Divergent Skeletal Muscle Metabolomic Signatures of Different Exercise Training Modes Independently Predict Cardiometabolic Risk Factors

The prevalence of type 2 diabetes (T2D) in the United States has now reached ∼10% of the population (1) and costs nearly $300 billion annually in health care (2). Insulin resistance is a pivotal contributor to the development of T2D and confers many negative health consequences, even before the onset of T2D (3,4). Both aerobic and resistance training (RT) exercise are effective strategies for enhancing SI in both young and older individuals (5,6) and may reverse many of the negative outcomes of insulin resistance.

Aerobic exercise training is characterized by relatively recurrent and repeated low-intensity muscle contractions, requiring aerobic metabolic pathways for the generation of ATP. RT is characterized by relatively rapid muscle contractions against a large weight load (resistance), requiring immediate ATP-producing metabolic pathways. Both aerobic exercise training and RT equally enhance SI (5), but mounting evidence suggests that the mechanisms by which these different exercise stimuli enhance SI are distinct. For example, aerobic exercise training has been shown to concomitantly enhance skeletal muscle SI and mitochondrial function (7), whereas RT has minimal impact on muscle mitochondria and aerobic metabolism (5,8,9). We (10) and others (11) have provided evidence for the role of RT-induced enhancement of muscle glycolytic metabolism as an underpinning mechanism potentially leading to improved SI. It remains to be determined how the distinct metabolic pathways related to these different exercise programs with similar impacts on SI could be deciphered using targeted and untargeted metabolomic approaches.

The overarching goal of the current study was therefore to identify skeletal muscle metabolomic predictors of improvements in SI by aerobic exercise training, RT, and combined exercise training (CT). We began by performing RNA sequencing (RNA-seq) analysis of human muscle tissue data from our previous publication, which demonstrated an insulin-sensitizing effect of different exercise training modes (5), and identified metabolic pathways linked to SI in muscle. However, because transcriptomic changes likely cannot entirely explain the exercise training–induced improvement in muscle metabolism, we performed targeted metabolomic analysis of amino and tricarboxylic acid (TCA) metabolites, acylcarnitines, and ceramides, as well as untargeted lipidomic analysis, to identify the association of various muscle metabolites with muscle SI and other cardiometabolic variables after different modes of exercise training.

Recruited participants were between ages 18 and 30 or 65 and 80 years, had a BMI ≤32 kg/m², did not regularly exercise >2 days per week for >20 min per session, and were generally healthy (fasting blood glucose ≤110 mg/dL, did not have cardiovascular disease, nonsmokers) (Supplementary Table 1). All other exclusion criteria are provided in our previous report (5). All participants underwent baseline testing, which included hyperinsulinemic–euglycemic clamp, DEXA scan, Vo_2max test, strength test, and vastus lateralis muscle biopsies. Participants then performed 12 weeks of either aerobic exercise training (high-intensity interval training [HIIT]; n = 19), RT (n = 18), or CT (n = 15) and repeated baseline testing, which indicated improved cardiometabolic parameters across all training groups and ages (Table 1 and Supplementary Table 2). After 12 weeks of exercise training, all follow-up tests were performed 72 h after the last bout of exercise. The full details of all study visits and exercise training sessions are provided in the online Supplementary Material and our previous report (5). The study design was approved by the Mayo Clinic Institutional Review Board, and participants were informed of study procedures and provided written informed consent.

Muscle gene expression differences in response to various exercise training modes were assessed using RNA-seq technology, as previously described (5). RNA-seq analysis details are provided in the Supplementary Material.

Amino metabolites were measured by liquid chromatography–mass spectrometry (LC-MS), TCA analytes were measured by gas chromatography–MS, acyl carnitines (specifically C0–C18:1) were measured by LC-MS, and tissue ceramides, sphinganine, sphingosine, and sphingosine-1-phosphate were measured by electrospray ionization–MS/MS. Full details of targeted metabolomic analyses can be found in the Supplementary Material. Amino metabolites were measured from muscle samples from all participants in the HIIT and RT groups (Supplementary Table 3). TCA analytes were not measured from five participants because there was no more tissue sample remaining after all other previous analyses. Therefore, both the HIIT (men n = 10; women n = 6) and RT (men n = 9; women n = 7) groups had 16 participants, respectively, for TCA analyses (Supplementary Table 3). All acylcarnitine values from one sample (young woman from HIIT group) fell below the standard curve because of sample processing issues; therefore, the number for the HIIT samples for acylcarnitines is 18 rather than 19 (Supplementary Table 3). Ceramides were not measured from four participants because there was no more tissue sample remaining after all other previous analyses. Therefore, the HIIT (men n = 10; women n = 7) and RT (men n = 9; women n = 7) groups had 17 and 16 participants, respectively, for ceramide analyses (Supplementary Table 3).

