Hyperglycaemia is associated with impaired muscle signalling and aerobic adaptation to exercise

Abstract

Increased aerobic exercise capacity, as a result of exercise training, has important health benefits. However, some individuals are resistant to improvements in exercise capacity, probably due to undetermined genetic and environmental factors. Here, we show that exercise-induced improvements in aerobic capacity are blunted and aerobic remodelling of skeletal muscle is impaired in several animal models associated with chronic hyperglycaemia. Our data point to chronic hyperglycaemia as a potential negative regulator of aerobic adaptation, in part, via glucose-mediated modifications of the extracellular matrix, impaired vascularization and aberrant mechanical signalling in muscle. We also observe low exercise capacity and enhanced c-Jun N-terminal kinase activation in response to exercise in humans with impaired glucose tolerance. Our work indicates that current shifts in dietary and metabolic health, associated with increasing incidence of hyperglycaemia, might impair muscular and organismal adaptations to exercise training, including aerobic capacity as one of its key health outcomes.

Increased aerobic exercise capacity, measured clinically as VO_2peak, is one of the key health-promoting adaptations induced by regular aerobic exercise training. High exercise capacity has emerged as one of the best predictors of health and longevity in animals and humans, and strongly protects against mortality^1–4. Conversely, low exercise capacity greatly increases risk for chronic disease and early death by up to eight-fold^5,6, independently of other risk factors such as obesity^7,8. The exercise capacity phenotype is equally attributed to heritable factors and environmental influence⁹, suggesting that this complex trait is highly modifiable. As a result, an individual’s ability to improve exercise capacity after aerobic training is substantially variable in humans and animals^9–11. The fact that an estimated 20% of the population^12–14 is resistant to increases in exercise capacity while others attain marked improvements has been the source of considerable interest and controversy in the scientific community^15–17. The mechanisms underlying this striking heterogeneity remain largely unknown.

We previously studied rat models generated by selective breeding for low (LRT) or high (HRT) response to aerobic training to elucidate the mechanisms that inhibit improved exercise capacity in some individuals¹⁸. Genetic selection for low improvements in exercise capacity coselected with impaired glucose tolerance in LRT, identifying an association between aberrant glucose metabolism and low exercise capacity. In line with this, clinical data demonstrate that populations with chronically elevated blood glucose levels (that is, hyperglycaemia) undergo blunted improvements in exercise capacity with aerobic training versus normoglycemic participants¹⁹. In addition, several measures of impaired glycaemic control, including low glucose tolerance, high fasting glucose and insulin resistance are robustly associated with low exercise capacity in humans^20–22. Thus, there is substantial evidence from humans and genetic models of exercise response that hyperglycaemia and low exercise capacity are somehow linked.

As rates of metabolic disease skyrocket globally, the incidence of hyperglycaemia is growing rapidly^23,24, which may lead to a population that is increasingly resistant to improved exercise capacity with training. A primary contributor to rising rates of hyperglycaemia is a recent shift towards consumption of a ‘Western’ style diet, which has become widespread since the industrial revolution^25–27. One defining characteristic of the Western diet (WD) is high consumption of processed carbohydrates and sugars that quickly increase blood glucose levels, thus directly contributing to hyperglycaemia²⁸. In addition, the WD is typically high in saturated fats that can cause insulin resistance, a condition that exacerbates hyperglycaemia by impairing glucose removal from the circulation^29,30. As such, chronic consumption of a WD is independently associated with hyperglycaemia and impaired glucose tolerance in humans and animal models^25,31.

Here, we determined how consumption of a WD, in addition to diet-independent induction of hyperglycaemia using streptozotocin (STZ), affects health-promoting adaptations in response to exercise training. We demonstrate that mice with chronic hyperglycaemia have blunted improvements in exercise capacity in response to aerobic training, as compared to mice with normoglycaemia. Furthermore, we find that the failure to improve exercise capacity in these models is independent of obesity or insulin levels, but instead is associated with hyperglycaemia-induced changes to the extracellular matrix (ECM) of skeletal muscle, and altered molecular signal transduction with exercise. We provide evidence that these fundamental changes in muscle structure and signalling impair skeletal muscle remodelling towards an aerobic phenotype, which may contribute to the mechanism for low improvements in exercise capacity with hyperglycaemia. We demonstrate that similar changes in exercise-induced molecular signalling occur in humans with impaired glucose metabolism. Taken together, our data identify important interactions between exercise and metabolic health that may influence one of the key health benefits of exercise training: improvements in exercise capacity.

Results

Animal models of hyperglycaemia.

Data from humans and animal models suggest that increases in exercise capacity with aerobic training may be blunted by hyperglycaemia^18,19. To determine whether induction of hyperglycaemia can impair adaptation to aerobic training, we performed exercise training studies using two distinct mouse models: (1) WD feeding and (2) STZ-induced hyperglycaemia. WD-fed mice consumed a diet higher in sucrose (29% total kcal) and saturated fat (38% kcal) compared to control diet-fed mice (CON). STZ-treated mice were injected with low doses of STZ (two consecutive daily injections of 40 mg kg⁻¹) to reduce insulin secretion. Eight weeks after treatment initiation, glucose tolerance was similarly impaired in WD and STZ compared to CON (Fig. 1a,b). Impaired glucose tolerance in WD and STZ was accompanied by an increase in blood glucose concentrations of ~50–70 mg dl⁻¹ during the light and dark cycles, confirming the induction of physiological levels of hyperglycaemia in both models (Fig. 1c).

Fig. 1 |. Glucose intolerance and hyperglycaemia develop in WD-fed and STZ-injected mice.

CD-1 mice were fed a WD, injected with STZ (2 × 40 mg kg⁻¹), or fed a control diet (CON). a,b, After 8 weeks of treatment, all mice underwent GTT (2 g kg⁻¹ glucose) (a) and AUC for glucose over 120 min was calculated (b) (CON n = 52, WD n = 51, STZ n = 52). c, Blood was sampled from unfasted mice under resting conditions during the light (7:00; CON n = 10, WD n = 12, STZ n = 15) and dark (20:00; CON n = 12, WD n = 12, STZ n = 15) cycles to assess the degree of hyperglycaemia. d, Serum insulin was measured by enzyme-linked immunosorbent assay after 12 h of fasting (CON n = 21, WD n = 11, STZ n = 21). e, Body mass was recorded for 8 weeks following initiation of CON (n = 14), WD (n = 12) or STZ (n = 12) treatment in a cohort of mice. f, To determine voluntary wheel running behaviour in CON, WD and STZ groups, running distance was measured daily and expressed as cumulative weekly running distance for 6 weeks (CON n = 25, WD n = 25, STZ n = 26). Two-tailed unpaired t-test with Welch’s correction was used to compare groups in c or by one-way ANOVA followed by Holm–Sidak post hoc testing in all other panels. Data are represented as a point for the result of each individual animal; bars are mean ± s.e.m.

WD mice had elevated fasting serum insulin levels compared to CON, indicating that insulin resistance contributes to hyperglycaemia in this model (Fig. 1d). In contrast, STZ mice had a 40% decrease in fasting insulin compared to CON, confirming reduced insulin secretion as a mechanism for hyperglycaemia (Fig. 1d). STZ is typically directed towards ablating insulin secretion and inducing overt diabetes; however, we chose to administer a relatively mild dose of STZ for a partial reduction in insulin secretion. This was done to match the levels of hyperglycaemia induced by WD feeding, and avoid the complications of untreated diabetes such as weight loss, muscle wasting and premature mortality³². In addition to divergent insulin phenotypes, body weight was 18% higher in WD versus CON mice following 8 weeks of feeding, while STZ body weights were similar to CON (Fig. 1e). Thus, despite similar levels of hyperglycaemia in WD and STZ, these models have vastly different insulin and body weight phenotypes. The use of these contrasting models was designed to distinguish the effects of hyperglycaemia from other variables, as only phenotypes observed in both models could potentially be attributed to hyperglycaemia.

Aerobic training by voluntary wheel running.

To determine whether hyperglycaemia affects wheel-running behaviour, we measured cumulative voluntary wheel-running distance in WD, STZ and CON mice. Running distance was similar among all treatment groups over a 6-week wheel-running period (Fig. 1f). Using a smaller cohort of mice, we did not detect significant differences in circadian running behaviour among our experimental groups (Extended Data Fig. 1), as has been observed in other studies^33,34. Thus, we identify wheel running as an appropriate training modality to determine differences in aerobic adaptation in our models.

Metabolic adaptations to training are preserved in models of hyperglycaemia.

A schematic diagram showing the timeline of our experimental design appears in Fig. 2a. Total training distance was not affected by hyperglycaemia, with all mice running ~500 km over a 6-week exercise training period (Fig. 2b). In line with this, skeletal muscle training markers such as OXPHOS mitochondrial subunits I–V, hexokinase II and GLUT4 protein levels were similarly increased by training in all groups (Fig. 2c; representative blots in Fig. 2d). Aerobic exercise training also improved whole-body metabolic control, with a partial restoration of glucose tolerance in WD and STZ towards the level of CON (Fig. 2e). In WD, training also caused a reduction in body weight (Fig. 2f), which was due to decreased fat mass (Fig. 2g). Exercise training decreased visceral adipocyte size by ~30% (Fig. 2h,i) and liver triglycerides were reduced by 35% in WD (Fig. 2j,k). These results demonstrate that the aerobic training stimulus was similar among treatment groups, and aerobic training produced a series of positive metabolic adaptations expected from exercise training.

