Influence of Fibroblast Growth Factor 21 on Delayed Glycemic Improvement Following Acute Exercise in Type 2 Diabetes Mellitus
Article information
Abstract
Background
Exercise positively influences glycemic control. Some individuals experience greater glycemic stability on the day after exercise, even without additional physical activity. However, the mechanisms underlying this delayed glycemic improvement remain unclear.
Methods
Seventy-one patients with type 2 diabetes mellitus were assigned to either a 60-minute exercise or resting group. Serum fibroblast growth factor 21 (FGF21) levels and untargeted metabolomic profiling were assessed at multiple time points before and after exercise. Interstitial glucose levels were monitored using continuous glucose monitoring system. FGF21 knockout mice and wildtype littermates on a high-fat diet, underwent a 3-week exercise intervention, and supplemented with recombinant mouse FGF21.
Results
Individuals exhibiting delayed glycemic improvement (responders) displayed a significantly stronger FGF21 response compared to non-responders. Baseline metabolites, including p-cresol sulfate and dimethylglycine, differed between responders and non-responders and were associated with the FGF21 response. Longitudinal time-series analyses revealed post-exercise differences in acylcarnitines, fatty acids, and complex lipids between responders and non-responders. Dynamic correlation and mediation analyses supported that FGF21 modulates delayed glycemic improvement via regulation of lipid metabolism. In vivo FGF21 knockout and rescue experiments demonstrated that FGF21 is necessary for these metabolic shifts and for the associated improvements in glucose tolerance and insulin sensitivity.
Conclusion
This study finds that the baseline metabolome is associated with the magnitude of the post-exercise FGF21 response, which influences delayed glycemic improvements through regulation of lipid metabolism pathways.
Highlights
• Heterogeneous post-exercise glycemic responses were observed in T2DM.
• Delayed glycemic improvement is associated with a greater FGF21 response.
• Glycemic responders show distinct shifts in lipid metabolism after exercise.
• FGF21 knockout eliminates the glycemic benefit of exercise, rescued by recombinant FGF21.
INTRODUCTION
Exercise elicits both acute and long-term changes in blood pressure, heart rate, and various hematological parameters [1]. Acute aerobic exercise increases whole-body oxygen consumption, with some studies reporting a sustained elevation in oxygen uptake (VO2) for up to 24 hours or longer post-exercise [2,3]. The advent of continuous glucose monitoring (CGM) has shown that even a single session of exercise can improve 24-hour glycemic control in individuals with obesity and type 2 diabetes mellitus (T2DM) [4,5]. However, the mechanisms underlying the delayed glycemic benefits of exercise remain poorly understood.
Fibroblast growth factor 21 (FGF21) has emerged as a promising therapeutic target for obesity and diabetes due to its rapid and potent effects on improving insulin sensitivity [6]. FGF21 has recently been identified as an exercise-responsive factor, with circulating levels rising after endurance exercise [7,8]. Nevertheless, evidence suggests that exercise-induced FGF21 secretion is attenuated in individuals with obesity and T2DM [9,10]. This raises the need to investigate whether the FGF21 response to exercise varies among individuals with T2DM and whether this variation influences post-exercise glycemic control.
Acute exercise triggers complex molecular responses and modifies biological processes, including cellular energy metabolism (e.g., glycolysis and the tricarboxylic acid [TCA] cycle), whole-body substrate metabolism (e.g., amino acid breakdown and fatty acid [FA] oxidation), oxidative stress, inflammation, and tissue repair [11,12]. Metabolomics has provided valuable insights into these metabolic profiles post-exercise, revealing that many of these processes are dampened or reversed in insulin-resistant individuals [11].
In this study, we conducted an acute exercise test in patients with T2DM, measuring serum FGF21 levels at multiple time points before and after exercise, along with untargeted metabolomic profiling. Given the observed variability in post-exercise glucose responses, we aimed to elucidate the mechanisms underlying the delayed effects of exercise on glycemic improvement in patients with T2DM.
METHODS
Study participants
This study was based on a clinical trial focused on exercise intervention for patients with T2DM (ChiCTR2100046148). Inclusion criteria included an age of 35 to 65 years, abdominal obesity, body mass index (BMI) ≤35 kg/m2, use of ≤three antidiabetic agents for at least 6 weeks, and no regular physical activity. Exclusion criteria included glycosylated hemoglobin (HbA1c) <6.5% or ≥9%, insulin usage, severe proliferative diabetic retinopathy or worse, macroalbuminuria, renal dysfunction, history of major adverse cardiovascular events, cerebrovascular diseases, musculoskeletal injuries preventing exercise, or pregnancy. At baseline, 71 participants were randomized in a 4:1 ratio to either an acute exercise group (n=56), which performed a 60-minute aerobic exercise session or a resting group (n=15), which remained inactive (Fig. 1A). The trial was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Shanghai Sixth People’s Hospital. All participants provided written informed consent.