Full methodological details of untargeted lipidomics performed by LC/MS are provided in the Supplementary Material. Untargeted lipidomics was not performed for one HIIT participant and five RT participants because of sample limitations after previous analyses. Therefore, the HIIT (men n = 11; women n = 7) and RT (men n = 7; women n = 6) groups had 18 and 13 participants, respectively, for untargeted lipidomics (Supplementary Table 3).

Approximately 10–15 mg frozen muscle tissue was homogenized and used for immunoblotting to measure branched-chain ketoacid dehydrogenase (BCKDH) abundance. Complete immunoblotting procedures are provided in the Supplementary Material. For immunoblotting, a subset of 14 participants in the HIIT group (men n = 8; women n = 6), 14 participants in the RT group (men n = 7; women n = 7), and 13 participants in the CT group (men n = 6; women n = 7) was used (Supplementary Table 3) because of sample limitations after analysis in our initial study (5). All full-length blots are shown in Supplementary Fig. 1.

Data for immunoblotting, amino metabolites, and TCA metabolites are expressed as mean ± SEM, with two-tailed significance levels of α < 0.05. Paired two-tailed Student t tests were used to determine the exercise training effect on BCKDH protein abundance and phosphorylation. Two-way ANOVAs were run to identify the influence of training (pre vs. post) and exercise type (HIIT vs. RT) on all targeted metabolites. Analyses were performed using Prism 9 software (GraphPad, San Diego, CA).

This study involved two different types of metabolite measurements: 1) targeted quantitative measurements of amino acids (AAs), TCA metabolites, acylcarnitines, and ceramides and 2) untargeted qualitative measurements of lipids (lipidomics). Measurements from each cohort (i.e., each combination of age group and training modality) were collated. Next, we assessed the association of each targeted metabolite with each phenotypic variable of interest using a generalized linear regression with random effects framework. For this, a linear regression model with metabolite as the dependent variable and time point (postexercise minus pre-exercise), participant ID, and phenotypic variable as independent variables was built. A null model was also built by removing the phenotypic variable. ANOVA was used to compare the full model (with phenotypic variable) with the null model (without phenotypic variable). A metabolite was considered significantly associated with a phenotypic variable if the resulting ANOVA had a P value ≤0.001. In parallel, we also computed the Spearman correlation between Δ change (postexercise minus pre-exercise) of each metabolite with Δ change of each phenotypic variable (to meaningfully display the relationship between each metabolite and phenotypic variable, because regression coefficients of models are not comparable across metabolite–phenotype pairs). We used a similar but slightly modified method for assessing the significance of the relationship between phenotypic variables and lipidomic data. The intensities of lipidomic data were first normalized using the trimmed means of M values method and log2 transformed. The normalized and transformed intensities were subjected to the same modeling method described above.

We used weighted correlation network analysis (WCNA) (12) to integrate targeted quantitative measurements (described above) with muscle gene expression. To accomplish this, gene expression data were normalized using the trimmed mean of M values method and log2 transformed. Normalized and transformed gene expressions were combined with metabolite data; Δ abundance (postexercise minus pre-exercise) was computed for each molecule (gene and metabolite). Resulting data were filtered to remove molecules with a coefficient of variation <20%. The remaining gene expression and metabolite abundance data were combined and z score transformed. Signed adjacency matrices were constructed from the metabolite and gene expression biweight midcorrelations using a soft-thresholding power of 19. Average linkage hierarchical clustering was computed for the dissimilarity matrices using unsigned topological overlap matrices and cutreeDynamic using a minClusterSize of 30, deepSplit of 2, and cutHeight of 0.99. Modules with a similarity score ≥0.8 were combined. Remaining module eigen features were extracted, and Pearson correlation with Δ phenotypic variables was computed individually for each of the age and training type groups. Transcripts (gene symbols) from significantly correlated clusters to phenotypic variables were subjected to pathway analysis (overrepresentation analysis) and cellular component analysis to make biological interpretations. WebGestalt (Web-based Gene Set Analysis Toolkit) software (https://www.webgestalt.org) was applied for pathway analysis using the Reactome database. Cellular component category analysis (GOTERM_CC_4) was performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) software (https://david.ncifcrf.gov/home.jsp). Pathways or cellular components with a false discovery rate <0.05 were considered significant.