Fig. 2 |. Metabolic outcomes in response to exercise training.

a, A schematic diagram in outlining the experimental timeline of the aerobic exercise training study. b, Total training distance was recorded from mice with access to wheels after 6 weeks (CON n = 25, WD n = 25, STZ n = 26). c, Skeletal muscle mitochondrial and training markers OXPHOS (complexes I–V), hexokinase II and GLUT4 were assessed by western blotting in gastrocnemius muscle (CON SED n = 12, WD SED n = 12, STZ SED n = 12; CON EXT n = 12, WD EXT n = 12, STZ EXT n = 10). d, Representative blots. e, In the final weeks of exercise training, mice underwent GTT and this shows AUC for glucose over 120 min (CON SED n = 27, WD SED n = 25, STZ SED n = 27; CON EXT n = 25, WD EXT n = 23, STZ EXT n = 25). f,g, Body mass was measured before and after training (f) (CON SED n = 27, WD SED n = 25, STZ SED n = 28; CON EXT n = 25, WD EXT n = 24, STZ EXT n = 26) and body composition was determined in a subset of animals by dual-energy X-ray absorptiometry (g) (CON SED n = 10, WD SED n = 9, STZ SED n = 6; CON EXT n = 10, WD EXT n = 9, STZ EXT n = 6). h, Visceral adipose tissue was collected and stained with H&E and representative images are shown (STZ EXT n = 5; all other groups n = 6). Scale bar, 100 µm. i, Adipocyte size was quantified and expressed as cross-sectional area (STZ EXT n = 5; all other groups n = 6). j, Liver was harvested at dissection, stained with H&E and imaged using bright field microscopy. Scale bar, 50 µm. Images are representative of n = 6 per group. k, Hepatic glycerol was measured by colorimetric assay (CON SED n = 6, WD SED n = 6, STZ SED n = 8; CON EXT n = 6, WD EXT n = 5, STZ EXT n = 9). In cases where a subset of mice was analysed for training effects, exercise-trained mice from each group were matched for training distance. Main effects were determined by one-way ANOVA in b or two-way ANOVA in all other panels, followed by Holm–Sidak post hoc testing. Data are represented as a point for the result of each individual animal; bars are mean ± s.e.m. In a, “Mouse” icon by Iconic, “Test tube” icon by H. Alberto Gongora, “Muscle” icon by P. J. Witt from the Noun Project.

Improved exercise capacity with training is blunted in hyperglycaemic mice.

To determine how WD and STZ treatment affects improvements in exercise capacity, all mice (sedentary and trained) underwent a graded exercise test before, and 6 weeks into the exercise training period. Pretraining exercise capacity was similar among treatment groups (Extended Data Fig. 2a,b). The change (Δ) in exercise capacity in response to training was calculated for each mouse (Δ exercise capacity). The magnitude of improvement in exercise capacity with training, expressed as Δ distance run, was significantly lower in WD and STZ groups (Fig. 3a). These results indicate that induction of hyperglycaemia, independently of whether it is caused by diet or reduced insulin secretion, is associated with blunted improvements in aerobic exercise capacity. Accordingly, absolute exercise capacity after training, expressed as distance run, was roughly two-fold higher in CON compared to WD and STZ; and only CON mice achieved a statistically significant improvement in exercise capacity relative to their age-matched sedentary group (Fig. 3b). Similar results were seen when exercise capacity was expressed as time to exhaustion (Fig. 3c). Maximal oxygen consumption rate (VO_2peak) was measured as an additional index of exercise capacity in a subset of CON, WD and STZ animals. In line with distance and time to exhaustion results, only CON animals improved VO_2peak beyond the extent of their time-matched sedentary counterparts (Fig. 3d). No differences in exercise capacity were noted among sedentary mice in all treatment groups, demonstrating that blunted exercise capacity in WD and STZ results from failed adaptation to exercise, rather than a basal defect (Fig. 3b–d). Clinical studies have previously identified an inverse association between elevated fasting glucose and/or glucose intolerance and VO_2peak. Our data extend these findings by demonstrating that induction of hyperglycaemia by two independent mechanisms can blunt improvements in exercise capacity with training, thus strengthening evidence that a causative relationship may underlie these two traits. Moreover, we demonstrate that metabolic improvements with aerobic training (Fig. 2) can occur in the absence of improvements in exercise capacity (Fig. 3).

Fig. 3 |. Exercise capacity and muscle phenotype in response to exercise training.

CD-1 mice were fed a control diet (CON), WD or treated with STZ for 8 weeks, and placed in voluntary wheel running cages for a further 8 weeks (exercise-trained). A separate set of mice remained in cages without wheel access to act as controls (sedentary). Aerobic exercise capacity was tested before and after exercise training. a, Individual changes in exercise capacity (Δ) were calculated by subtracting posttraining exercise capacity from pretraining exercise capacity and plotted from lowest to highest response in all trained groups. b,c, Absolute exercise capacity was measured 6 weeks into the training period for exercise-trained and sedentary age-matched controls in distance (b) and time to exhaustion (c). a–c, CON SED n = 17, WD SED n = 16, STZ SED n = 22; CON EXT n = 17, WD EXT n = 16, STZ EXT n = 20). d, Maximal oxygen consumption, VO_2peak, was measured in a cohort of sedentary and exercise-trained mice by using a graded treadmill test as an indicator of cardiorespiratory fitness (WD SED n = 4; all other groups n = 5). Gastrocnemius muscles from a subset of animals that were matched for total wheel running distance were sectioned for histology. e, FITC-conjugated Griffonia simplicifolia lectin immunofluorescent staining was used to calculate capillary density in muscle. Scale bar, 100 µm. f, Capillary density (capillaries per mm²) was quantified in FIJI (CON SED n = 5; all other groups n = 6). g, Sections were stained with MHC type I (pink) and MHC type IIA (blue) to calculate oxidative versus glycolytic fibre distribution, and WGA to visualize the ECM (pink). Scale bar, 200 µm. Images are representative of n = 6 per group. h, Combined type I and type IIa fibre proportion was counted and plotted as percentage of total fibres per animal (all groups n = 6). Main effects were determined by one-way ANOVA relative to CON group in a and by two-way ANOVA otherwise, followed by Holm–Sidak post hoc testing. Data are represented as a point for the result of each individual animal, bars are mean ± s.e.m.

Aerobic remodelling of skeletal muscle is impaired in hyperglycaemic mice.

Aerobic training induces adaptive remodelling events in skeletal muscle that contribute to improved exercise capacity. Specifically, a shift towards a more oxidative muscle fibre type (type I/IIA), and an increase in muscle capillary density enhance the delivery and utiuselization of oxygen and nutrients during exercise, leading to improved exercise capacity^35–37. To determine whether impaired muscle remodelling may contribute to blunted improvements in exercise capacity in WD and STZ, we measured muscle fibre type and capillary density in sedentary and exercise-trained mice. Muscle capillary density and the proportion of oxidative (type I/IIA) muscle fibres were similar among sedentary mice. However, exercise training resulted in a 44% increase in muscle capillary density in CON mice, while capillary density was not significantly increased in WD or STZ (Fig. 3e,f). These data indicate that exercise-induced angiogenesis is impaired in WD and STZ. Likewise, the proportion of oxidative muscle fibres (type I and IIA) was increased to a greater degree by exercise training in CON mice, compared to WD and STZ (Fig. 3g,h). Our work identifies impaired aerobic remodelling of skeletal muscle as a potential contributor to blunted improvements in exercise capacity in models of hyperglycaemia.

Hyperglycaemia is associated with ECM accumulation in muscle.

There is well-established evidence that chronic hyperglycaemia in people with diabetes can lead to accumulation of the ECM in organs such as the kidney and heart^38,39. Recent evidence also demonstrates that insulin resistance induced by a high-fat diet can lead to the build-up of collagen, a chief ECM component, in skeletal muscle⁴⁰. As ECM accumulation can impair tissue remodelling and inhibit angiogenesis^41,42, we hypothesized that hyperglycaemia-induced alterations to the ECM may underlie blunted aerobic remodelling in WD and STZ. Fractional ECM area was measured in gastrocnemius muscle cross-sections stained with fluorescently labelled wheat germ agglutinin (WGA). We observed thickening of the ECM compartment surrounding myofibres in both WD and STZ, compared to CON (Fig. 4a). When fractional ECM area was calculated, the ECM was found to comprise 27 and 36% more of the muscle cross-sectional area in WD and STZ, respectively (Fig. 4b). Muscle ECM accumulation in individual mice was significantly correlated with blood glucose concentration (Fig. 4c; R = 0.477, P = 0.045), thus supporting the hypothesis that glucose can drive ECM accumulation in muscle. In addition, we measured total collagen content in hydrolysed muscle by using a hydroxyproline assay, and found that both WD and STZ displayed muscle collagen accretion compared to CON (Fig. 4d). These results demonstrate that blunted aerobic remodelling in WD and STZ is associated with increased skeletal muscle ECM density, which is due in part to increased collagen content.

Fig. 4 |. Impaired glucose tolerance and hyperglycaemia are associated with muscle ECM accretion.