Individual variability in delayed glycemic improvement following exercise. (A) The study design. The changes in 24-hour (B) coefficient of variation (CV), (C) standard deviation (SD), (D) mean sensor glucose (MSG), (E) time in range (TIR), and (F) time in tight range (TITR) before and after exercise exhibit individual heterogeneity. (G) The interstitial glucose level trends throughout post-exercise days in the response (n=30) and non-response groups (n=26). (H-L) The changes in 24-hour (H) CV, (I) SD, (J) MSG, (K) TIR, and (L) TITR between the response and non-response groups after adjusting for clinical variables and exercise intensity. Data are presented as mean±SD. Gray trace and area, non-response group; orange trace and area, response group. T2DM, type 2 diabetes mellitus; Pre_Day, pre-exercise day; EX_Day, exercise day; Post_Day, post-exercise day. aP<0.05 for significant differences between groups, bP<0.01, cP<0.001.
To validate our findings, we employed an independent cohort consisting of patients with T2DM (n=16), who underwent a single aerobic exercise intervention similar to the primary cohort. Serum FGF21 levels and untargeted metabolomic profiling were measured before and after exercise. For further information about participants in this cohort, please contact the corresponding author.
Detailed methods for the study design and clinical and laboratory measurements are outlined in the Supplementary Methods.
Animals
FGF21 knockout (KO) mice on a C57BL/6J background were generated as previously described [13]. All the mice were housed in a room at controlled temperature (23°C±2°C) with a 12-hour light-dark cycle and had free access to water and rodent diet. At 8 weeks of age, mice were switched to a high-fat, high-cholesterol (HFHC) diet (Research Diet, D12079B, containing 40% fat, 17% protein, and 43% carbohydrate [kcal%]) for 12 weeks to establish diet-induced obesity. All animal experiments were conducted with male mice and were approved by the Animal Ethics Committee of Shanghai Sixth People’s Hospital. Detailed methods for the animal studies are provided in the Supplementary Methods.
Statistical analysis
Clinical characteristics of participants were shown as mean± standard deviation (SD) or median (interquartile range) for continuous variables, and frequencies (%) for categorical variables. Normal distribution (Shapiro-Wilk test) and homoscedasticity (Levene’s test) were checked before any statistical treatment. Untargeted metabolomics data were standardized using logarithm transformation and auto-scaling. Mixed-effects linear regression models, adjusted for clinical variables and exercise intensity, were used to analyze post-exercise glucose-related indices, cytokines response, and metabolite changes. Bonferroni post hoc tests assessed differences between conditions at each time point. Fuzzy c-means clustering was performed using the R package ‘Mfuzz.’ Anthropometric measures, the area under the curve (AUC) for cytokines, and glucose or insulin tolerance test results were compared by unpaired t-tests or Welch’s t-tests. Pearson correlation coefficient evaluated the relationship between FGF21 and metabolites. Time-delayed analysis was used to dynamically investigate the correlation between FGF21 and metabolites at different time points in the time series. Mediation effects of metabolites on the relationship between FGF21 levels and participant grouping were investigated using the R package ‘mediation.’ All analyses were performed using MetaboAnalyst version 6.0 (https://www.metaboanalyst.ca/), GraphPad Prism 10 (GraphPad Software Inc., San Diego, CA, USA), and R software version 4.3.3 (R Foundation for Statistical Computing, Vienna, Austria).
RESULTS
Individual variability in delayed glycemic improvement following exercise
A total of 71 participants with T2DM were included in the analysis: 56 assigned to the acute exercise group and 15 to the resting group (Supplementary Table 1). After 60 minutes of aerobic exercise, the acute exercise group exhibited significant reductions in blood glucose, insulin, and C-peptide levels, whereas the resting group showed a gradual decrease (Supplementary Fig. 1A-C). No significant differences in 24-hour glycemic control were observed between the acute exercise and resting groups on the following day (Supplementary Fig. 1D-H). Interestingly, within the acute exercise group, substantial individual differences in 24-hour coefficient of variation (CV) and SD of glucose levels were observed after exercise (Fig. 1B-F). Approximately half of the participants demonstrated a reduction in 24-hour CV compared to pre-exercise levels, indicating improved glucose variability, while the others did not (Fig. 1B). Based on these findings, participants in the acute exercise group were classified into two categories: those exhibiting delayed glycemic improvement (response group, n=30) and those without (nonresponse group, n=26). Immediate changes in blood glucose, insulin, and C-peptide levels post-exercise were similar between both groups (Supplementary Table 2). However, on the following day, the response group exhibited lower amplitudes of glucose fluctuation than the non-response group (Fig. 1G). Specifically, the response group showed significant reductions in 24-hour CV (P<0.001) and SD (P<0.001), whereas the nonresponse group exhibited significant increases in both 24-hour CV and SD after acute exercise (Fig. 1H and I). Moreover, significant improvements in 24-hour mean sensor glucose (MSG), time in range, and time in tight range were observed exclusively in the response group on the following day (Fig. 1J-L).