RNA-seq data have been deposited in Gene Expression Omnibus under accession number GSE97084. Specific metabolomic data and group and individual values are provided in the Supplementary Material. All other data sets generated and/or analyzed in this study are available from the corresponding author upon request.

In our previous study, we found that either HIIT, RT, or CT resulted in greater peripheral SI (5), but the metabolic pathways of these different exercise training modes were not identified. Using RNA-seq data from muscle biopsies after HIIT, RT, and CT, we performed univariate analysis correlating gene transcript expression with SI and found that the only gene pathways associated with exercise-induced SI were those of AA metabolism (Fig. 1A). Specifically, gene set enrichment analysis revealed that lysine metabolism–(Fig. 1A) as well branched-chain AA (BCAA) metabolism–related genes were positively associated with the exercise training–induced increase in SI (Fig. 1A–C). To identify specific gene pathways associated with SI induced by different exercise training modes (HIIT vs. RT vs. CT), we performed multivariate analysis and found that SI induced by either HIIT or RT was positively associated with AA metabolism in general and with BCAA metabolism specifically, but only HIIT-induced SI was associated with lysine metabolism (Fig. 1D). We also found that CT-induced SI was not significantly associated with any gene pathways (Fig. 1D). Although CT resulted in enhanced SI in the young cohort, no significant increase in SI was observed in the older cohort (Supplementary Table 2), which we previously hypothesized was due to a lower intensity/volume of both aerobic exercise training and RT, respectively, during CT (5). Because the BCAA metabolism gene pathway was positively associated with SI after either HIIT or RT, we measured the protein abundance and phosphorylation of BCKDH, the rate-limiting enzyme of BCAA metabolism, in skeletal muscle before and after HIIT, RT, and CT. HIIT significantly increased the abundance of BCKDH in muscle (Fig. 1E and G), indicating greater capacity of BCAA metabolism after HIIT. BCKDH phosphorylation, a marker of BCKDH activity, was increased by HIIT and CT. The ratio of phosphorylated to total BCKDH was unchanged in all groups (Supplementary Fig. 1).

A strong association between elevated plasma BCAAs and insulin resistance has been previously reported (13,14), suggesting impaired BCAA metabolism in insulin-resistant peripheral tissues and the contribution of BCAAs to insulin resistance. However, plasma BCAAs are consistently shown to be unresponsive after exercise training (15–17), likely because the plasma amino metabolites represent the net balance of exchanges between muscle and liver (18). We confirmed that plasma BCAAs before and after HIIT or RT were unaltered (Supplementary Table 4), indicating that the significant association between muscle BCAA metabolism gene pathways and SI in muscle after exercise training detected in the current study was not reflected by changes in the plasma BCAAs.

Efficient mitochondrial function to produce ATP by the TCA cycle and electron transport chain is critical to sustain moderate to intense exercise. Glucose and fatty acids have been known to provide fuel for mitochondria to meet energy needs during exercise (19), and multiple AAs can also be used as substrates in the TCA cycle in muscle, generating electrons to be transported to inner mitochondrial electron transport and ATP production (20). We previously reported a more profound effect of HIIT on muscle mitochondrial respiration and mitochondrial biogenesis (based on transcriptome, synthesis rate, and protein content) compared with RT (Supplementary Table 2) (5). Twelve weeks of HIIT and CT induced a significant increase in the concentrations of TCA intermediates, including citrate, ketoglutarate, 2-hydroxyglutarate, fumarate, and malate (Fig. 2 and Supplementary Table 5). Additionally, the concentration of succinate as well as that of multiple amino metabolites, including all BCAAs, glutamate, phenylalanine, γ-aminobutyric acid, aspartate, asparagine, methionine, and serine, was significantly increased in muscle with HIIT but not with RT or CT (Fig. 2 and Supplementary Table 5). Lysine, citrulline, sarcosine, and 1-methylhistidine concentrations decreased after HIIT. In contrast, RT did not significantly increase the concentration of any TCA intermediate or amino metabolite in muscle, and CT reduced the concentration of sarcosine (Fig. 2 and Supplementary Table 5). These data suggest that HIIT results in enhanced muscle TCA and amino metabolite availability to support metabolically demanding bouts of HIIT exercise, CT enhanced only TCA but not amino metabolites, and RT did not increase TCA or amino metabolite concentrations, potentially because of the lower metabolic fuel requirements of RT. Of importance, these are training effects and not acute exercise effects, which may be different.