Gastrocnemius muscles from sedentary CON, WD and STZ mice were collected to assess ECM accumulation. a, WGA staining demonstrates a thicker ECM compartment in WD and STZ compared to CON. Images are representative of n = 6 per group. Scale bar, 100 µm. b, Fractional (%) ECM area was calculated (n = 6); boxplot shows mean with error spanning minimum to maximum values. c, The relationship between muscle ECM accumulation in b was correlated with fasting glucose for each animal by Spearman’s correlation (n = 6). d, Hydroxyproline content was measured in gastrocnemius muscle as a marker of skeletal muscle collagen (CON n = 25, WD n = 20, STZ n = 12). e, WGA staining was performed on soleus muscle sections from rats bred for LRT or HRT, demonstrating higher ECM thickness in LRT. Scale bar, 200 µm. f, Relative ECM fractional area was calculated for n = 8 HRT and n = 9 LRT, boxplot shows mean with error spanning minimum to maximum values. g, Hydroxyproline content of LRT and HRT muscles was measured (n = 20 per group). h, Muscle sections were stained for AGE (green) and counterstained with WGA to visualize the ECM (red). i, The glycation signal was quantified (n = 6 per group, three images per individual rat) for pixel intensity and represented as a percentage of LRT; boxplot shows mean with error spanning minimum to maximum values. Scale bar, 200 µm. Main effects were determined by one-way ANOVA in b and d followed by Holm–Sidak post hoc testing, or by two-tailed unpaired t-test with Welch’s correction in remaining panels. Data are represented as a point for the result of each individual animal, bars are mean ± s.e.m.

Muscle ECM is altered in genetic models of low aerobic adaptation.

To determine whether ECM accumulation may be causally linked to low improvements in aerobic exercise capacity, we examined skeletal muscle from rats genetically selected over 15 generations for low or high response to aerobic training (LRT or HRT). We have previously published that LRT and HRT had similar exercise capacity in the untrained state, but LRT failed to improve exercise capacity after 8 weeks of aerobic training, while HRT achieved large improvements in exercise capacity¹⁸. Moreover, we demonstrated that selection for the LRT phenotype led to impaired aerobic remodelling in muscle, including a less oxidative skeletal muscle fibre type and blunted exercise-induced angiogenesis¹⁸. Consistent with a role for muscle ECM accumulation in impaired aerobic remodelling, we found thickening of skeletal muscle ECM in LRT rats, compared to HRT (Fig. 4e). Similarly to WD and STZ mice, LRT rats had higher total ECM fractional area (Fig. 4f) and muscle collagen content (Fig. 4g) versus HRT. Thus, our results demonstrate skeletal muscle ECM accumulation as a predictor of LRT in two distinct models of hyperglycaemia (WD, STZ), and a genetic model of LRT.

High-glucose-induced ECM modifications can impair angiogenesis.

In addition to causing increased ECM component expression, exposure to glucose and its metabolites can promote non-enzymatic modification of the ECM, known as glycation^43,44. Glycation can induce cross-linking of ECM molecules and prevent their turnover and degradation, thus contributing to ECM accumulation^43,45. We have previously demonstrated that LRT rats have impaired glucose tolerance compared to HRT¹⁸. To determine whether ECM glycation may contribute to ECM accumulation and impaired aerobic remodelling, we stained LRT and HRT muscle cross-sections with antibodies against glycated protein residues, also known as advanced glycation endproducts (AGE) (Fig. 4h). LRT muscle had significantly higher levels of glycation compared to HRT (Fig. 4i). In LRT, much of the glycation signal colocalized with the ECM compartment (Fig. 4h, merge), demonstrating that both ECM accumulation and glycation are caused by selection for low improvements in exercise capacity with aerobic training. Collectively, our data show that relatively modest perturbations in glucose metabolism, even in the absence of diabetes, are associated with profound changes to the ECM compartment of skeletal muscle.

To determine whether ECM glycation itself is sufficient to impair angiogenesis, we performed an assay using human umbilical vein endothelial cells (HUVECs), which can form capillary-like tubes in vitro when plated on ECM extract (Matrigel). Tissue culture wells were coated with Matrigel and treated with the glycating agent, methylglyoxal (0 or 3 mM). Methylglyoxal is a glucose metabolite that increases in the circulation of humans with hyperglycaemia, and more potently induces protein glycation than glucose itself⁴⁶. Following a 24-h glycation period, the wells were washed of excess methylglyoxal, and HUVECs were plated on the Matrigel with growth medium to induce tube formation (Fig. 5a). Tube formation was inhibited by ECM glycation, and cotreatment with the glycation inhibitor aminoguanidine (100 mM) restored tube formation to the level of the 0 mM control (Fig. 5b). When considered collectively (Figs. 4 and 5), our results point to ECM glycation and accumulation as possible mechanisms for impaired aerobic remodelling in muscle in models of impaired glucose tolerance and hyperglycaemia.

Fig. 5 |. In vitro angiogenesis and muscle gene expression changes in response to hyperglycaemia.

Cell culture experiments were performed to analyse the effect of hyperglycaemia on endothelial and muscle cell behaviour and gene expression. a, In vitro capillary tube formation was performed in HUVECs to determine the effect of ECM glycation by 0 and 3 mM methylglyoxal (MG) or treatment with glycation inhibitor (aminoguanidine, AG; 100 mM) over 16 h. Images are representative of n = 3 independent experiments. Scale bar, 250 µm. Tube formation was quantified using FIJI and corresponding analysis overlays are shown. b, Quantification for HUVEC total tube length for n = 3 independent experiments. c, Tube formation was measured in HUVECs grown in medium supplemented with 20% serum collected from normoglycemic CON (n = 6) mice or hyperglycemic STZ (n = 5) mice for 16 h. Serum experiments were run in duplicate for each mouse. Representative images from n = 4 mice per group are shown. Scale bar, 500 µm. c, Serum-conditioned HUVEC total tube length was quantified. FIJI representative overlays are shown in c for CON (top) and STZ (bottom). d, Quantification for HUVEC total segment length is shown for CON n = 6 and STZ n = 5 independent animals. e, HUVEC proliferation was measured over 4 d in cells cultured in low (LG, 5 mM) or high (HG, 25 mM) glucose media (n = 6). Fold change is expressed as the increase versus day 1 for each glucose condition. f, MHC and ECM gene expression were measured in C2C12 cells differentiated for 6 d in low (LG; 5 mM, n = 6) or high (HG; 25 mM, n = 6) glucose media. Fold change is expressed relative to the LG mean for each gene. Specific groups were compared via one-way ANOVA in e followed by Holm–Sidak post hoc testing, or by two-tailed unpaired t-test to respective controls with in remaining panels. Data are represented as a point for the result of each independent experiment in b, individual animal d or individual cell culture well in e and f; bars are mean ± s.e.m.

Hyperglycaemia alters endothelial cell function in vitro.

Previous investigations have demonstrated that exposure to high glucose can impair endothelial tube formation in vitro^47,48. To further investigate how the physiological environment in models of hyperglycaemia may affect angiogenesis, we plated HUVECs on Matrigel after suspension in basal medium supplemented with 20% serum from hyperglycaemic mice treated with STZ 6 weeks before blood collection or normoglycemic control mice (Fig. 5c). After 16 h, total capillary tube length was lower in wells treated with serum from hyperglycaemic mice, indicating impaired angiogenic potential (Fig. 5d). One mechanism by which hyperglycaemia may inhibit angiogenesis is by affecting endothelial cell proliferation, as expansion of this stem cell population is important for increases in muscle capillary density with exercise training^49,50. To determine how hyperglycaemia affects endothelial cell proliferation, we exposed HUVECs to low (5 mM) or high (25 mM) glucose conditions over 4 d and found that HUVEC proliferation was significantly impaired by exposure to high glucose (Fig. 5e). Thus, our results and those of other laboratories demonstrate that hyperglycaemia can have profound negative effects on endothelial cell function. Although relatively large differences in glucose concentration were used for in vitro studies, these data support our in vivo findings that exercise-induced angiogenesis is blunted in models of hyperglycaemia.

Hyperglycaemia is sufficient to alter muscle phenotype and ECM gene expression.

We have shown that, in vivo, hyperglycaemia is associated with a glycolytic muscle phenotype and ECM accumulation. To test whether hyperglycaemia is sufficient to induce changes in muscle phenotype, we differentiated C2C12 cultures in low (5 mM) or high (25 mM) glucose conditions for 6 d. Differentiation under high glucose conditions promoted an increase in glycolytic myosin heavy chain (MHC) isoform messenger RNA (Myh1, type IIx) and a concomitant decrease in oxidative MHC isoform expression (Myh7, type I) (Fig. 5f). In line with this, the expression of myostatin, which is a negative regulator of glycolytic muscle phenotype, was significantly down-regulated by high glucose (Fig. 5f). In addition, we demonstrate that high glucose caused significant increases in the expression of several collagen isoforms in cultured myotubes, including Col1a1, Col3a1, Col4a1 and Col8a1 (Fig. 5f). This evidence supports a direct role for hyperglycaemia in the phenotype of ECM accretion that is characteristic of our WD and STZ models and genetic model of LRT. In addition, we demonstrate that exposure to high glucose can promote a glycolytic muscle phenotype and impair oxidative fibre-type transformation, as we demonstrate in vivo.

Hyperglycaemia is associated with increased mechanical signalling in muscle.

