difference between the two groups (all P > 0.05).
3.2. Baseline clinical characteristics for predicting ΔHbA1c in patients with T2DM after exenatide treatment
To determine whether baseline clinical characteristics alone could separate responders from non-responders, univariate and multivariate logistic regression analyses and ROC analysis were utilized. As shown in Table 2, baseline HDL-C had a significant positive association with ΔHbA1c, while baseline HbA1c and total cholesterol displayed marginal negative association with ΔHbA1c (P = 0.053 and P = 0.055, respectively). Stepwise multivariate logistic analysis was further performed to investigate the independent effects associated with ΔHbA1c. Baseline HbA1c (OR = 0.447; 95% CI: 0.209, 0.952) and HDL-C (OR = 7.134; 95% CI: 1.371, 37.131) were associated with ΔHbA1c after exenatide treatment. The AUC values of baseline HbA1c, HDL-C and combination of these two parameters were 0.631 (0.513–0.749), 0.254 (0.143–0.365), and 0.771 (0.658–0.883), respectively (Supplementary Table 1 and Supplementary Fig. 1). The relatively low AUC values indicated that only baseline clinical characteristics might be not optimized for predicting HbA1c reduction in patients with T2DM after treatment with exenatide for 16 weeks.
Table 2.
Univariate and multivariate logistic regression analysis for the baseline clinical characteristics associated with ΔHbA1c value.
| Univariate |
Multivariate |
|||||
|---|---|---|---|---|---|---|
| OR | 95% CI | P-value | OR | 95% CI | P-value | |
| Age | 1.038 | 0.981, 1.098 | 0.198 | |||
| Sex | 0.480 | 0.177, 1.307 | 0.480 | |||
| Duration of diabetes | 0.951 | 0.844, 1.071 | 0.406 | |||
| Body weight | 0.975 | 0.942, 1.009 | 0.148 | |||
| Body mass index | 0.943 | 0.830, 1.071 | 0.365 | |||
| Systolic blood pressure | 0.990 | 0.946, 1.036 | 0.661 | |||
| Diastolic blood pressure | 1.029 | 0.968, 1.094 | 0.358 | |||
| Fasting blood glucose | 1.027 | 0.789, 1.335 | 0.845 | |||
| Postprandial 2 h blood glucose | 0.890 | 0.767, 1.033 | 0.124 | |||
| HbA1c | 0.522 | 0.270, 1.008 | 0.053 | 0.447 | 0.209, 0.952 | 0.037 |
| Total cholesterol | 1.418 | 0.867, 2.321 | 0.164 | |||
| LDL-C | 1.388 | 0.788, 2.445 | 0.256 | |||
| HDL-C | 7.968 | 1.942, 32.694 | 0.004 | 7.134 | 1.371, 37.131 | 0.020 |
| Triglycerides | 0.540 | 0.288, 1.013 | 0.055 | 0.686 | 0.324, 1.450 | 0.323 |
| Uric acid | 1.001 | 0.995, 1.006 | 0.846 | |||
| Alanine aminotransferase | 0.982 | 0.954, 1.011 | 0.219 | |||
| Aspartate aminotransferase | 0.999 | 0.958, 1.041 | 0.958 | |||
| Alkaline phosphatase | 1.002 | 0.977, 1.029 | 0.855 | |||
| Creatinine | 0.993 | 0.962, 1.025 | 0.659 | |||
| Blood urea nitrogen | 1.364 | 0.931, 1.999 | 0.112 | |||
Abbreviations: HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol.
3.3. Differences in baseline metabolite profiles between responders and non-responders
An OPLS-DA model built on baseline metabolomic data demonstrated clear separation between responders and non-responders (Supplementary Fig. 2). Subsequent S-plot analysis of the model revealed 53 metabolites with VIP value > 1 (Supplementary Table 2 and Supplementary Fig. 3). Further comparison analysis between the two groups revealed that 20 of these metabolites differed significantly (all P < 0.05, Supplementary Table 3). By using the dual criteria of a VIP >1 and P < 0.05, we identified 15 metabolites as potential predictive biomarkers. Among these metabolites, 12 were elevated and three were reduced in responders relative to non-responders (Table 3).
Table 3.