It has been reported that skeletal muscle concentrations of acylcarnitines are associated with altered insulin resistance and T2D (21), but it remains to be determined whether exercise-induced improvements in SI after HIIT, RT, and CT correspond with changes in acylcarnitine concentrations in muscle. Muscle concentrations of long- and medium-chain acylcarnitines were significantly increased after 12 weeks of HIIT, but the concentrations of free carnitine (C0) and the BCAA-derived short-chain C3 and C5 acylcarnitines were significantly reduced after HIIT (Fig. 3A–D). Although RT and CT had no effect on free carnitine, the effects of RT and CT on C3 and C5 acylcarnitines were similar to those of HIIT (Fig. 3A and B), suggesting a common impact of different types of exercise training on BCAA metabolism. Like HIIT, CT increased the concentrations of all identified medium and long chain acylcarnitines, but RT had no effect on these longer acylcarnitine species (Fig. 3C and D). Together, these results demonstrate that overall, HIIT and CT have a more robust impact on acylcarnitines than RT.

There is substantial support for the idea that ectopic accumulation of fatty acids and lipid intermediates in muscle is a strong contributor to obesity-associated insulin resistance (22,23). Ceramides are sphingolipids composed of a fatty acid linked to a sphingoid base, and they have been strongly linked to antagonism of insulin signaling and increased inflammatory signaling (24). In individuals with obesity, aerobic exercise training reduces muscle ceramide content (25). However, less is known about the impact of exercise training on ceramides in lean individuals without insulin resistance (26). Moreover, to our knowledge, only one report has evaluated the impact of RT on muscle ceramide content (27). Therefore, we assessed the abundance of muscle ceramides after 12 weeks of HIIT or RT in relatively lean healthy individuals. In general, the abundance of ceramide species in muscle was increased with either HIIT or RT (Fig. 4A–C), and although significance was not reached for any ceramide species with CT, most were trending in the upward direction after CT. Specifically, both HIIT and RT increased the abundance of C14 and C24 ceramides, whereas C16 and C18 ceramides were only increased by HIIT, and sphingosine and C22 ceramides were only increased by RT (Fig. 4A–C). Contrary to what has been observed in individuals with obesity, muscle ceramide abundance is increased after either HIIT or RT in healthy lean individuals.

Because our study cohort consisted of older (age 65–80 years) and younger (age 18–30 years) cohorts, we separated the analyses of targeted metabolites by age group. We found that age group did not have an impact on the effect of exercise training on targeted muscle metabolite (organic acids, AAs, acylcarnitines, or ceramides) concentration (Supplementary Tables 6–8). We next split our analyses by biological sex. Interestingly, women had greater increases after HIIT than men in several targeted muscle metabolites, including glutamate, glutamine, anserine, histidine, 3-methylhistidine, ethanolamine, and C18 ceramide (Supplementary Table 9). Women also had less of a decrease than men after HIIT in citrulline and propionylcarnitine (C3). After RT, women had a greater increase than men in the TCA metabolite ketoglutarate but a lower response in cis-aconitate (Supplementary Table 10). With CT, similar changes in metabolites were observed in men and women (Supplementary Table 11).

Various lipid species have been identified and may contribute to SI (28). We used an untargeted lipidomic approach to gain a broader understanding of the contributions of lipid species to muscle metabolism after exercise training. Of the 540 different lipid species identified, 113 were triglyceride (TG) species, 104 were phosphatidylcholine (PC) species, 63 were phosphatidylethanolamine (PE) species, 50 were acylcarnitine species, 31 were sphingomyelin species, 31 were fatty acyl species, 27 were diglyceride species, and 31 were lyso-PC species; all other lipids had <30 identified species (Fig. 5A).