Remodelling of skeletal muscle in response to exercise results from molecular signalling events initiated during each bout of acute exercise^51,52. To determine whether changes in muscle signal transduction contribute to failed aerobic adaptation in WD and STZ, we performed acute exercise experiments consisting of 30 min of moderate-intensity treadmill running. Our previous work identified a new signalling network centred around activation of the c-Jun N-terminal kinase (JNK) that was hyperactivated by aerobic exercise in rats that fail to improve aerobic exercise capacity with training (LRT)¹⁸. Using muscle-specific JNK knockout mice, it was also determined that JNK activation during exercise can blunt aerobic adaptations with training⁵³. Consistent with a role for JNK in impairing aerobic muscle remodelling and exercise capacity, we found significantly higher activation of JNK’s upstream kinase, MKK4 (Ser257), with acute aerobic exercise in WD compared to CON mice (Fig. 6a). In addition, JNK, along with a second MKK4 substrate, P38, were hyperactivated in muscle from WD mice with aerobic exercise (Fig. 6a). Phosphorylation of the JNK substrate, SMAD2, at its Ser245/250/255 residues was also increased with aerobic exercise in WD (Fig. 6a), indicating inhibition of the myostatin pathway and reduced repression of muscle growth. Moreover, phosphorylation of recently discovered JNK target sites on the ribosomal protein P70S6 kinase (Thr421/Ser424)⁵⁴ was higher in WD versus CON (Fig. 6a). Hyperactivation of this extensive JNK signalling cascade in WD was only observed after aerobic exercise, and not under basal conditions. Thus, our work further identifies significant diet-exercise interactions at the molecular level in skeletal muscle. Moreover, hyperphosphorylation of MKK4, P38, JNK, SMAD2 and P70S6K was also observed with acute exercise in the STZ model (Fig. 6b), demonstrating that hyperglycaemia, rather than increased body weight and insulin, may be responsible for aberrant signalling in WD mice.

Fig. 6 |. Hyperglycaemia is associated with altered muscle signalling with acute exercise.

WD-fed, STZ or CON diet-fed mice underwent a bout of moderate-intensity treadmill running for 30 min (AEX). A group of sedentary (SED) mice did not undergo treadmill running and acted as basal controls. Western blotting was used to determine acute exercise-induced signal transduction in gastrocnemius muscles. a,b, Blots are quantified from WD mice (a) and STZ-treated mice (b). Representative images are shown in a and b. In a, CON SED, CON AEX, WD SED, AD EAX n = 6 for all signalling proteins except p-JNK, n = 10 per group. In b, CON SED n = 8 (p-AKT, p-P70S6K n = 4), CON AEX n = 8 (p-MKK4, p-SMAD2L n = 7; p-AKT, p-P38, p-P70S6K n = 4), STZ SED n = 12 (p-AKT, p-P70S6K n = 6), STZ AEX n = 12 (p-AKT, p-MKK4, p-P70S6K n = 6, p-P38 n = 5). C, control diet; S, STZ-treated; SED, resting control; AEX, acute exercise. c, EDL muscles excised from CD-1 mice on control diet were placed in oxygenated Krebs–Henseleit buffer (B, basal) or stretched by 10% of optimal length for 10 min (S, stretch). Muscles were snap frozen for western blotting to determine phosphorylation and activation of mechanosensitive proteins; n = 6 per group. d,e, Mass of gastrocnemius (Gastroc.) (d) and tibialis anterior (TA) muscles were measured (e) (CON n = 10, WD n = 9, STZ n = 6). f, Muscle fibre cross-sectional area was measured in gastrocnemius after WGA staining (CON n = 10, WD n = 6, STZ n = 10). g, JNK activation after 30 min of treadmill running was correlated with blood glucose in a separate cohort of CON, WD, STZ and NOD mice (n = 38). Data in g were analysed using Spearman’s correlation; R and P values are indicated. p-SEK1/MKK4, phosphorylated mitogen-activated protein kinase kinase 4; p-P38, phosphorylated mitogen-activated protein kinase 38; p-JNK, phosphorylated c-JNK; p-SMAD2L, linker region phosphorylated SMAD2 and p-P70S6K T421/S424, phosphorylated P70S6 kinase on residues Thr421/Ser424. Group differences were determined by two-way ANOVA in a and b, or by one-way ANOVA in d, e and f followed by Holm–Sidak post hoc testing. Data are represented as a point for the result of each individual animal; bars are mean ± s.e.m.

To determine whether altered signalling with exercise in hyperglycaemic models was specific to this JNK signalling network, we measured the phosphorylation of other common exercise-activated proteins. 5′ adenosine monophosphate-activated protein kinase (AMPK) and extracellular signal-regulated kinase (ERK), which can be activated by more intense exercise^55,56, were not activated by our relatively moderate aerobic exercise protocol (Fig. 6a,b). In addition, phosphorylation of Akt at Ser473 was increased by exercise in all groups to a similar extent. These results indicate that hyperactivation of exercise-responsive signalling in hyperglycaemic models, was at least specific to JNK-mediated growth pathways, as other pathways were similarly regulated by exercise in CON, WD and STZ.

To determine whether hyperglycaemia itself is associated with aberrant JNK activation with exercise, we performed an acute exercise experiment in a separate cohort of mice with a range of blood glucose concentrations (127–245 mg dl⁻¹). Four groups were used for this experiment: (1) male CD-1 CON, (2) male CD-1 WD, (3) male CD-1 STZ and (4) male and female non-obese diabetic mice (NOD). In contrast to WD and STZ, NOD is an inbred genetic model that spontaneously develops hyperglycaemia in early adulthood. Random glucose measurements were taken from all groups in the days before undergoing an acute exercise bout, and the highest recorded glucose was correlated to JNK activation, calculated as fold change versus sedentary, in response to a bout of moderate treadmill running exercise (representative blots in Extended Data Fig. 3a). We demonstrate that blood glucose concentration was a significant predictor of JNK activation with exercise in mice, irrespective of strain, sex or aetiology of hyperglycaemia (Fig. 6g, R = 0.439, P = 0.006). These data support our hypothesis that hyperglycaemia can contribute to changes in muscle signalling and phenotype in models of metabolic disease.

While we demonstrate aberrant signal transduction in models of chronic hyperglycaemia, it is unknown whether increased blood glucose concentrations at the time of exercise can also alter exercise-induced signal transduction. To test this, we used a separate cohort of chow-fed CD-1 mice and injected half with a bolus of glucose to determine the effect of acute hyperglycaemia on JNK activation with exercise (Extended Data Fig. 3b,c). We found no effect of acute hyperglycaemia on exercise-induced JNK activation (Extended Data Fig. 3b), indicating that defects due to chronic, rather than acute, hyperglycaemia are potentially responsible for JNK hyperactivation in WD and STZ. In support of that conclusion, we performed a time course experiment in WD mice, and found that hyperactivation of JNK with exercise emerges between 8 and 16 weeks of treatment and becomes exacerbated with longer duration of hyperglycaemia, up to 32 weeks (Extended Data Fig. 3d).

Exercise and muscle contraction can stimulate adaptive signalling via pathways responsive to metabolic or mechanical stress. One key role of the ECM in muscle is to regulate mechanical signal transduction. Given our data demonstrating profound changes to the ECM compartment in WD and STZ mice, we hypothesize that changes in signal transduction with exercise are due to increases in mechanical stress resulting from ECM accumulation. To determine the responsiveness of signalling pathways altered in WD and STZ to mechanical stimulation, we performed an in vitro muscle stretch experiment. As anticipated, AMPK, which is typically responsive to metabolic stressors such as reduced ATP availability, was not activated by stretch in muscle (Fig. 6c). Likewise, the mitogen-activated kinase, ERK, was minimally activated by stretch. In contrast, MKK4, P38, JNK, SMAD2 and P70S6K were all robustly phosphorylated in response to mechanical stress (Fig. 6c). Therefore, our data demonstrate that muscle ECM accumulation in WD and STZ is associated with hyperphosphorylation of proteins activated by mechanical stress in response to endurance exercise (Fig. 6a,b). Further study would be needed to determine whether preventing hyperglycaemia-induced increases in muscle ECM can restore mechanical signalling in models of hyperglycaemia.

Activation of mechanical signalling is typically associated with resistance exercise that induces high levels of contractile force and mechanical stress compared to endurance exercise. Indeed, we recently demonstrated that activation of the JNK/SMAD2 signalling axis occurs with resistance exercise, but not endurance exercise in healthy humans⁵³. Consistent with a role for JNK/SMAD2 signalling in resistance training adaptations, we found that this signalling axis is a negative regulator of myostatin, and is necessary for muscle growth with mechanical overload⁵³. In addition, P70S6K, which we found to be phosphorylated on JNK-specific sites Thr421/Ser424 in WD and STZ with aerobic exercise, is typically activated with overload or resistance exercise^54,57,58 as it stimulates protein synthesis and muscle growth. In line with growth/resistance pathway activation, we found the mass of the gastrocnemius and tibialis anterior muscles to be greater in WD and STZ compared to CON (Fig. 6d,e). Consistent with our findings, humans and animal models of diabetes or prediabetes can have increased muscle mass compared to controls^59–61, although advanced diabetes with neuropathy can lead to muscle wasting⁶². Increased muscle mass with hyperglycaemia was due, at least in part, to a shift in muscle fibre size towards a larger cross-sectional area (Fig. 6f). Thus, our data clearly demonstrate the inappropriate activation of muscle growth and resistance exercise pathways, specifically during moderate aerobic exercise, in two models of hyperglycaemia. As JNK activation in muscle is sufficient to blunt improvements in exercise capacity with training⁵³, our data identify this aberrant signalling as a potential mechanism for impaired aerobic adaptation with hyperglycaemia.

Hyperactivation of JNK with exercise in humans with impaired glucose tolerance.