Metabolites present at different levels in responders versus non-responders at baseline.
| Number | Metabolite | P-value | VIP value | q value |
|---|---|---|---|---|
| Higher in responders | ||||
| 1 | Cholesterol sulfate | 0.021697 | 1.318433 | 0.318224 |
| 2 | 3-Hydroxyhexadecadienoylcarnitine | 0.017767 | 1.496698 | 0.291545 |
| 3 | LysoPC(18:2(9Z,12Z)) | 0.001750 | 1.524314 | 0.042703 |
| 4 | LysoPC(18:1(11Z)) | 0.000826 | 1.790424 | 0.102651 |
| 5 | PC(18:1(9Z)/20:4(5Z,8Z,11Z,14Z)) | 0.015911 | 1.588322 | 0.283925 |
| 6 | PC(18:3(6Z,9Z,12Z)/16:0) | 0.006768 | 1.691401 | 0.198517 |
| 7 | LysoPC(16:1(9Z)) | 0.008026 | 1.977284 | 0.201804 |
| 8 | SM(d18:0/18:1(11Z)) | 0.031872 | 1.811141 | 0.400673 |
| 9 | PC(14:0/22:5(4Z,7Z,10Z,13Z,16Z)) | 0.012049 | 1.484474 | 0.265084 |
| 10 | PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) | 0.000139 | 2.606379 | 0.024412 |
| 11 | SM(d18:1/18:0) | 0.034178 | 1.673699 | 0.400673 |
| 12 | LysoPC(22:5(7Z,10Z,13Z,16Z,19Z)) | 0.003582 | 2.942223 | 0.126094 |
| Lower in responders | ||||
| 1 | Pidolic acid | 0.015641 | 1.127047 | 0.291545 |
| 2 | Uracil | 0.041037 | 1.120202 | 0.409528 |
| 3 | Butenylcarnitine | 0.000085 | 1.050049 | 0.007531 |
Note: The group of responders represents patients with ΔHbA1c ≤ −0.3%, and the group of non-responders represents patients with ΔHbA1c > −0.3%. VIP value was obtained from OPLS-DA with a threshold of 1.0.
Abbreviation: VIP, variable importance in projection.
3.4. Combination of the baseline clinical characteristics and metabolite profiles for predicting ΔHbA1c in patients with T2DM after exenatide treatment
Differential metabolites were subjected to multivariate logistic regression analysis and ROC assessment across four models, and their quality was evaluated by calculating the sensitivity, specificity and AUC value (Table 4 and Fig. 1). Using backward stepwise elimination method, butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) were retained in Model 4. The sensitivity, specificity and AUC value were 72.7%, 85.7% and 0.818 respectively, indicative of a good predictive performance. Furthermore, we combined baseline clinical variables with these metabolites to construct a prediction model. In Model 5, the clinical variables HbA1c and HDL-C as well as the metabolites butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) were included. This model showed a significant improvement, with the sensitivity, specificity and AUC value of 78.1%, 90.0% and 0.895, respectively.
Table 4.
AUC parameters of clinical characteristics and metabolites for discriminating responders from non-responders at baseline.
| Model Groups | AUC (95% CI) | P-value | Sensitivity (%) | Specificity (%) |
|---|---|---|---|---|
| Model 1 | 0.708 (0.472, 0.944) | <0.001 | 70.4 | 75.0 |
| Model 2 | 0.695 (0.438, 0.951) | <0.001 | 83.3 | 57.1 |
| Model 3 | 0.714 (0.415, 1.000) | <0.001 | 86.7 | 71.4 |
| Model 4 | 0.818 (0.611, 1.000) | <0.001 | 72.7 | 85.7 |
| Model 5 | 0.895 (0.805, 0.984) | <0.001 | 78.1 | 90.0 |
Note: The group of responders represents patients with ΔHbA1c ≤ −0.3%, and the group of non-responders represents patients with ΔHbA1c > −0.3%.
Model 1 contains all metabolites in Table 3.
Model 2 contains metabolites with P < 0.05 and VIP >1.5 in Table 3.
Model 3 contains metabolites with P < 0.01 and VIP >1.5 in Table 3.
Model 4 contains butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)), as optimal model.
Model 5 contains clinical characteristics (HbA1c and HDL-C) and metabolites [butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z))].
Abbreviations: AUC, area under the curve; HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol.
Fig. 1.
AUC analysis of clinical characteristics and metabolites for discriminating responders from non-responders
Note: The group of responders represents patients with ΔHbA1c ≤ −0.3%, and the group of non-responders represents patients with ΔHbA1c > −0.3%.