The muscle concentrations of many lipid species were altered by either HIIT, RT, or CT (Fig. 5B–D). Although HIIT increased some and decreased other lipid species (Fig. 5B), RT and CT resulted in almost exclusively increased lipid species abundance (Fig. 5C and D). Each exercise mode seemed to differentially affect the abundance of various lipid classes, but the primary class of lipids affected by each exercise mode was phospholipids (Fig. 5E). HIIT robustly increased cardiolipin (CL) species, and CT moderately increased CLs, but only two CLs were increased by RT (Fig. 5F). In agreement with our targeted analysis in Fig. 3, a greater number of acylcarnitines (by untargeted lipidomics) were increased with HIIT and CT than with RT (Fig. 5E). Finally, whereas HIIT reduced the concentration of PC and increased the concentration of PE, RT and CT increased both PC and PE (Fig. 5F).

We determined the associations between the change in targeted metabolite concentrations with each mode of exercise training and multiple cardiometabolic parameters, including SI, aerobic capacity, mitochondrial protein synthesis, fat-free mass, strength, and mitochondrial respiration (Fig. 6A). Most categories of metabolites did not show a clear and consistent association between the changes in cardiometabolic parameters across all exercise training modes in either young or older individuals. Of interest, most ceramide species tended to be negatively associated with the exercise-induced changes in mitochondrial protein synthesis rate. In the young cohort, BCAAs were positively associated with mitochondrial respiration in the HIIT and CT groups but not in the RT group, consistent with changes in mitochondrial respiration (Supplementary Table 2).

Next, we examined the associations between changes in untargeted lipid metabolites after exercise training with changes in cardiometabolic parameters (Fig. 6B). Because untargeted lipidomics in the CT group was assessed in a different batch from the HIIT and RT groups, there were different numbers of identified lipid species within each lipid class, indicated by dashed gray lines. However, we were able to detect the change among the lipid species within each exercise mode. The most prominent finding was an age effect on the association between untargeted lipid metabolites and physiological parameters after exercise training (Fig. 6B). For example, in the young cohort, acylcarnitines, diacylglycerols, and phospholipids species were negatively associated with mitochondrial protein fractional synthesis rate after RT, but these same lipid species were positively associated with mitochondrial protein fractional synthesis rate in the older cohort after RT. The divergence in metabolite association with physiological parameters depending on exercise type and age emphasizes the complexity of muscle metabolism and exercise training.

WCNA revealed that 17 clusters of muscle gene transcripts and metabolites (modules) could be identified from our data set (Fig. 7A). Among these modules, 16 were associated with the exercise-induced change in at least one cardiometabolic parameter within one of the exercise training groups (HIIT, RT, or CT in either young or older cohort). Some modules did not contain any metabolites, such as the dark red and green/yellow. The dark red module, which included genes involved in muscle contraction pathways and vesicle/cytosolic compartments, was positively associated with fat-free mass in the young RT group (Fig. 7B). The green/yellow module also included genes involved in muscle contraction as well as multiple muscle microstructure compartments and was positively associated with SI in the older RT group but was negatively associated with Vo_2max in the young HIIT group (Fig. 7C). The gray module included all BCAAs and aromatic AAs and was positively associated with muscle strength but negatively associated with SI in the older CT group (Fig. 7D). The light cyan module included gene pathways involved in lipid metabolism and the mitochondrial compartment and multiple acylcarnitine metabolites. Interestingly, the light cyan module was positively associated with SI in two groups (young HIIT and older CT), with mitochondrial protein synthesis in the young HIIT group, and with fat-free mass in the young RT group (Fig. 7E). The turquoise module included genes involved in multiple signaling pathways in the cytosolic region and multiple amino metabolites, especially some that are highly involved in cellular signaling through posttranslational modifications, such as serine and threonine. The turquoise module was negatively association with SI and positively associated with fat-free mass in the older CT group and negatively associated with mitochondrial respiration in the older RT group (Fig. 7F).