We next sought to determine whether increased mechanical signalling might also be a mechanism for impaired aerobic adaptation in human participants. In a cohort of 24 individuals, we found that impaired glucose tolerance, indicated by high glucose area under the curve (AUC), was a strong predictor of low VO_2peak (Fig. 7a). The relationship between impaired glucose tolerance and exercise capacity also remained when absolute VO_2peak in ml min⁻¹, not adjusted for body weight, was used (R = −0.55; P = 0.005). As insulin resistance is a primary contributor to impaired glucose tolerance in humans, insulin sensitivity was calculated using the Si_OGTT index²². In line with our glucose AUC data, we demonstrate a significant negative correlation between VO_2peak and insulin resistance in human participants (Fig. 7b). Insulin sensitivity predicted VO_2peak more robustly than body mass index (BMI), despite the fact that VO_2peak is calculated per body weight (Fig. 7c). Thus, our data are consistent with other studies demonstrating robust relationships between hyperglycaemia and low exercise capacity^19,22. To assess whether impaired glucose tolerance is associated with increased mechanical signalling with exercise, all participants underwent a moderate bout of aerobic exercise at an intensity of 60% of their calculated VO_2peak, and muscle biopsies were taken before and after exercise. Similarly to our findings in animal models, human participants with impaired glucose tolerance had higher activation of JNK with aerobic exercise, leading to greater phosphorylation of its downstream substrate, SMAD2 (Fig. 7d). JNK activation with exercise displayed a strong negative correlation with Si_OGTT (R = −0.612, P = 0.003; Fig. 7e), thus identifying this index of insulin sensitivity as a predictor of exercise-induced signal transduction in addition to VO_2peak. JNK activation was also correlated with BMI in our participants (Fig. 7f, R = 0.442, P = 0.04); however, as with VO_2peak, we found clinical markers of glucose homeostasis, such as Si_OGTT, to be more robust predictors of JNK activation with exercise in humans. Intervention studies will be needed to determine whether hyperglycaemia and JNK activation with acute exercise in humans can predict low improvements in exercise capacity with training, as we demonstrate in mouse models of metabolic disease. When considered collectively, our data demonstrate that chronic hyperglycaemia and impaired glucose tolerance are associated with significant alterations in the molecular response to exercise in humans and animal models, which may contribute to impaired aerobic capacity (Fig. 8).

Fig. 7 |. Exercise-induced JNK activation increases with glucose intolerance in humans.

Participants completed a 75 g OGTT on visit one, and a maximal oxygen uptake test, or VO_2peak test, on a cycle ergometer during visit two. a, The correlation between aerobic exercise capacity (VO_2peak, ml kg⁻¹ min⁻¹) and glucose tolerance (glucose AUC) was calculated. b,c, VO_2peak was also correlated with OGTT-derived insulin sensitivity (Si_OGTT) (b) and BMI (c). On visit 3, each participant completed a standardized acute exercise bout at 60% of their VO_2peak. Muscle biopsies were taken from the vastus lateralis before and after acute exercise. Muscle was frozen immediately and processed for western blotting. JNK activation and downstream p-SMAD2L activation were determined by western blotting. d, Representative images for JNK activation are shown for n = 2 participants with low OGTT AUC (glucose tolerant) and n = 2 participants high OGTT AUC (glucose intolerant). e, Individual JNK activation (fold change from pre- to postexercise) was robustly correlated with insulin sensitivity, calculated as Si_OGTT. f, Individual JNK activation (fold change from pre- to postexercise) was also significantly correlated with BMI, although to a lesser extent than Si_OGTT. Pearson correlation coefficients were calculated in a–c, e and f; R and P values are indicated for each statistical comparison; n = 24 for a–c and n = 22 for e and f. Data are represented as a point for the result of each individual participant.

Fig. 8 |. Hypothesized mechanisms by which hyperglycaemia may blunt aerobic adaptations with exercise.

Chronic hyperglycaemia is associated with skeletal muscle ECM accumulation, including increased total ECM area and glycation, and collagen content. We hypothesize that these hyperglycaemia-associated changes in muscle structure promote increased mechanical stress during muscle contraction, and hyperactivation of the MKK4–JNK signalling axis, and downstream targets SMAD2 and p70S6K during acute aerobic exercise. This signalling cascade—more typically activated by resistance exercise—is inappropriately initiated during moderate aerobic exercise in hyperglycaemic models. Our data indicate that long-term hyperglycaemia promotes a persistent glycolytic muscle phenotype and impairs aerobic remodelling adaptations that occur in response to aerobic exercise training. We propose that these physiological and molecular mechanisms underlie blunted responses to exercise training and a phenotype of low exercise capacity in humans and animal models with metabolic disease. “Western Diet” icon by Blaise Sewell, “Runner” icon by Michael Scott Fischer from the Noun Project. The medical art images were produced using Servier Medical Art (https://smart.servier.com). Servier Medical Art by Servier is licensed under a Creative Commons Attribution 3.0 Unported License (https://creativecommons.org/licenses/by/3.0/).

Discussion

Epidemiological studies have highlighted that attaining high exercise capacity is paramount for protection against mortality⁸, but that exercise capacity tends to be lower in individuals with metabolic diseases resulting in hyperglycaemia^20,21. Here, we demonstrate that inducing hyperglycaemia in the weeks before exercise training is associated with blunted gains in exercise capacity and aerobic remodelling, suggesting that hyperglycaemia may impart a resistance to aerobic training adaptations. The existence of a subset of the population that achieves minimal improvements in aerobic exercise capacity with training is supported by significant evidence^{9,10,13,16,17}. As a result, some individuals require a larger training stimulus than others to achieve even modest gains¹⁵. Our data support seminal clinical observations by Bouchard et al.⁹, that the exercise response phenotype is largely modifiable by environmental changes, and identify WD consumption and STZ treatment as mediators of the training response phenotype. Given recent increases in the incidence of hyperglycaemia due to dietary shifts and dramatic elevations in rates of metabolic disease^23,63, our results suggest that the population may become increasingly resistant to improvements in exercise capacity.

One limitation of studying chronic hyperglycaemia in animal models and humans is that it cannot be studied in isolation due to inevitable homeostatic adjustments in other variables such as body weight and insulin concentration. To strengthen our conclusions regarding hyperglycaemia, we chose to study two independent models that partially reflect the pathologies leading to hyperglycaemia in humans. Both WD and STZ mice had similar impairments in exercise capacity, which were accompanied by commensurate changes in muscle phenotype and signal transduction. The fact that these models are divergent with respect to body weight and insulin levels supports the conclusion that our results may be attributable to chronic hyperglycaemia. In support of that conclusion, our in vitro data demonstrate that exposure to high glucose is sufficient to inhibit critical aerobic remodelling processes such as angiogenesis and an oxidative fibre phenotype, which were impaired in our animal models of hyperglycaemia. These results suggest that hyperglycaemia may be one mechanism behind previous reports that individuals with poor glycemic control have a less aerobic muscle phenotype, including lower capillary density and a smaller proportion of oxidative muscle fibres, than those without metabolic impairments^64–67. Although we identify impaired muscle remodelling as one mechanism underlying blunted improvements in exercise capacity, future studies should elucidate whether other components that contribute to VO_2peak are divergently regulated by exercise training in healthy versus metabolically impaired populations.

Our data identify glucose-induced changes in muscle ECM composition and ECM glycation as potential mechanisms for broad maladaptive muscle remodelling and persistently low exercise capacity in models of hyperglycaemia. This hypothesis is supported by clinical data demonstrating that ECM gene expression in skeletal muscle can predict improvements in aerobic exercise capacity in humans⁶⁸, and that ECM genes are differentially regulated by exercise in people with metabolic impairments compared to healthy individuals⁶⁹. High glucose enhances ECM density and stiffness, thus increasing mechanical stress in muscle, which is transmitted via mitogen-activated protein kinases (MAPKs)⁷⁰ to the nucleus to influence remodelling events⁷¹. We hypothesize that increases in ECM density and glycation in WD, STZ and genetic models of low aerobic response (LRT) cause the muscle to misinterpret exercise that normally induces low levels of mechanical stress (for example, modest treadmill running) as an exercise with higher mechanical stress (for example, resistance exercise). As a result of this inappropriate signalling program, the muscles fail to undergo aerobic remodelling, but instead take on a glycolytic phenotype more congruent with resistance exercise-induced signalling.

Our previous work identified JNK activation with exercise as a molecular ‘switch’ that inhibits aerobic adaptations and promotes resistance-type adaptations in muscle^18,53. JNK probably exerts its effects on muscle phenotype via phosphorylation of SMAD2 that can inhibit myostatin activity⁵³ and activation of targets such as P70S6K, which can promote muscle growth^54,58. Phosphorylation of SMAD2 (Ser245/250/255) and P70S6K (T421/S424) was elevated in WD and STZ models, as was the upstream kinase of JNK, MKK4. In addition, we have previously demonstrated that JNK/SMAD signalling is hyperactivated with aerobic exercise in rat models of low response to training¹⁸ and now show that activation of this pathway is predicted by impaired glucose metabolism in humans (Fig. 7). JNK/SMAD signalling is normally activated by resistance exercise, and not aerobic exercise, in healthy humans⁵³. Thus, our work identifies an extensive growth/resistance signalling network that is inappropriately activated with aerobic exercise in multiple rodent models of hyperglycaemia and human participants, and predicts impaired aerobic adaptation. The hyperactivation of resistance exercise-induced signalling with aerobic exercise presents the intriguing possibility that although aerobic adaptations are impaired, individuals with hyperglycaemia may respond well to resistance training, which is also recommended for maintaining health.

Despite blunted improvements in exercise capacity, mice with hyperglycaemia had increases in muscle protein content of critical metabolic regulators (for example, GLUT4) along with improved glucose tolerance with training (Fig. 2). These data indicate that metabolic improvements with exercise are disassociated from improvements in exercise capacity, a contention that is supported by clinical evidence⁷². This presents the possibility that exercise-induced improvements in glucose metabolism may, over time, reduce the physiological barriers such as ECM glycation and accretion that are associated with low improvements in aerobic capacity (Fig. 4). However, the time course for reversal of glucose-induced changes to ECM is unknown. We anticipate that reversal of ECM defects may be slow to occur, even following normalization of glucose homeostasis, as the half-lives of ECM proteins are very long (that is, months to years), leading to slow turnover, a process that is attenuated further by cross-linking induced by glycation^73,74. Future investigations should determine whether amelioration of glycemic control, if it is through prolonged exercise training or pharmacological means, can prevent hyperglycaemia-associated ECM accumulation and restore aerobic response to training.