Baseline metabolic profiles associate with delayed postexercise glycemic improvement
No significant differences were observed between the response and non-response groups in age, gender, BMI, HbA1c levels, as well as VO2max and maximum heart rate during the cardiopulmonary exercise testing, or actual heart rate during acute exercise (Table 1, Supplementary Fig. 2). Both groups maintained consistent dietary patterns over the 3-day period (Supplementary Table 3). Interestingly, we identified 20 metabolites that differed significantly at baseline between the response and non-response groups (Supplementary Table 4). These include gut microbiota-derived metabolites, such as phenol sulfate, pcresol sulfate, and indoxyl sulfate; dimethylglycine; N-oleoyltaurine; 2-hydroxyarachidate; propionylcarnitine; 6-bromotryptophan, hexanoylglutamine, and phenylacetylglutamine; as well as the bilirubin metabolite L-urobilin. These results suggest that baseline metabolic differences may be associated with delayed glycemic improvement following exercise.
Summary of clinical characteristics of participants in both response and non-response groups
Association between FGF21 and delayed glycemic improvement
FGF21 is a known exercise-responsive factor. In this study, we found that serum FGF21 levels significantly increased after exercise, peaking at 30 minutes post-exercise before gradually declining (Supplementary Fig. 1I). Notably, the post-exercise FGF21 response in the response group was significantly higher than those in the non-response group (P<0.001) (Fig. 2A). Similarly, free fatty acids (FFAs) levels also increased significantly following acute exercise (Supplementary Fig. 1J), with the response group demonstrating a significantly higher AUC for FFA elevation compared to the non-response group (Fig. 2B). These findings suggest that post-exercise elevations in serum FGF21 and FFA are closely associated with delayed glycemic improvement in individuals with T2DM. To validate this, we included an external cohort of patients with T2DM who underwent aerobic exercise intervention and had serum FGF21 levels measured before and after exercise. Results were consistent with our primary cohort, with some participants showing delayed glycemic improvement one day post-exercise, and these participants exhibited a more pronounced FGF21 response post-exercise (Supplementary Fig. 3).
Post-exercise serum fibroblast growth factor 21 (FGF21), free fatty acid (FFA), and fibroblast activation protein (FAP) responses in both the response (R) and non-response (NR) groups. Time-course changes and the area under the curve (AUC) for the fold-change of (A) serum FGF21, (B) FFA, and (C) FAP levels following exercise are presented for the R and NR groups. Data are presented as mean±standard error of the mean. Analyses were conducted after adjusting for clinical variables and exercise intensity. Blue points and lines, NR group; red points and lines, R group; shaded areas indicate exercise (EX) periods. (D) The correlation of baseline serum FAP levels with baseline serum FGF21 levels. (E) The correlation of baseline serum FAP levels with the AUC for the fold-change of serum FGF21 response post-exercise. aP<0.05 for significant differences between groups, bP<0.01, cP<0.001.
Fibroblast activation protein (FAP) has been reported to cleave FGF21 and potentially impair it signaling pathways. In this study, we observed no significant change in serum FAP levels before and after exercise (Fig. 2C), suggesting that acute aerobic exercise does not immediately alter circulating FAP. However, baseline serum FAP levels were inversely correlated with baseline FGF21 (P=0.038) (Fig. 2D) and with the AUC for post-exercise FGF21 response (P=0.009) (Fig. 2E), suggesting that inter-individual variability in the post-exercise FGF21 response may be related to circulating FAP.
Specific baseline metabolites associated with post-exercise FGF21 response
To investigate potential associations between baseline differential abundant metabolites and the post-exercise FGF21 responses, we conducted additional analyses, correlating baseline metabolite levels with the AUC for post-exercise FGF21 response. We observed significant positive correlations between the AUC for post-exercise FGF21 responses and three baseline metabolites: p-cresol sulfate, a gut microbiota-derived metabolite (r=0.301, P=0.042), dimethylglycine, a methylation-related metabolite (r=0.338, P=0.022), and 6-bromotryptophan, an amino acid metabolism-related metabolite (r=0.373, P=0.011) (Supplementary Fig. 4). These findings suggest that specific baseline metabolic differences may contribute to delayed glycemic improvement by modulating the FGF21 response following acute exercise.
Post-exercise metabolomic profiles differed between responders and non-responders
In addition to baseline measurements, we constructed longitudinal metabolomics datasets at multiple time points before and after exercise. The datasets were used to evaluate differential metabolomic responses to acute exercise between the response and non-response groups and to explore associations among FGF21, metabolites, and delayed glycemic improvement. After data processing and annotation, the final datasets included 632 metabolites. Comparative analyses of the metabolomic profiles between the acute exercise group and resting group were shown in Supplementary Fig. 5.