Exercise training promotes SI and other markers of cardiometabolic health, but the potential impact of altered skeletal muscle metabolism with different modalities of exercise training on improved cardiometabolic health remains to be fully understood. Here, we first identified based on skeletal muscle gene pathways that AA metabolism, specifically BCAA metabolism with RT and HIIT and lysine metabolism with HIIT, is significantly associated with exercise training–induced enhancement of SI. Measurement of targeted metabolites (AAs, TCA intermediates, acylcarnitines, and ceramides) and untargeted lipidomics in muscle revealed that the predominant metabolic pathways altered by HIIT and RT are distinct, and although CT involved lower-intensity forms of both HIIT and RT, the alterations in metabolic pathways were also distinct from HIIT and RT. Furthermore, we found that age and sex also influenced the association between muscle metabolites and the exercise-induced enhancement of cardiometabolic markers.

Increased appearance of plasma BCAAs has been proposed as an indicator of mitochondrial overload and an inability to efficiently metabolize fuels in individuals with obesity and T2D (29). Here, we find that concurrently with enhancement of SI and mitochondrial respiration after 3 months of HIIT, BCAA concentrations were higher in muscle but not in plasma. The differences in plasma BCAA concentrations may vary from those in tissue concentrations because the appearance of BCAAs in plasma reflects not only metabolism in muscle but also the net appearance of BCAAs from different tissues in the body, especially the liver and other splanchnic tissues (30). The liver is the main site of oxidation of BCAA transamination products (e.g., keto isocaproate) (31) and many other aromatic and glucogenic AAs, which may explain why BCAAs and other amino metabolites are different in muscle and plasma after exercise training. Muscle BCKDH abundance was also increased with HIIT but not with RT, whereas the phosphorylation of BBCKDH was increased with both HIIT and CT, supporting that aerobic exercise is important to enhance BCAA metabolism in muscle. These metabolic changes in muscle AAs occurred concurrently with an increase in mitochondrial transcripts, protein synthesis (representing translation), and respiration (function), as well as CL levels, indicating that aerobic exercise, especially HIIT, expands mitochondrial biogenesis, content, and function, potentially enabling muscle capacity to metabolize and store BCAAs. Another group previously reported that CT has no impact on muscle BCAA concentration in individuals with BMI >25 kg/m² (15), consistent with the current study, in which we found higher muscle BCAA concentrations with HIIT but not with CT.

Supporting the concept that HIIT increases mitochondrial capacity and substrate turnover, TCA cycle intermediates were elevated in muscle after 12 weeks of HIIT, indicating enhanced citric acid cycle. The HIIT-induced increase in muscle glutamate and aspartate also suggests greater anaplerosis via conversion to α-ketoglutarate and oxaloacetate, respectively. Of interest, BCKDH is involved in transamination of BCAAs and plays a role in maintaining glutamine–glutamate levels (32). CT also increased TCA intermediate concentration, suggesting greater mitochondrial capacity. On the other hand, we found that RT, which does not significantly enhance mitochondrial respiration, did not increase the concentration of TCA intermediates (or any amino metabolites). These results are consistent with the notion that the higher aerobic energy cost of HIIT and CT compared with RT enhances citric acid flux and anaplerosis, potentially facilitated by a greater mitochondrial volume after HIIT and CT but not after RT. It is known that RT derives energy cost primarily through anaerobic glycolysis (10).

Although RT did not increase amino or TCA metabolites in muscle (Fig. 2), we observed decreased concentrations of C3 and C5 acylcarnitines, derivatives of incomplete BCAA metabolism, which are often higher in patients with obesity or T2D with insulin resistance (14,33). These results suggest that RT may also to a lesser degree enhance muscle capacity to metabolize BCAAs, but incompletely, even though phosphorylated BCKDH and BCKDH abundance were unaltered with RT, unlike with HIIT. A potential reason for the discrepancy between BCKDH phosphorylation and BCAA metabolism with CT versus HIIT is the magnitude of increased mitochondrial function and volume that occurred with HIIT compared with CT, resulting in less impact on metabolites in general that can feed into the TCA cycle.

Of interest, after HIIT, the lysine metabolism gene pathway was positively associated with greater SI (Fig. 1D). We found that muscle lysine concentration was reduced with HIIT but that α-amino adipic acid (AAAA), a metabolic product of lysine catabolism, was unaltered with HIIT (Fig. 2). Because lysine conversion to AAAA occurs in the liver (18), and AAAA concentration was unchanged by HIIT, these data suggest that the lower muscle lysine concentration with HIIT was likely due to either increased incorporation of lysine into proteins or decreased release of lysine from proteins (or a combination of both) (34).