In summary, we identify WD consumption and STZ treatment as negative regulators of the exercise capacity phenotype, which is a key predictor of health and longevity in humans. Our work has uncovered a physiological and molecular axis connecting hyperglycaemia and skeletal muscle ECM to altered signalling and impaired aerobic adaptations. These results demonstrate that diet and metabolic disease can fundamentally change how skeletal muscle signals and adapts to aerobic training, and may have important implications for a growing population with hyperglycaemia.

Methods

Animal experiments.

Mice.

For exercise training and acute exercise experiments, 8-week-old CD-1 mice (strain IGS, stock no. 022) were purchased from Charles River Laboratories. We selected this strain specifically for its outbred background, to increase translation of results to other species and models. All mice were housed in a specific-pathogen-free facility and maintained on a 12-h light/dark cycle (6:30 and 18:30 eastern standard time) with ad libitum access to diet. Control mice (CON) consumed a low-fat diet (Research Diets D14020502) consisting of 17% kcal from protein, 73% carbohydrate (0% sucrose) and 10% fat. To induce glucose intolerance and modest hyperglycaemia, a subset of mice were fed a WD for 8 weeks (Research Diets D12079B), consisting of 17% kcal from protein, 43% carbohydrate (34.9% sucrose) and 40% fat (38% saturated fat). WD and CON diets were matched for protein, micronutrient and fibre content. A second model of glucose intolerance/hyperglycaemia was induced with low-dose STZ (catalogue no. 1621 Tocris Bioscience) dissolved in diluted citrate buffer (114 mM; 0.5 M stock catalogue no. 2034, Boston Bioproducts). STZ was administered in 40 mg kg⁻¹ intraperitoneal injections on 2 consecutive days after a 4 h fast. STZ-treated mice consumed a control diet. Body composition was measured in anaesthetized mice by using a Lunar PIXImus2 mouse densitometer (DEXA). Male mice were used for all experiments in CON, WD and STZ models. Male and female NOD mice were purchased from Jackson Laboratories at 8 weeks of age, and maintained for >5 weeks on control diet (strain NOD/ShiLtJ; stock no. 001976). NOD females were included, as they develop hyperglycaemia at an earlier age than male NOD. Blood glucose was monitored weekly. For tissue collection, mice were fasted for a minimum of 2 h before being anaesthetized with isofluorane and killed by cervical dislocation. All mouse experiments were approved by the Institutional Animal Care and Use Committee of the Joslin Diabetes Center.

Rats.

A founder generation of N:NIH outcrossed stock rats were tested for exercise capacity at ~10 weeks of age and subsequently underwent a standardized aerobic exercise training protocol for 8 weeks. Individual improvements in exercise capacity following the aerobic training intervention were calculated as Δ = (posttraining exercise capacity – pretraining exercise capacity) to assess the magnitude of training response. Rats with the highest response to aerobic training (HRT) were selectively bred; conversely the lowest responders (LRT) were bred to create a separate line¹¹. Data shown here are from female LRT and HRT rats following 15 generations of selective breeding. General characteristics of this cohort were published previously¹⁸. Rats were fed rodent pellet diet (diet no. 5001; Purina Mills) and given free access to water. All procedures were performed in accordance with the University Committee on Use and Care of Animals at the University of Michigan.

Glucose tolerance tests.

To assess whole-body glucose tolerance, CON, WD and STZ mice were administered an intraperitoneal bolus of glucose (2 g kg⁻¹) after a 5 h fast. Blood glucose was sampled from the tail vein and recorded using an Infinity (US diagnostics) glucose-monitoring system for specified time intervals after injection. GTTs were performed 8 weeks after diet or STZ treatment and repeated near the end of the exercise-training period (see Fig. 2a for timeline). A separate cohort of age-matched sedentary mice were tested at the same timepoints and acted as controls. AUC was calculated using GraphPad Prism 8 software (GraphPad Software). Access to running wheels was restricted for 24 h preceding any GTT to washout the effects of acute exercise on glucose tolerance.

Circulating glucose and insulin.

Blood was sampled from the tail vein of unfasted mice during their light (7:00) and dark (20:00) cycles and blood glucose was measured using an Infinity glucometer (US Diagnostics). Fasting blood glucose (5 h) was measured at the 0 min timepoint of the posttraining GTT (~14 weeks of treatment in CON, WD or STZ animals) in sedentary animals. Glucose values were also used to demonstrate the correlation with muscle ECM accumulation in the same animals (methods described next; ECM fractional area). For insulin measurements, whole blood was collected from the tail vein by capillary tube (catalogue no. 16440100, Sarstedt) after a 12 h fast. Serum was isolated by centrifugation at 10,000g for 20 min and insulin concentration was measured using a commercially available kit (Millipore, EZRMI-13K).

Exercise capacity testing.

Mice were acclimated to a modular treadmill for 2 consecutive days before any testing (Columbus Instruments). Each animal rested in the treadmill lane for 15 min, followed by 5 min at 5 m min⁻¹ and 5 min at 10 m min⁻¹ at a 0° angle. Following acclimation, mice underwent aerobic exercise capacity testing to exhaustion on a modular treadmill set to a 5° angle. Mice were motivated to run with a shock grid operating at 0.56 mA. The starting speed was 5 m min⁻¹ and was increased by 1 m min⁻¹ until exhaustion. Speed was capped at a maximum of 25 m min⁻¹, to ensure that performance on the test was based on aerobic fitness, rather than biomechanical limitations that can arise with excessive treadmill speed. Mice that achieved a speed of 25 m min⁻¹ during the test were permitted to run at that speed until exhaustion. Exhaustion was defined as failure to return to the treadmill from the rest platform after three consecutive attempts to continue within the last ~1 min of running. Aerobic capacity was expressed as total distance run during the test. Pretraining aerobic exercise capacity was measured after 8 weeks of CON, WD or STZ treatment and was repeated in the final weeks of the exercise training period. Access to running wheels was restricted for at least 24 h before testing. Age-matched sedentary control mice were tested at the same timepoints to control for the effects of age, repeated testing and experimental conditions on exercise capacity. It was not possible to blind the capacity tester to CON versus WD mice, given obvious body weight differences, however, in each CON, WD and STZ group, sedentary or exercise-training allocations were blinded until data entry.

VO_2peak testing.

Mice were acclimated to a modular treadmill for 2 consecutive days before any testing, as described above (Columbus Instruments). Following acclimation, mice underwent individual VO_2peak testing with gas exchange monitoring in the Comprehensive Laboratory Animals Monitoring System on a modular treadmill set to a 5° angle. Mice were motivated to run with a shock grid operating at 0.56 mA). The starting speed was 5 m min⁻¹ and was increased by 1 m min⁻¹ until exhaustion without a maximum speed cap. VO₂ and VCO₂ were collected in 1 min intervals. Exhaustion was defined as failure to return to the treadmill from the rest platform after three consecutive attempts to continue within the last ~1 min of running. All mice reached a plateau in VO₂ at or before the last gas collection interval and VO_2peak values were normalized to body mass.

Voluntary wheel running.

Sedentary and exercise-trained groups within each control or hyperglycemic model were matched for GTT AUC, baseline aerobic exercise capacity and body weight at baseline. Statistical approaches to assess ‘non-response’ to exercise training have been the source of scrutiny and controversy, in large part due to the absence of a non-exercised control group and test/retest considerations in human studies^16,75. Our investigation included sedentary controls with each cohort of exercise-trained mice to account for variance in the exercise capacity measure due to age, experimental conditions and repeated testing. Sedentary mice were singly housed in a cage without a running wheel, and cage enrichment was provided to minimize stress. Mice assigned to exercise training were singly housed in a cage with a running wheel and voluntary running behaviour was recorded daily using a wireless cycling computer (catalogue no. 425309, Bontrager). Mice were allowed to run uninterrupted for the first 6 weeks of the exercise training period. During the final 2 weeks of training, wheels were removed intermittently for 24 h to accommodate washout periods before metabolic testing (for example, GTT and DEXA) and terminal dissections. Individual wheel running behaviour for mice in each experimental group during the 6-week uninterrupted training period is shown in Fig. 1f, while the cumulative aerobic training distance is shown in Fig. 2b. Due to the large number of animals required for data collection from our six experimental groups, treatments were performed in four separate cohorts that contained n = 4–10 animals from each treatment group. Combined data from each cohort are displayed in figures.

Running behaviour.

In a cohort of mice with access to running wheels, temporal running behaviour was collected over a 24 h diurnal period (light/dark, 6:30/18:30). A Columbus Instruments system was used to collect data using a Hall Effect Sensor (0297–0501), Wheel Counter 8 Channel Interface (0297–0050) and a Quad CI-Bus to interface with CI Multi Device Software (v.1.5.5). Wheel revolutions were recorded in 10-min intervals and accumulated over 24 h to calculate the percent of time spent running in the dark and light phases. Average wheel speed was collected throughout the entirety of the training period (Gotime Computer).

Acute exercise experiments.