Comparison of metabolomic responses to exercise revealed clear segregation between the response and non-response groups in the partial least square-discriminant analysis (PLSDA) score plot (Fig. 3A). A total of 216 differential metabolites were identified (Fig. 3B, Supplementary Table 5), approximately half of which were lipids and lipid-like molecules. Based on longitudinal trajectories, metabolites were categorized into six distinct temporal patterns (Fig. 3C). The AUC for metabolite clusters (patterns 1, 2, 3, and 5) differed significantly between response and non-response groups (Supplementary Table 6). Metabolites in pattern 1 exhibited rapid responses, peaking immediately post-exercise. These included glycolysis and TCA cycle-related metabolites; intermediates of amino acid metabolism like glutaric acid; as well as adenosine triphosphate-turnover-related xanthine. These metabolites accumulated to a greater extent in responders and correlated significantly with body-fat and lean mass percentages. Patterns 2 and 3 were enriched for metabolites involved in FA oxidation and complex lipid metabolism, including acylcarnitines (e.g., carnitine(2:0), carnitine(8:0)), phosphatidylcholines [PC] (e.g., PC(33:2), PC (34:1)), and FAs (e.g., FA(20:5), FA(20:3)). Metabolites in pattern 2 increased immediately post-exercise and remained elevated for up to 2 hours, and were significantly associated with exercise-induced changes in 24-hour SD and CV. Metabolites in pattern 3 showed a delayed increase and were closely associated with changes in the 24-hour glucose CV, waist circumference, and baseline insulin and C-peptide levels (Supplementary Table 7). Metabolites in patterns 4, 5, and 6 were down-regulated post-exercise, with a more pronounced decline observed in the response group. These included L-carnitine and certain lysophosphatidylcholines (LPC). These findings suggest that individuals in the response group experience a more pronounced activation of energy metabolism, particularly lipid metabolism, following acute exercise.
Untargeted metabolomics pipeline identified metabolic differences between responders and non-responders following acute exercise. (A) The score plot of partial least squares-discriminant analysis for participants in the response and non-response groups at seven time points. Blue points, non-response group; red points, response group. (B) Heatmap of the 216 metabolites with significant inter-group differences. Each metabolite is annotated according to the superclass to which it belongs. (C) Post-exercise response patterns of differentially abundant metabolites identified in responder and non-responder groups using Fuzzy cmeans clustering. The solid lines represent the mean and the dashed lines represent the 95% confidence interval. Blue trace and area, non-response group; red trace and area, response group. ‘n’ indicates the number of metabolites categorized within each pattern. (D) Pearson correlation coefficients of 126 differential metabolites correlated with fibroblast growth factor 21 (FGF21).
Post-exercise FGF21 response influences delayed glycemic improvement through lipid metabolism
To explore the relationship between specific metabolite profiles and FGF21 responses following acute exercise, correlation analyses were conducted. We observed significant correlations between FGF21 and 125 of the 216 differential metabolites, with over 70% showing positive correlations (Fig. 3D). By calculating the correlation between FGF21 and each metabolite at various temporal lag points, we identified 79 metabolites with positive lag coefficients associated with FGF21 (Supplementary Table 8), suggesting that the FGF21 response acts as an upstream event triggering fluctuations in these metabolites.
Mediation analysis identified 19 metabolites as mediators in the association between FGF21 and participant grouping (response vs. non-response groups) (Supplementary Table 9). Key mediators included L-acetylcarnitine with a mediation proportion of 12.56% (P=0.01) (Fig. 4A) and L-carnitine at 11.58% (P=0.016) (Fig. 4B), both associated with FA oxidation. Additionally, these included FA(24:6) with a mediation proportion of 12.54% (P=0.01) (Fig. 4C), and FA(20:5) at 8.46% (P=0.026) (Fig. 4D), which are linked to fat mobilization and lipolysis. Complex lipids, such as LPC(15:0) at 9.29% (P=0.024) (Fig. 4E) and LPC(18:0) at 16.50% (P=0.022) (Fig. 4F), also served as mediators. These metabolites were more activated in the response group, with L-acetylcarnitine, FAs, and PCs showing a sustained increase following exercise. This indicates that FGF21 enhances delayed glycemic improvement through pathways involving fat mobilization, lipolysis, and FA oxidation. In addition to lipid metabolism, several intermediates of the TCA cycle and amino acid metabolism also mediated the relationship between FGF21 and participant grouping. These included aminoadipic acid (mediation proportion: 15.36%, P<0.001), threonine (16.89%, P<0.001), and tyrosine (16.71%, P=0.036). These findings suggest that FGF21 influences post-exercise glucose homeostasis through multiple pathways, as illustrated in Fig. 4G. Analysis of metabolite time-course changes revealed distinct temporal patterns in responders. Glucose metabolism-related pathways were activated shortly after acute exercise, returning to baseline within 60 minutes. In contrast, lipid-metabolism responses were activated later, following the FGF21 response, and continued to rise for up to 2 hours post-exercise. Additionally, we also performed untargeted metabolomic profiling in the independent validation cohort (Supplementary Fig. 6). The findings mirrored those of the primary cohort, providing additional evidence that the post-exercise FGF21 response primarily modulates delayed glycemic improvement by regulating lipid metabolism.