Other amino metabolites with potential interactions with the brain were altered with HIIT. Sarcosine, a metabolic product of glycine, was decreased with HIIT and CT, indicating an impact of aerobic training, although at different intensities, with HIIT and CT. Sarcosine is interconvertible with glycine and plays an important role in making glycine available for brain synapse function (35). Furthermore, an HIIT-induced increase in γ-aminobutyric acid, a metabolic product of glutamine and inhibitory neurotransmitter in the brain and spinal cord (36), was also noted. Future investigations linking the exercise-induced changes in these muscle metabolites with cardiovascular and brain metabolism/function are warranted.

In 2001, Goodpaster et al. (37) reported a so-called athlete’s paradox, in which both exercise-trained insulin-sensitive individuals and individuals with T2D had elevated skeletal muscle lipid content. The greater capacity of exercise-trained muscle to oxidize fatty acids has been offered as an explanation for the capability of exercise-trained muscle to retain SI while also increasing substrate storage (38). In the current study, we found that the concentrations of multiple lipid species were increased in muscle without a concurrent increase in fatty acids with either HIIT, RT, or CT, indicating increased capacity to store and metabolize fat. During excess fuel availability, incomplete β oxidation and inadequate TCA cycle activity have been postulated to be the potential causes of greater appearance of acylcarnitines and other lipid intermediates in muscle in individuals with T2D (39), and increased acylcarnitines or other lipid species in muscle are reported to be associated with insulin resistance (33,40). Using targeted metabolomics, the current study measured a significant increase in medium- and long-chain acylcarnitines in muscle after HIIT or CT and increased ceramides after both HIIT and RT. However, RT did not alter muscle concentration of medium- or long-chain acylcarnitines. Together, these results support the concept that greater mitochondrial function results in greater muscle transport of fatty acids, β oxidation, and ATP production. WCNA revealed a cluster of genes involved in lipid metabolism and mitochondrial compartments involving multiple acylcarnitine species, which are linked to enhanced SI after HIIT or CT, providing evidence for a role of acylcarnitines to contribute to mitochondrial-related impacts on SI.

A decrease in muscle TG concentration after exercise training is typically observed in individuals with T2D (41), but in the current study, our untargeted lipidomics in lean healthy individuals show that muscle TG concentrations were unchanged by HIIT, RT, and CT. A disassociation between exercise-induced decrease in TGs and SI has been reported in healthy individuals (41), and the current study in healthy participants without insulin resistance agrees with the above report. Mitochondrial membrane phospholipids have been linked to aerobic capacity (42) and SI (43), and the abundance of multiple phospholipids was significantly altered by HIIT, RT, and CT, concurrently with increased SI, except in CT in older individuals. PC is the most abundant phospholipid on both the inner and outer mitochondrial membranes (42), and alterations in PC can lead to altered mitochondrial function (44,45). We found that although RT and CT increased PC species, there was a significant reduction in PC species with HIIT. Therefore, we speculate that the differential regulation of PC after HIIT or CT may indicate disparate mitochondrial maintenance/turnover between the different exercise modes. Additionally, it has been shown that aerobic exercise training decreases the ratio of PC to PE and increases PE concentration (46), which is in agreement with the current study. Like HIIT, RT and CT also robustly increased PE concentration. PE has many functions, including membrane fusion and autophagy (47), processes that are important for muscle protein turnover (48). Although not as abundant as PC or PE, CLs are important phospholipids that have a far greater increase after HIIT than after RT. CLs are specifically located in the inner mitochondrial membrane and play a critical role in the maintenance of mitochondrial membrane potential, enzyme activity, and Ca²⁺ homeostasis (49,50). Increased abundance of CLs has been suggested to be sufficient for increased electron transport chain activity and mitochondrial respiration (42). The potentially important role of muscle phospholipid maintenance with exercise training will need to be further studied to gain a better understanding of how exercise influences these critical lipids during different conditions (e.g., aging, metabolic disease).