Model 1 contains all metabolites in Table 3.
Model 2 contains metabolites with P < 0.05 and VIP >1.5 in Table 3.
Model 3 contains metabolites with P < 0.01 and VIP >1.5 in Table 3.
Model 4 contains butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)), as optimal model.
Model 5 contains clinical characteristics (HbA1c and HDL-C) and metabolites [butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z))].
Abbreviations: AUC, area under the curve; HbA1c, glycated hemoglobin; HDL-C, high-density lipoprotein cholesterol.
4. Discussion
This study showed that in patients with T2DM whose HbA1c was not adequately controlled with metformin and/or insulin secretagogues, baseline HbA1c and HDL-C levels were associated with the magnitude of HbA1c reduction after 16 weeks of exenatide therapy. Metabolite profiles at baseline were significantly different between patients who had greater and smaller reductions in HbA1c level after treatment with exenatide. Importantly, combination of the baseline HbA1c, HDL-C, butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) could serve as predictive characteristics for estimating the glucose-lowering efficacy of exenatide treatment.
Despite the availability of newer glucose-lowering agents, economic considerations remain relevant. In a cost-effectiveness evaluation involving U.S. population aged 25–64 years, using exenatide as second-line therapy for glycemic control (compared with glibenclamide) was associated with an additional lifetime cost of $23,849 per patient [17]. Moreover, although exenatide resulted in a reduction in the level of HbA1c, only 46% of participants achieved the glycemic control target [18]. Therefore, correctly identifying which patients will benefit from the GLP-1RA treatment can improve precision therapy and reduce the associated financial burden. Our previous study revealed that higher baseline fibroblast growth factor-21 level was associated with poor glycemic control in patients with T2DM following the exenatide treatment [19]. Another study revealed that a poor glucose-lowering response to GLP-1RAs was associated with several baseline clinical variables: longer diabetes duration, insulin combination therapy, lower fasting C-peptide, lower postmeal urine C-peptide/creatinine ratio, and positive islet autoantibodies [20]. This post hoc analysis combined baseline clinical features with metabolomic profiling to identify parameters capable of anticipating the glucose-lowering benefit induced by exenatide treatment.
Herein, two baseline clinical characteristics could distinguish the exenatide treatment response. The responders had higher HbA1c and lower HDL-C, and their change directions were associated with better glucose-lowering response. Similarly, another study revealed that higher baseline HbA1c was associated with better glucose-lowering efficacy of insulin glargine [21]. The association between a lower baseline HDL-C and a better exenatide treatment response may be linked to the involvement of HDL-C in insulin sensitivity. Low HDL-C is a hallmark of insulin resistance [22]. GLP-1RAs have been demonstrated to ameliorate insulin resistance through various mechanisms, including loss of weight and decrease of hepatic glucose output [23,24]. Therefore, patients with greater insulin resistance, as characterized by lower HDL-C, may experience a more pronounced improvement in glycemic control upon treatment with exenatide.
Metabolomics help biomarker identification and contribute to the development of precise therapy [25]. Our previous study showed metabolite profiles could anticipate the exenatide-related improvements in endothelial injury in patients with T2DM [12]. In the present study, the results from no-targeted metabolomic analysis revealed that butenylcarnitine, LysoPC(18:2(9Z,12Z)) and PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) had a ability to help predict the glucose-lowering efficacy of exenatide treatment. Previous studies reported that the level of butenylcarnitine (O-butenyl-carnitine) in cord blood was associated with maternal prepregnant obesity and its level in feces was increased in individuals with obesity following the fecal microbiota transplantation from lean donors, suggesting the relevance of this metabolite to metabolic regulation [26,27]. LysoPC(18:2(9Z,12Z)), an unsaturated lysophosphatidylcholine, is a potential biomarker for distinguishing maturity-onset diabetes of the young type 3 from type 1 diabetes [28]. The origin and function of PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) are still unknown. To our best knowledge, no in vitro studies have yet demonstrated a direct functional involvement of these specific metabolites in the action of GLP-1RAs. Therefore, our findings primarily position these metabolites as predictive biomarkers. Further mechanistic studies are needed to determine whether they play a causal role or simply reflect the underlying pathophysiological states that are more amenable to GLP-1RA therapy. In our study, lower level of butenylcarnitine, higher level of LysoPC(18:2(9Z,12Z)) and higher level of PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z)) at baseline indicated better glucose-lowering efficacy of exenatide treatment. Therefore, measuring the levels of these three biomarkers may aid in determining the appropriate treatment regimen prior to administering exenatide.