Fibroblast growth factor 21 (FGF21)-driven metabolomic mediators and post-exercise pathway divergence between responders and non-responders. (A) L-acetylcarnitine, (B) L-carnitine, (C) fatty acid (FA)(24:6), (D) FA(20:5), (E) lysophosphatidylcholine (LPC)(15:0), and (F) LPC(18:0) significantly mediate the relationship between serum FGF21 and participant grouping. The estimates are given as standardized coefficients, with P values and 95% confidence intervals (CIs). (G) Schematic representations of selected metabolites are provided: amino acid metabolism (orange), glucose metabolism (yellow), FA oxidation and mitochondrial energy production (blue), and complex lipid metabolism (green). The individual metabolite responses are shown for each of the colored pathways above. Data are presented as mean±standard error of the mean. Blue point and line, non-response group (NR); red point and line, response group (R); shaded areas indicate exercise (EX) periods. ACME, average causal mediation effect; ADE, average direct effect; CoA, coenzyme A; TCA, tricarboxylic acid; FFA, free fatty acids; TG, triglycerides; SM, sphingomyelin; DG, diacylglycerol; PC, phosphatidylcholine; LPS, lysophosphatidylserine; PA, phosphatidic acid; PE, phosphatidylethanolamine; LPE, lysophosphatidylethanolamine; LPA, lysophosphatidic acid; PS, phosphatidylserine; LPC, lysophosphatidylcholine. aP<0.05, bP<0.01, cP<0.001 indicate significant differences between the response and non-response groups, assessed using the mixed model with Bonferroni post hoc test for multiple comparisons.
FGF21 KO and rescue experiments establish its necessity for exercise-induced metabolic and glycemic benefits
To further validate the necessity and mechanistic role of FGF21 in mediating exercise-induced metabolite changes and glycemic regulation, we compared FGF21 KO mice with their wild-type (WT) littermates. Eight-week-old male mice of each genotype were fed a HFHC diet for 12 weeks; thereafter, one cohort underwent 3 weeks of treadmill training while sedentary controls remained untrained. In parallel, a subset of FGF21 KO mice received subcutaneous osmotic pumps delivering recombinant mouse FGF21 (rmFGF21) throughout the training period to restore circulating FGF21 levels (Fig. 5A). After training, WT mice exhibited significant improvements in glucose tolerance and insulin sensitivity compared with sedentary controls (Fig. 5B and E), whereas FGF21 KO mice showed markedly attenuated responses (Fig. 5C and F). Remarkably, rmFGF21 supplementation rescued these glycemic benefits in FGF21 KO mice, indicating that FGF21 signaling is required for the exercise-induced improvements in glycemic control (Fig. 5D and G).
Fibroblast growth factor 21 (FGF21) is required for exercise-induced improvements in glucose tolerance and metabolomics response in obese mice. (A) Experimental workflow. (B) Glucose tolerance tests (GTT) in wild-type sedentary (WT-SE, n=6) and exercised (WT-EX, n=6) mice. (C) GTT in FGF21 knockout sedentary (KO-SE, n=6), exercised (KO-EX, n=6), and exercised with recombinant mouse FGF21 (rmFGF21) supplemented (KO-rmFGF21-EX, n=6) mice. (D) Area under the glucose curve (AUC) quantification corresponding to panels (B) and (C). (E) Insulin tolerance tests (ITT) in WT-SE and WT-EX mice. (F) ITT in KO-SE, KO-EX, and KO-rmFGF21-EX mice. (G) AUC quantification corresponding to panels (E, F). Data are presented as mean±standard error of the mean. (H) The score plot of partial least squares-discriminant analysis of serum metabolomic profiles from five groups (n=9). Ellipses represent 95% confidence intervals. (I) Volcano plots of differential metabolites between WT-EX and WT-SE. (J) Volcano plots of differential metabolites between KO-EX and KO-SE. (K) Volcano plots of differential metabolites between KO-rmFGF21-EX and KO-EX. (L) Volcano plots of differential metabolites between KO-EX and WTEX. Green points indicate metabolites mapped from key human metabolites. (M) Heatmap of metabolites that are associated with exercise and exhibit FGF21-related patterns. HFHC, high-fat, high-cholesterol; FDR, false discovery rate; FC, fold change; LPC, lysophosphatidylcholine; NS, not significant; PC, phosphatidylcholine; FA, fatty acid. aP<0.05, bP<0.01.
We then performed untargeted metabolomic profiling of post-intervention serum samples. PLS-DA analysis revealed clear separation among the five experimental groups (Fig. 5H). Exercise significantly altered 46 metabolites in WT mice, whereas only four changed in exercised FGF21 KO mice (Fig. 5I and J), underscoring FGF21’s critical role in exercise-induced metabolic remodeling. Under exercise conditions, the KO metabolome deviated from WT, and rmFGF21 largely reversed this divergence (Fig. 5K and L). Pathway enrichment analysis in exercised WT mice identified significant enrichment of pantothenate and coenzyme A biosynthesis, glutathione metabolism, ether lipid metabolism, and thiamine metabolism. These responses were attenuated or altered in KO mice but were restored towards WT levels by rmFGF21 treatment (Supplementary Fig. 7). Through group comparisons, we identified 23 exercise-related metabolites regulated by FGF21. Heatmap visualization clearly reveals coherent patterns across the five groups (Fig. 5M). To link mouse and human data, we mapped 19 key human metabolites onto the mouse dataset; seven showed significant group-wise differences (Supplementary Table 10). Notably, two metabolites highlighted in the human analysis, L-acetylcarnitine and LPC(18:0), decreased with exercise in WT, showed no significant change in KO, and were rescued by rmFGF21 (Supplementary Fig. 8), exemplifying an FGF21-dependent lipid-carnitine signature. Collectively, these results demonstrate that FGF21 is essential for exercise-induced remodeling of energy and lipid metabolism, and rmFGF21 restores these effects toward WT levels.