To assess the effect of chronic hyperglycaemia on molecular signalling in response to acute exercise, a separate subset of CD-1 mice were maintained on the CON diet, WD-fed or STZ treated (2 × 40 mg kg⁻¹) for 8–32 weeks in the absence of exercise training. All mice were treadmill acclimated for 2 consecutive days. On the morning of experiments, mice were fasted for 2 h before completing an acute exercise bout of 30 min moderate-intensity running at a 5° incline and speed of 11 m min⁻¹. A shock grid was activated on contact (0.1 mA intensity) during the acute exercise but was required only minimally at this moderate running speed. Following completion of the exercise bout, mice were immediately removed from the treadmill and anaesthetized with isofluorane. Gastrocnemius muscle was quickly harvested, snap frozen in liquid N₂ and stored at −80 °C for subsequent analysis. An equal number of animals did not undergo treadmill running and acted as sedentary controls.

To determine whether JNK activation with exercise is affected by acute increases in blood glucose in addition to chronic hyperglycaemia, a separate subset of CD-1 mice were maintained on a control chow diet and underwent treadmill acclimation for 2 consecutive days. On the third day, mice were fasted for 2 h before being split into four groups: (1) control sedentary, (2) acute glucose sedentary, (3) control 30 min exercise and (4) acute glucose 30 min exercise. Mice allocated to glucose treatment were injected with 3 g kg⁻¹ glucose to rapidly increase blood glucose levels. Acute treadmill exercise was completed at a 5° incline, and speed of 11 m min⁻¹ for 30 min. Mice were immediately removed from the treadmill and anaesthetized with isofluorane. Gastrocnemius muscle was quickly harvested, snap frozen in liquid N₂ and stored at −80 °C for subsequent analysis.

Western blotting.

Skeletal muscle signalling was analysed using methods as previously published⁵³. The following antibodies were used for the detection of total or phosphorylated protein levels: Hexokinase II (sc 6521), GLUT4 (Abcam 654; samples unboiled), OXPHOS (Abcam 110413), p-Akt Ser473 (CST 9271), p-AMPK (CST 2531), p-ERK 1/2 (CST 4370), p-SEK1/MKK4 (CST 4514), p-P38 (CST 4511), p-JNK (CST 4668), p-SMAD2-L (CST 3104) and p-P70S6K T421/S424 (CST 9204). All primary antibodies were used at 1:1,000 dilution in TBST, with the exception of p-SMAD2-L, which was diluted at 1:500.

Adipose tissue and liver histology.

Mice were euthanized and visceral adipose tissue was collected from the perigonadal area in CON, WD and STZ sedentary and exercise-trained groups. Liver was harvested from the same animals. Tissues were stored in 10% neutral buffered formalin, transferred to PBS, paraffin embedded and sectioned for histological staining with hematoxylin and eosin (H&E). Images were acquired at ×10 magnification (Olympus, DP74). Adipocyte cross-sectional area was calculated in FIJI with at least 200 adipocytes counted per sample.

Liver glycerol.

Frozen liver samples were processed into saponified neutralized extracts using volumes of Ethanolic KOH (two parts EtOH: one part 30% KOH), 1 M MgCl₂ × 6H₂O and H₂O: EtOH (1:1) in combination with sequential centrifugation. Liver glycerol was assayed using a commercially available kit (catalogue no. 10010755, Cayman Chemical Company). Glycerol concentrations were expressed per mg of wet tissue.

Muscle histology.

To assess muscle phenotype histologically in mice and rats, whole gastrocnemius (mice) or soleus (rats) muscles were frozen with ice-cold isopentane during terminal dissections and sectioned at 20 °C in a cryostat at a thickness of 6 µm. Capillary density was measured with fluorescein-conjugated Griffonia simplicifolia lectin (1:100; catalogue no. FL-1101–2, Vector Laboratories) in PBS for 1 h. For fibre-type quantification, slides were incubated with primary antibodies against MHC I (MHCI (1:25; A4.951, hybridoma bank) and MHCIIa (1:25; SC-71, hybridoma bank) in 1% BSA in TBS + 1% Normal Goat Serum (NGS) overnight at 4 °C. Mouse IgG1 fluorescent conjugate (A21124; red) and Mouse IgM (A21042; green) secondary antibodies were added at 1:1,000 in 1% BSA in TBS + 1% NGS for 1 h. WGA (Texas Red 1:1,000, Thermo Fisher W7024) was added with secondary antibodies to counterstain for visualization of unstained fibres. Fibre-type distribution was completed by a researcher blinded to the study design. Absolute fibre counts were tracked in FIJI and expressed as a percent of the total number of fibres, for a total of 400–700 per sample. For cross-sectional area analysis, at least 300 individual fibres per sample were counted via FIJI. Muscle fibre size distribution was expressed as a proportion of the total number of fibres per image. All skeletal muscle images were obtained with an Olympus VS120 slidescanner at ×10 and an numerical aperture (NA) lens (×10, 0.4 NA), and exposure times were kept constant for each type of analysis. For visualization of AGE, soleus muscle from LRT-HRT rats was stained with anti-AGE antibodies (1:100, Abcam (ab23722)) and a simple pipeline for total pixel intensity was used to average 3–4 separate images per sample using CellProfiler software.

ECM fractional area.

In CON, WD and STZ animals, and LRT and HRT rats, WGA (1:1,000, Texas Red) in 1% BSA in TBS + 1% NGS was applied to slides for 1 h to calculate fractional area of the ECM via FIJI. Images were taken on an Olympus IX51 inverted fluorescent microscope and exposure time was held constant across samples (×20, Olympus D972 camera).

Muscle stretch experiments.

To measure skeletal muscle markers of mechanotransduction, a subset of CON CD-1 mice were killed by cervical dislocation for in vitro muscle stretch experiments. Extensor digitorum longus (EDL) muscles from both hindlimbs were rapidly dissected and tendon ends were tied with silk sutures. Muscles were vertically mounted to a force transducer with slack to limit muscle stretching. To determine the optimal muscle length, each EDL was lengthened by 1-mm increments and a short electrical stimulus applied at each tension/length (1 ms at 100 V) at ~30-s intervals until force production plateaued. The tension/length at which maximal force was achieved was considered L_o for that muscle and used for any further experimental procedures. Muscles were left to equilibrate at resting tension and incubated at 30 °C in a water bath enveloping oxygenated Krebs–Henseleit buffer with 5 mM glucose for 5 min. Following pre-incubation and equilibration, one muscle from each mouse was stretched 1 mm past its resting length (approximately 10% stretch) for 10 min while the contralateral muscle acted as a basal control. Immediately following the experiment, muscles were frozen in liquid N₂ and stored at −80 °C until processing for western blotting.

Hydroxyproline.

To measure hydroxyproline content in muscle, ~20–30 mg pulverized tissue was hydroylzed in a 15× volume of 6 N HCl for 3 h. Samples were then assayed on a 96-well plate using chloramine T reagent (dissolved in acetate-citrate buffer, pH 6.5 and 50% propanol) and Ehrlich’s reagent (p-dimethylaminobenzaldeyhde dissolved in 2:1 v/v propanol: perchloric acid). Hydroxyproline content was measured colorimetrically and expressed per mg of muscle tissue.

Tissue culture.

HUVEC.

HUVECs (passage 2–5, catalogue no. 32519 A, Lonza) were used for all experiments and maintained in Endothelial Growth Medium-2 (Lonza). HUVECs were passaged on collagen (catalogue no. 354236, Corning) coated 10 cm dishes until 70% confluence. To assess the effect of ECM glycation on endothelial tube formation, 96-well plates were coated with Matrigel (catalogue no. 354230, Corning) treated with 0–3 mM of the glycating agent, methylglyoxal (catalogue no. MO252, Sigma) in the presence of absence of 100 mM aminoguanidine, which is a glycation inhibitor (catalogue no. 396494, Sigma) for 24 h. Following glycation treatment, the Matrigel-coated wells were washed three times with PBS to remove unbound reagents and HUVECs were seeded at equal density (2 × 10⁴ cells per well) across experimental wells. The images shown are representative of n = 3 independent experiments.

In a separate set of experiments, HUVECs were treated with serum-conditioned media from normoglycemic CON or STZ-treated mice that received their last dose of STZ 6 weeks before serum collection (n = 5–6 mice per group). Cells were maintained in endothelial growth medium and passaged as above. On the day of plating, media were supplemented with 20% (by volume) serum from CON or STZ animals, and HUVECs were seeded at equal density (2 × 10⁴ cells per well) on Matrigel. In all experiments, tube formation proceeded overnight for 16 h, after which, each well was fixed with methanol and stained with methylene blue dye to be visualized using ×4 or ×10 bright field microscopy. Total capillary network length was quantified in an unbiased manner using the angiogenesis plugin on FIJI, with three iterations computed per image (http://image.bio.methods.free.fr/ImageJ/?Angiogenesis-Analyzer-for-ImageJ).

HUVEC proliferation.

HUVECs were pretreated and maintained in low (5 mM) or high (5 mM) glucose medium for 3 d before starting the proliferation assay. HUVECs in each low or high glucose condition were seeded (2 × 10⁴ cells per well) at equal density on 96-well plates precoated with collagen. Cell proliferation was determined by the MTS tetrazolium compound assay. High and low glucose cells were incubated with MTS solution for 3 h, then the reaction was stopped by adding 10% SDS. The absorbance was measured at 490 nm using a microplate reader. The average optical density of six wells was calculated and normalized by their corresponding day 1 average optical density. Presented results are representative of three independent experiments.

C2C12.