In any case, our study underscored the synergistic value of combining traditional clinical assessments with novel metabolomic profiling for substantial improvement in predictive accuracy when compared with use of clinical variables alone. This approach aligns with the growing trend in precision medicine, where multi-omics data integration has been shown to improve risk stratification and therapeutic decision-making beyond the use of classic biomarkers alone [29]. By providing a more comprehensive pathophysiological snapshot, the combined models could facilitate the identification of patients most likely to benefit from exenatide treatment, thereby optimizing the treatment outcomes and resource allocation in clinical settings.
Several limitations should be considered. First, the major limitation is the small sample size (only 93 participants included in this post hoc analysis) and the imbalanced distribution between responders and non-responders (an approximate ratio of 3:1). The studies should expand the sample size or implement a well-matched design in future. However, the responder/non-responder ratio is also imbalanced in clinical real-world scenarios. This means the predictive indicators in our study may be suitable for most patients who want to initiate the exenatide treatment. Second, only LC‒MS was utilized in this study. Adding gas chromatography–mass spectrometry would significantly increase the coverage of detectable metabolites, thereby providing a more comprehensive metabolomic profiling to elucidate the benefits of accurate drug usage in therapeutic decision-making. Third, the exclusion of juvenile (<20 years) and elderly (>70 years) patients limits the generalizability of our predictive model. Future studies are needed to validate or adapt this model in these specific age groups. In addition, the intervention period was only 16 weeks, and thus the ability of the baseline biomarkers to predict the long-term efficacy of exenatide treatment remained uncertain.
5. Conclusions
Combination of the baseline clinical characteristics (higher HbA1c and lower HDL-C) and metabolomic features [lower butenylcarnitine, higher LysoPC(18:2(9Z,12Z)) and higher PC(20:3(5Z,8Z,11Z)/20:3(5Z,8Z,11Z))] can effectively predict better glucose-lowering efficacy of exenatide treatment.
CRediT authorship contribution statement
Yunyi Le: Writing – review & editing, Writing – original draft, Formal analysis, Data curation. Jin Yang: Writing – review & editing, Writing – original draft, Investigation, Funding acquisition, Formal analysis, Data curation. Qi Wu: Visualization, Validation, Data curation. Fei Li: Visualization, Resources, Data curation. Wei Fu: Visualization, Validation, Resources. Wenhua Xiao: Methodology, Investigation, Formal analysis, Data curation. Haining Wang: Supervision, Resources, Methodology, Formal analysis. Tianpei Hong: Visualization, Validation, Supervision, Project administration, Conceptualization. Rui Wei: Writing – review & editing, Project administration, Funding acquisition, Conceptualization.
Disclosure of interest
The authors declare that they have no competing interests.
Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding authors upon reasonable request.
Funding
This research was supported by the National Key Research and Development Program of China (2024YFA1802902), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0506901), the National Natural Science Foundation of China (82270843, 82371588, 82470839, 82574470), the special fund of the National Clinical Key Specialty Construction Program (2023), and the Talent Project of Clinical Key Project of Peking University Third Hospital. The funders of this study played no role in either its design or conduct; the collection, management, analysis or interpretation of data; the preparation, review or approval of the manuscript; or the decision to submit the manuscript for publication.
Acknowledgments
The authors thank Lixin Guo from the Department of Endocrinology, Beijing Hospital; Quanmin Li from the Department of Endocrinology, PLA Rocket Force Characteristic Medical Center; Liyong Zhong from the Department of Endocrinology, Beijing Tiantan Hospital, Capital Medical University; Jinkui Yang from the Department of Endocrinology, Beijing Tongren Hospital, Capital Medical University; and Jing Yang from the Department of Endocrinology, The First Hospital of Shanxi Medical University; and Yongyi Gao from the Department of Endocrinology, People's Hospital of Hainan Province, for their help with patient enrollment and data collection.
Footnotes
Supplementary data to this article can be found online at https://doi.org/10.1016/j.metop.2026.100453.
Contributor Information
Tianpei Hong, Email: tpho66@bjmu.edu.cn.
Rui Wei, Email: weirui@bjmu.edu.cn.