DISCUSSION
In this study, only a subset of patients with T2DM exhibited a reduction in 24-hour glucose CV the day after a 60-minute aerobic exercise session, while others did not. Untargeted metabolomics identified 20 baseline metabolites that significantly differed between the two groups. Responders displayed a significantly greater post-exercise increase in serum FGF21 than non-responders. Three baseline metabolites, p-cresol sulfate, dimethylglycine, and 6-bromotryptophan, were strongly associated with the magnitude of FGF21 response. Longitudinal metabolomic profiling revealed that metabolic divergence between responders and non-responders was driven chiefly by lipid metabolism. Mediation analysis implicated L-acetylcarnitine, L-carnitine, several FAs, and complex lipids as intermediaries linking FGF21 to participant classification. In vivo experiments supported that FGF21 is indispensable for the activation of energy and lipid metabolic pathways induced by exercise. Genetic deletion of FGF21 abolished exercise-induced metabolic reprogramming, whereas administration of rmFGF21 reinstated these effects. These findings indicate a mechanistic link between baseline metabolites, the post-exercise FGF21 surge, downstream metabolic responses, and subsequent glycemic outcomes.
Several studies show that the glucose-lowering effect of exercise is highly heterogeneous [14,15]. This variability is modulated by baseline glycemic status (e.g., higher HbA1c), cardiorespiratory fitness (e.g., VO2max), exercise timing, and molecular features including gut microbiota composition and the plasma proteome [16-18]. Notably, elevated day-to-day glucose variability-reflected by higher 24-hour CV and SD predicts more frequent hypoglycemia and increased vascular stiffness, even when mean glucose levels are comparable [19,20]. Clarifying the determinants of this variability is therefore of considerable clinical importance.
In our cohort, exercise exerted a greater effect on 24-hour CV and SD than on MSG, indicating a delayed improvement in glycemic stability. Correlation and mediation analyses attributed this benefit to FGF21-driven lipid reprogramming. In responders, the post-exercise surge in FGF21 activated the carnitine shuttle (e.g., L-acetylcarnitine, L-carnitine) and stimulated FA oxidation (e.g., FA(20:5), FA(24:6)). Consistent with its known actions, FGF21 enhances hepatic FA oxidation, suppresses lipogenesis and gluconeogenesis, and improves dyslipidemia and hyperglycemia by promoting adipose lipolysis, mitochondrial activity, and glucose uptake [21-24]. Mechanistically, FGF21 engages several key signaling nodes, including adenosine monophosphate-activated protein kinase-sirtuin 1 (AMPK-Sirt1) [25], peroxisome proliferator-activated receptor (PPAR) alpha [26,27] and gamma [28], Jumonji domain-containing protein 3 [29], the mechanistic target of rapamycin complex 1 (mTORC1) [30], and the cyclic adenosine monophosphate response element-binding protein (CREB)/CREB-regulated transcription coactivator 2 [31]. By accelerating mitochondrial lipid utilization, this program limits the accumulation of lipotoxic intermediates and enhances hepatic and skeletal-muscle insulin sensitivity [32,33]. Enhanced peripheral insulin sensitivity may therefore explain post-exercise glucose stability, with improvements persisting for up to 24 to 48 hours [34,35]. Collectively, these adaptations probably underpin the improved glucose variability observed on the following day. Our murine experiments corroborated this mechanism: FGF21 KO abolishes the lipid-metabolism response to exercise, whereas rmFGF21 reinstated it and restored glucose tolerance. Taken together, these data position FGF21 as a central regulator of post-exercise metabolic adaptation, orchestrating a lipid-centered network that safeguards subsequent glycemic control.
Several baseline metabolites were strongly correlated with the post-exercise increase in FGF21, suggesting that they may reflect the gut microbiota or hepatic metabolic status which, via the gut-liver axis, modulates hepatic FGF21 expression. For instance, dimethylglycine is an intermediate in betaine catabolism [36]. Dietary betaine supplementation elevates hepatic and circulating FGF21, improves glucose homeostasis, reduces hepatic lipid accumulation, and enhances white adipose tissue oxidation in mice [37]. Similarly, the gut-derived metabolite pcresol sulfate increases when the intestinal barrier is compromised or microbial metabolic pathways are altered, and its accumulation may potentiate the hepatic response to exercise [38]. FAP, a member of the dipeptidyl peptidase 4 (DPP4) gene family, cleaves both the N- and C-termini of FGF21, thereby attenuating its signaling pathways [39]. DPP4 has recently been identified as a microbial-host isoenzyme; gut microbial DPP4 disrupts glucose homeostasis by reducing active glucagon-like peptide 1 levels in high-fat diet mouse models [40]. Collectively, these observations implicate specific microbial and hepatic metabolites as potential upstream modulators of the FGF21 response to exercise.