C2C12 myoblasts (ATCC, passage 4–9) were used for all experiments. Cells were maintained and passaged in DMEM (4.5 g l⁻¹ glucose) containing 10% FBS and 1% penicillin/streptomycin. Before differentiation, cells were resuspended in growth medium (DMEM, 10% FBS, 1% penicillin/streptomycin) with low (1 g l⁻¹, 5 mM) or high (4.5 g l⁻¹, 25 mM) glucose concentration and plated at ~75% confluence on six-well plates. Cells were allowed to adhere for 24 h, after which they were switched to differentiation medium (DMEM, 2% horse serum, 1% penicillin/streptomycin) containing low or high glucose. Cells were allowed to differentiate into myotubes for 6 d and media were changed every 48 h. mRNA levels were determined using real-time PCR, as described next.

Real-time PCR.

RNA was extracted using TRIzol reagent and purified using Direct-zol columns (Zymo). RNA samples underwent reverse transcription and complementary DNA levels were measured using target-specific primers. Primer sequences used are outlined in Table 1. Expression was normalized to β2-microglobulin as a housekeeping gene. Fold changes were calculated using the 2^−∆∆Ct method⁷⁶ and expressed relative to low glucose for each gene.

Table 1 |.

Mouse primer sequences used for PCR analysis

Gene	Forward (5′ to 3′)	Reverse (5′ to 3′)
β2m	TTCTGGTGCTTGTCTCACTGA	CAGTATGTTCGGCTTCCCATTC
Ppargc1α	TATGGAGTGACATAGAGTGTGCT	CCACTTCAATCCACCCAGAAAG
Myh1	CTCTTCCCGCTTTGGTAAGTT	CAGGAGCATTTCGATTAGATCCG
Myh4	CCGCATCTGTAGGAAGGGG	GTGACCGAATTTGTACTGAGTGT
Myh7	ACTGTCAACACTAAGAGGGTCA	TTGGATGATTTGATCTTCCAGGG
Col1a1	GCTCCTCTTAGGGGCCACT	CCACGTCTCACCATTGGGG
Col3a1	CTGTAACATGGAAACTGGGGAAA	CCATAGCTGAACTGAAAACCACC
Col4a1	CTGGCACAAAAGGGACGAG	ACGTGGCCGAGAATTTCACC
Col6a1	GGATCTATTCTTCGTCGAC	TCTCAGGTTGTCAATGAAGCG
Col8a1	ACTCTGTCAGACTCATTCAGGC	CAAAGGCATGTGAGGGACTTG
Mstn	AAGATGACGATTATCACGCTACC	CCGCTTGCATTAGAAAGTCAGA

Human participants.

Twenty-four participants (age, mean ± s.d. 24.5 ± 6.57 years) were recruited to Joslin Diabetes Center (11 male and 13 female). All participants were screened for the absence of diabetes (mean HbA1c 5.35 ± 0.06%), and included on the basis of BMI ≥ 26 and ≤35 kg m⁻² (mean BMI 29.55 ± 2.83 kg m⁻²). This relatively narrow BMI range in the overweight to obese categories was included to control for large differences in BMI, while ensuring a wide distribution of glucose tolerance values, which would probably not be present had only lean participants been recruited. Participants were excluded if they: did physical exercise >150 min of moderate-intensity weekly, experienced current or recent large-scale weight loss, were pregnant, had a known chronic disease or condition precluding aerobic exercise, were taking beta blockers or anticoagulants or had donated blood in the last 3 months. Participants were asked to refrain from strenuous exercise for 24 h before each study visit.

On visit 1, participants arrived after an overnight fast for an oral GTT (75 g), and blood samples were drawn at regular intervals over 2 h to measure circulating glucose and insulin. For visit 2, all participants fasted overnight and consumed a standardized meal (Glucerna 1.5Cal) before undertaking an incremental cycling test to volitional fatigue, to determine peak oxygen uptake (VO_2peak, 29.42 ± 6.93 ml kg⁻¹ min⁻²). On visit 3, all participants arrived overnight fasted and consumed a standardized meal before completing an acute bout of moderate aerobic cycling exercise for 45 min at 60% of previously determined individual VO_2peak. Skeletal muscle microbiopsies (catalogue no. TT1411 CareFusion, Temno Evolution) were taken from contralateral vastus lateralis muscles at similar anatomical positions, before and immediately after acute exercise. Muscle samples were removed from the needle sampling window, and stored in Allprotect reagent (Qiagen) at −80 °C before being processed for western blotting to assess exercise-induced activated of JNK and SMAD signalling. Signalling data are shown from n = 22 participants due to insufficient muscle sample obtained from two participants. Written informed consent was obtained for all participants enroled. The study and its procedures were approved by the Joslin Diabetes Center Committee on Human Studies.

Statistics and calculations.

All data are presented as individual data points. Bars are mean ± s.e.m. In the case of two group comparisons, an unpaired, two-tailed t-test with Welch’s correction was used. For three independent groups being compared (for example, Fig. 1), a one-way analysis of variance (ANOVA) was used to determine the main effect of glycemia with Holm–Sidak post hoc testing. For all comparisons between sedentary and exercise-trained groups, a two-way ANOVA was used to determine main effects of glycemia or training, with Holm–Sidak post hoc testing to test statistical differences within each group (that is, CON, WD or STZ) between sedentary and trained conditions. Pearson or Spearman correlation coefficients were calculated where appropriate. In all analyses, significance was accepted as P < 0.05.

Glucose tolerance and insulin sensitivity in human participants.

Glucose AUC was calculated using GraphPad Prism 8 (GraphPad Software). Following methods from Solomon et al.²², oral GTT- (OGTT-)derived insulin sensitivity was calculated as published, as a function of BMI, G₁₂₀ (glucose at 120 min) and I₃₀ and I₉₀ (insulin at 30 and 90 min) during the OGTT²², calculated according to equation (1):

$$\begin{gathered} {\text{Si}}_{OGTT}\left(\mathrm{\mu }\text{mol}\hspace{0.17em}{\text{kg}}^{-1}\hspace{0.17em}{\min }^{-1}\hspace{0.17em}{\text{pmol}}^{-1}\times l\right)=0.138-\left(0.00231\times \text{BMI}\right) \\ -\left(0.00118\times {\text{G}}_{120}\right)-\left(0.0000135\times {\text{I}}_{30}\right)-\left(0.00000678\times {\text{I}}_{90}\right) \end{gathered}$$ (1)

Reporting Summary.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1 |. 24 h wheel running behaviour.

A cohort of mice in control (CON, n=4), Western Diet-fed (WD, n=4) and Streptozotocin (STZ, n=4) groups undergoing exercise training were placed in cages to assess wheel running patterns. Wheel revolutions were counted per 1 h interval and recorded. a, An average trace over 24 h is shown; and mean for time spent running in the dark (%) is shown in Table b. Data are shown as mean ± SEM. Group differences in time spent dark running were compared by one-way ANOVA.

Extended Data Fig. 2 |. Baseline exercise capacity.

Three separate cohorts of CD-1 mice were fed with Western Diet (WD), injected with streptozotocin (STZ) or maintained on a control diet (CON). After 8 weeks of treatment, all mice underwent aerobic exercise capacity testing in a and were found to have similar exercise capacities prior to being allocated to treatment groups for the training intervention (CON n=34, WD n=32, STZ n=42). Mice from each treatment group (CON, WD, STZ) were then allocated to remain sedentary or undergo voluntary wheel running (exercise-training) for a further 8 weeks. Baseline (pretraining) exercise capacity, shown in b, was similar among all six treatment groups (CON SED n=17, WD SED n=16, STZ SED n=22; CON EXT n=17, WD EXT n=16, STZ EXT n=20). Main effects were determined by one-way ANOVA relative to CON group in a and by two-way ANOVA in b. Data is represented as a point for the result of each individual animal, or mean ± SEM.

Extended Data Fig. 3 |. JNK activation with treadmill running during acute and chronic hyperglycaemia.

a, To determine whether JNK activation with exercise is due to hyperglycemia, NOD mice completed an acute exercise bout (AEX n=18; 30 min treadmill running) or remained sedentary (SED n=6) and JNK signaling was measured in gastrocnemius muscle to determine correlation with random blood glucose. Representative blots (n= 1 SED; n= 2 AEX) shown here correspond with data shown in Figure 6g. To determine whether JNK activation with exercise is affected by acute increases in blood glucose, CD-1 mice were maintained on a control chow diet and fasted for 2 h prior to being split into four groups: a) control sedentary (n=5), b) acute glucose sedentary (n=5), c) control 30 min exercise (n=4) and d) acute glucose 30 min exercise (n=5). Mice allocated to glucose treatment were injected with 3 g/kg glucose to rapidly increase circulating glucose levels. b, JNK is activated with acute exercise (AEX) in gastrocnemius muscle, but acute glucose elevation does not alter JNK signaling with moderate treadmill running. c, Relative increases in blood glucose in control animals versus mice injected with 3 g/kg glucose. d, Mice were maintained on control diet (CON) or Western Diet (WD) for 8 (CON n=4; WD n=5), 16 (CON n=10; WD n=7), or 32 weeks (CON n=9, WD n=10) and an acute running experiment was performed at each time point. JNK activation with moderate treadmill running (30 min) was significantly higher in WD mice vs. CON mice by 16 weeks and worsened by 32 weeks; (right) representative blots showing n= 2/group per time point. Data is represented as mean ± SEM in all panels. Main effects were determined by two-way ANOVA in a and c. In d, differences in JNK activation between groups and over time were determined by two-way repeated measured ANOVA and indicated with “P”; differences between CON and WD groups at each 16 and 32 wk timepoint are indicated with “p.”.

Data availability

Source data for western blots are available online. All other data that support the findings of this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.