Appendix A. Supplementary data
The following is/are the supplementary data to this article:
References
- 1.Duncan B.B., Magliano D.J., Boyko E.J. IDF diabetes atlas 11th edition 2025: global prevalence and projections for 2050. Nephrol Dial Transplant. 2026;41(1):7–9. doi: 10.1093/ndt/gfaf177. [DOI] [PubMed] [Google Scholar]
- 2.Stratton I.M., Adler A.I., Neil H.A., et al. Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. BMJ. 2000;321(7258):405–412. doi: 10.1136/bmj.321.7258.405. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Khunti K., Zaccardi F., Amod A., et al. Glycaemic control is still central in the hierarchy of priorities in type 2 diabetes management. Diabetologia. 2025;68(1):17–28. doi: 10.1007/s00125-024-06254-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Zhong V.W., Yu D., Zhao L., et al. Achievement of guideline-recommended targets in diabetes care in China : a nationwide cross-sectional study. Ann Intern Med. 2023;176(8):1037–1046. doi: 10.7326/M23-0442. [DOI] [PubMed] [Google Scholar]
- 5.Campbell J.E., Drucker D.J. Pharmacology, physiology, and mechanisms of incretin hormone action. Cell Metab. 2013;17(6):819–837. doi: 10.1016/j.cmet.2013.04.008. [DOI] [PubMed] [Google Scholar]
- 6.McGowan B., Ciudin A., Baker J.L., et al. A systematic review and meta-analysis of the efficacy and safety of pharmacological treatments for obesity in adults. Nat Med. 2025;31(10):3317–3329. doi: 10.1038/s41591-025-03978-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.American Diabetes Association 8. Pharmacologic approaches to glycemic treatment: standards of medical care in diabetes-2018. Diabetes Care. 2018;41(Suppl 1):S73–S85. doi: 10.2337/dc18-S008. [DOI] [PubMed] [Google Scholar]
- 8.Heo C.U., Choi C.I. Current progress in pharmacogenetics of second-line antidiabetic medications: towards precision medicine for type 2 diabetes. J Clin Med. 2019;8(3):393. doi: 10.3390/jcm8030393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wishart D.S. Emerging applications of metabolomics in drug discovery and precision medicine. Nat Rev Drug Discov. 2016;15(7):473–484. doi: 10.1038/nrd.2016.32. [DOI] [PubMed] [Google Scholar]
- 10.Everett J.R. Pharmacometabonomics in humans: a new tool for personalized medicine. Pharmacogenomics. 2015;16(7):737–754. doi: 10.2217/pgs.15.20. [DOI] [PubMed] [Google Scholar]
- 11.Li B., He X., Jia W., Li H. Novel applications of metabolomics in personalized medicine: a mini-review. Molecules. 2017;22(7):1173. doi: 10.3390/molecules22071173. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Yang J., Le Y., Wei T., et al. Non-targeted metabolomic analysis predicts the therapeutic effects of exenatide on endothelial injury in patients with type 2 diabetes. J Diabet Complicat. 2021;35(2) doi: 10.1016/j.jdiacomp.2020.107797. [DOI] [PubMed] [Google Scholar]
- 13.Yang J., Xiao W., Guo L., et al. Efficacy and safety of generic exenatide injection in Chinese patients with type 2 diabetes: a multicenter, randomized, controlled, non-inferiority trial. Acta Diabetol. 2020;57(8):991–1000. doi: 10.1007/s00592-020-01510-y. [DOI] [PubMed] [Google Scholar]
- 14.Hsieh S.H., Shih K.C., Chou C.W., Chu C.H. Evaluation of the efficacy and tolerability of miglitol in Chinese patients with type 2 diabetes mellitus inadequately controlled by diet and sulfonylureas. Acta Diabetol. 2011;48(1):71–77. doi: 10.1007/s00592-010-0220-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Yang W.S., Jeng C.Y., Wu T.J., et al. Synthetic peroxisome proliferator-activated receptor-gamma agonist, rosiglitazone, increases plasma levels of adiponectin in type 2 diabetic patients. Diabetes Care. 2002;25(2):376–380. doi: 10.2337/diacare.25.2.376. [DOI] [PubMed] [Google Scholar]
- 16.Zhu X.X., Pan C.Y., Li G.W., et al. Addition of rosiglitazone to existing sulfonylurea treatment in chinese patients with type 2 diabetes and exposure to hepatitis B or C. Diabetes Technol Therapeut. 2003;5(1):33–42. doi: 10.1089/152091503763816445. [DOI] [PubMed] [Google Scholar]