Metabolic responses after exercise exhibit a temporal cascade: metabolites in pattern 1 (glycolytic and TCA cycle-related metabolites) demonstrate a rapid response (0 to 60 minutes), which may improve acute glycemic control; lipid metabolites in patterns 2 and 3 increase with a delay (60 to 120 minutes), thereby facilitating glycemic control the following day. It is well-established that the body prioritizes glycogen as an energy source during aerobic exercise, while lipid metabolism also serves as a significant energy source; moreover, the alterations in lipid-metabolism post-exercise often persist for an extended period [41,42]. The temporal cascade identified in our study reveals a potential action pattern of FGF21; its secretion occurs later than the immediate glucose metabolic response following exercise, but promotes the FA oxidation and activates the carnitine shuttle system, thereby shifting the principal energy substrates from carbohydrates to lipids during the postexercise recovery phase, ultimately enhancing insulin sensitivity in target tissues [43,44]. Therefore, from a clinical perspective, regulating specific microbial and hepatic metabolites, as well as the bioavailability of FGF21, may optimize the benefits of exercise on glycemic control.
The limitations should also be acknowledged. First, the study examined only a single session of aerobic exercise, the behavior of FGF21 during long-term exercise intervention or alternative exercise modalities remains unexplored. Second, although we identified several baseline metabolites, particularly of microbial origin, that were associated with delayed glycemic improvement, causality has not been established. Additional research is required to define the relationship between gut microbiota composition and post-exercise glycemic control. Third, several other candidate metabolites that may mediate the link between the FGF21 response and delayed glycemic improvement were identified only statistically. These associations require further validation.
SUPPLEMENTARY MATERIALS
Supplementary materials related to this article can be found online at https://doi.org/10.4093/dmj.2024.0814.
Characteristics of participants in both the resting and acute exercise groups
Variations in glucose metabolism indicators of participants in both the response and non-response groups
Dietary intake of participants in the response and non-response groups before and after the intervention over a period of 3 days
Metabolites showing significant differences between the response and non-response groups at baseline
The results of one-way ANOVA and mixed model between the response and non-response groups
Comparison of area under the curve for metabolite clusters between the response and non-response groups
Correlations between the area under the curve of each metabolite cluster and clinical characteristics
Time-delayed correlation analysis of differential metabolites with FGF21
Metabolites serve as mediators in the association between FGF21 with grouping (response or non-response groups)
Group-wise contrasts for human key metabolites mapped to the mouse model
Time-course changes in metabolic parameters following acute exercise. Time-course changes in (A) glucose, (B) insulin, and (C) C-peptide levels in the resting group (RE; n=15) and the acute exercise groups (AE; n=56) are depicted. (D-H) present the changes in 24-hour mean sensor glucose (MSG), standard deviation (SD), coefficient of variation (CV), time in range (TIR), and time in tight range (TITR) in the RE and AE groups. Time-course changes and the area under the curve (AUC) for the fold-change of (I) serum fibroblast growth factor 21 (FGF21) and (J) free fatty acids (FFA) post-exercise in the RE and AE groups. Data are presented as mean±standard error of the mean. Analyses were conducted after adjusting for clinical variables and exercise intensity. White circles represent the RE group, while black squares represent the AE group. Pre_Day, pre-exercise day; Post_Day, post-exercise day; EX, exercise. aP<0.05 for significant differences between groups, bP<0.01, cP<0.001.
Heart rates of participants in both the response (R) and non-response (NR) groups during the intervention. (A) Changes in heart rate before and after exercise for participants in the R and NR groups. (B) Mean heart rate and (C) maximum heart rate (HRmax) between the R and NR groups. Data are presented as mean±standard error of the mean. Statistical comparisons were performed using unpaired t-tests. Red traces and points represent the R group, while blue traces and points represent the NR group.
External cohort also found heterogeneity in the delayed glycemic improvement after exercise in patients with type 2 diabetes mellitus, and this heterogeneity is associated with the post-exercise response of fibroblast growth factor 21 (FGF21). (A-E) illustrate the changes in 24-hour mean sensor glucose (MSG), standard deviation (SD), coefficient of variation (CV), time in range (TIR), and time in tight range (TITR) between the response group (R; n=6) and the non-response group (NR; n=10) after adjustments for clinical variables and exercise intensity; (F, G) show the time-course changes and the area under the curve (AUC) for the fold-change of serum FGF21 and free fatty acids (FFA) post-exercise in the R and NR groups. Data are presented as mean±standard error of the mean. Pre_Day, pre-exercise day; Post_Day, post-exercise day; EX, exercise. aP<0.05 for significant differences between groups, bP<0.01.