- 17.Sinha A., Rajan M., Hoerger T., Pogach L. Costs and consequences associated with newer medications for glycemic control in type 2 diabetes. Diabetes Care. 2010;33(4):695–700. doi: 10.2337/dc09-1488. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Esposito K., Mosca C., Brancario C., Chiodini P., Ceriello A., Giugliano D. GLP-1 receptor agonists and HBA1c target of <7% in type 2 diabetes: meta-analysis of randomized controlled trials. Curr Med Res Opin. 2011;27(8):1519–1528. doi: 10.1185/03007995.2011.590127. [DOI] [PubMed] [Google Scholar]
- 19.Yang K., Wang H., Wei R., et al. High baseline FGF21 levels are associated with poor glucose-lowering efficacy of exenatide in patients with type 2 diabetes. Acta Diabetol. 2021;58(5):595–602. doi: 10.1007/s00592-020-01660-z. [DOI] [PubMed] [Google Scholar]
- 20.Jones A.G., McDonald T.J., Shields B.M., et al. Markers of β-cell failure predict poor glycemic response to GLP-1 receptor agonist therapy in type 2 diabetes. Diabetes Care. 2016;39(2):250–257. doi: 10.2337/dc15-0258. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Yan X., Jiang S., Lou Y., Zhou Z. Comparable efficacy and safety between LY2963016 insulin glargine and insulin glargine (Lantus) in Chinese patients with type 1 diabetes: a phase III, randomized, controlled trial. Diabetes Obes Metabol. 2021;23(10):2226–2233. doi: 10.1111/dom.14463. [DOI] [PubMed] [Google Scholar]
- 22.He S., Wang S., Chen X., et al. Higher ratio of triglyceride to high-density lipoprotein cholesterol may predispose to diabetes mellitus: 15-year prospective study in a general population. Metabolism. 2012;61(1):30–36. doi: 10.1016/j.metabol.2011.05.007. [DOI] [PubMed] [Google Scholar]
- 23.Blundell J., Finlayson G., Axelsen M., et al. Effects of once-weekly semaglutide on appetite, energy intake, control of eating, food preference and body weight in subjects with obesity. Diabetes Obes Metabol. 2017;19(9):1242–1251. doi: 10.1111/dom.12932. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Meier J.J. GLP-1 receptor agonists for individualized treatment of type 2 diabetes mellitus. Nat Rev Endocrinol. 2012;8(12):728–742. doi: 10.1038/nrendo.2012.140. [DOI] [PubMed] [Google Scholar]
- 25.Leslie R.D., Ma R.C.W., Franks P.W., Nadeau K.J., Pearson E.R., Redondo M.J. Understanding diabetes heterogeneity: key steps towards precision medicine in diabetes. Lancet Diabetes Endocrinol. 2023;11(11):848–860. doi: 10.1016/S2213-8587(23)00159-6. [DOI] [PubMed] [Google Scholar]
- 26.Schlueter R.J., Al-Akwaa F.M., Benny P.A., et al. Prepregnant obesity of mothers in a multiethnic cohort is associated with cord blood metabolomic changes in offspring. J Proteome Res. 2020;19(4):1361–1374. doi: 10.1021/acs.jproteome.9b00319. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Ghorbani Y., Schwenger K.J.P., Sharma D., et al. Effect of faecal microbial transplant via colonoscopy in patients with severe obesity and insulin resistance: a randomized double-blind, placebo-controlled phase 2 trial. Diabetes Obes Metabol. 2023;25(2):479–490. doi: 10.1111/dom.14891. [DOI] [PubMed] [Google Scholar]
- 28.Liu J., Fu J., Xie Z., et al. Serum metabolomics identified metabolite biomarkers and distinguished maturity-onset diabetes of the young from type 1 diabetes in the Chinese population. Clin Chim Acta. 2023;539:250–258. doi: 10.1016/j.cca.2022.12.019. [DOI] [PubMed] [Google Scholar]
- 29.Xu Z., Zhou Y., Xie R., Ning Z. Metabolomics uncovers the diabetes metabolic network: from pathophysiological mechanisms to clinical applications. Front Endocrinol. 2025;16 doi: 10.3389/fendo.2025.1624878. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets used and/or analyzed during the current study are available from the corresponding authors upon reasonable request.