Relationship between baseline differential metabolites and the post-exercise response of serum fibroblast growth factor 21 (FGF21). Correlations of (A) p-cresol sulfate, (B) dimethylglycine, and (C) 6-bromotryptophan with the area under the curve (AUC) for the fold-change in serum FGF21 are presented.
Changes in circulating metabolites following acute exercise in patients with type 2 diabetes mellitus. (A) The score plot of partial least squares-discriminant analysis across seven time points. (B) Top 50 of the important time-differential metabolites ranked by variable importance projection (VIP) scores from the partial least-squares discriminant analysis model. (C) Number and direction of metabolites with significant changes over intervention time in the acute exercise group and resting group (false discovery rate <0.05). (D) Clustering of longitudinal trajectories based on circulating metabolites. Post 2 min, 2-minute immediately after acute exercise. 8-ETE, 8-hydroxyeicosatetraenoic acid; FA, fatty acid; LPC, lysophosphatidylcholine.
Multi-timepoint untargeted metabolomics analysis of serum in the independent validation cohort before and after exercise. (A) The score plot of partial least squares-discriminant analysis for the response group (red) and non-response group (blue) at baseline and at 2, 15, 30, 60, 90, and 120 minutes after exercise. (B) Heatmap illustrating the dynamic changes of key differential metabolites between the two groups at each time point; each row represents a metabolite and the color indicates the normalized score. (C) Pathway enrichment analysis of key differential metabolites. Dot colors indicate P values and size represents enrichment ratio. (D-G) Mediation analysis demonstrates the role of specific metabolites in mediating the effect of fibroblast growth factor 21 (FGF21) on post-exercise glycemic improvement. Examples shown are (D) L-acetylcarnitine, (E) fatty acid (FA) (24:5), (F) FA(20:5), and (G) lysophosphatidylcholine (LPC)(15:0). Statistical results for each pathway—including average causal mediation effect (ACME), average direct effect (ADE), 95% confidence interval (CI), and P value—are marked next to the arrows. LPE, lysophosphatidylethanolamine; PE, phosphatidylethanolamine; CoA, coenzyme A; TCA, tricarboxylic acid; NR, non-response group; R, response group.
Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway over-representation analysis for key groupwise contrasts. (A) wild-type exercised (WT-EX) vs. wild-type sedentary (WT-SE), (B) knockout exercised (KO-EX) vs. knockout sedentary (KO-SE), (C) KO-EX vs. WT-EX, (D) exercised with recombinant mouse FGF21 (rmFGF21) supplemented (KO-rmFGF21-EX) vs. KO-EX. Dot colors indicate P values and size represents enrichment ratio. CoA, coenzyme A; TCA, tricarboxylic acid.
Comparisons of normalized metabolite intensities across mouse groups for two key cross-species-consistent metabolites. (A) L-acetylcarnitine and (B) lysophosphatidylcholine (LPC)(18:0). Boxes depict the median and interquartile range; points represent individual mice (n=9). WT-SE, wild-type sedentary; WT-EX, wild-type exercised; KO-SE, knockout sedentary; KO-EX, knockout exercised; KO-rmFGF21-EX, knockout exercised with recombinant mouse FGF21 (rmFGF21) supplemented. aFalse discovery rate (FDR) <0.05 for significant differences between groups, bFDR <0.01, cFDR <0.001.
Notes
CONFLICTS OF INTEREST
No potential conflict of interest relevant to this article was reported.
AUTHOR CONTRIBUTIONS
Conception or design: X.L., Q.F., Y.W., G.X., H.L.
Acquisition, analysis or interpretation of data: Y.Z., D.L., Y.L. (Yurun Lu), P.K., X.L., Q.W., A.C., D.C., L.W., Q.L., X.W., Y.L. (Yanli Li), Y.Y., J.Y., J.N.
Drafting the work or revising: Y.Z., D.L., X.L., Q.F., Z.H., A.X., W.J., Y.W., G.X., H.L.
Final approval of the manuscript: all authors.
FUNDING
This study was supported by the National Key Research and Development Program of China (2022YFA1004804) to Huating Li and Weiping Jia; the Excellent Young Scientists Fund of NSFC (82022012), General Fund of NSFC (82270907), Major Program of NSFC (92357305), Innovative Research Team of High-level Local Universities in Shanghai (SHSMU-ZDCX202 12700) to Huating Li; the Shanghai Municipal Key Clinical Specialty (2017ZZ01013), Shanghai Research Center for Endocrine and Metabolic Diseases (2022ZZ01002) to Weiping Jia; the Youth Innovation Promotion Association of the Chinese Academy of Sciences (2021186) to Xinyu Liu; the National Natural Science Foundation of China (82100879), Shanghai Pujiang Program (2020PJD044), and Exploration Fund Grant of Shanghai Sixth People’s Hospital (ynts202003) to Liang Wu; the National Natural Science Foundation of China (82500987) to Dan Liu.
ACKNOWLEDGMENTS
We thank all participants and investigators for their contributions to this study. We appreciate Nanjing Kuanyue Health Technology Co., Ltd. for supporting the exercise equipment. The graphical abstract was created using BioRender.com.