In the current study, we had older and younger cohorts. We segregated the effect of exercise on each metabolite by age group, and although age influenced cardiometabolic parameters with exercise training, we did not find a significant age effect on targeted metabolites in response to exercise training. However, when we segregated our cohort by biological sex, there were several metabolites that were differentially regulated by HIIT in women versus men. For instance, we found that muscle glutamate after HIIT was increased in women but was unchanged in men. Glutamate can be synthesized from multiple AAs and used in the TCA cycle after conversion to α-ketoglutarate (51). Sex differences in glutamate concentration after HIIT suggest a potentially greater adaptation to use this metabolite as fuel in exercise-trained women. Another interesting sex difference that we observed was that after HIIT, there was a significant sex effect on 3-methylhistidine concentration. Although neither men nor women independently had a significant HIIT effect on 3-methylhistidine, men tended to have decreased 3-methylhistidine and women tended to have increased 3-methylhistidine after HIIT. Urinary 3-methylhistidine has been used as a marker for myofibrillar protein breakdown (52), and differences in muscle 3-methylhistidine may suggest that women and men have a divergent muscle contractile protein breakdown in response to HIIT, potentially explaining sex differences in muscle mass responses to exercise training (53). Considering women have a greater muscle protein synthesis rate but lower muscle mass than men (54), it is logical that women also have a greater increase in protein breakdown in response to HIIT than men. The observed sex differences in muscle metabolites in response to HIIT are interesting and should warrant further investigation in larger cohorts.

Exercise training is an excellent therapy to combat insulin resistance and other cardiometabolic disorders. Different exercise modes affect physiological parameters by some distinct and some similar metabolic pathways. Further complicating our understanding of the mechanistic underpinnings of exercise, our results suggest that the age and sex of the individual influence how exercise affects muscle metabolism. In line with the current American College of Sports Medicine exercise guidelines, our results support that performing both HIIT and RT is critical for improving all cardiometabolic parameters. Although lower-intensity CT has many cardiometabolic improvements, it seems that the intensity of the exercise likely contributes greatly to the metabolic changes in muscle, especially in older individuals.

A limitation to the current study is the lack of a middle-aged cohort. The current study was able to make comparisons between young and old participants, although limited by the relatively smaller number of participants. Women in the current study were either pre- or postmenopausal, but a middle-aged cohort of women would have been interesting to determine whether the exercise-induced metabolomic signature is different in perimenopausal women. We also did not assess whether exercise responsiveness involving both beneficial and adverse effects is related to the intensity of exercise (55,56), because the number of participants in our study was not large. Another limitation to the current study was that it was exclusively conducted in lean healthy participants, and the results cannot be translated to populations with insulin resistance, such as those with T2D or obesity. Finally, the current study determined the metabolomic responses to exercise training (72 h after the last bout of exercise). The response to an acute bout of exercise in terms of the muscle and plasma metabolome is likely different from the effect of months of exercise training.

The importance of the current study is that it represents a seminal attempt to comprehensively gauge the skeletal muscle metabolomic signatures of individuals who have undergone three different modalities of exercise training. These measurements were performed in muscle samples collected before and thereafter 3 months of supervised training, and to ensure that our measurements represented the exercise training effect (vs. acute exercise effect), the second biopsies were performed 72 h after the last bout of exercise training. Moreover, the measurements of the current study were in skeletal muscle, in contrast to previous exercise metabolomic investigations in plasma. This offered us unique insight into the metabolomic signature in the skeletal muscle, an organ (like the heart) that is involved in contractile activities related to exercise training. The metabolomic measurements of muscle clearly indicated the distinct fuels used for energy needs after 3 months of RT and HIIT and less-intense CT. The results also inform the modulating effects of age and sex on the muscle metabolome after exercise training. Of interest, another report (57) demonstrated that sex hormones and age modulate the effects of distinct exercise training modalities at transcriptional, translational, and proteomic levels. The current study clearly indicates the value of measuring the metabolome in skeletal muscle to unravel the complexities of fuel metabolism that are not fully evident from transcriptome and proteome measurements.

In conclusion, the distinct metabolic adaptations of skeletal muscle after HIIT, RT, and CT resulted in differential metabolomic signatures in muscle of healthy individuals. Although greater concentrations of multiple metabolites at baseline are conventionally believed to be negatively associated with SI, increased abundance of metabolites (AAs, organic acids, acylcarnitines, and lipid species) after HIIT did not impair SI measured by hyperinsulinemic–euglycemic clamp (predominantly representing SI in muscle) in our lean study participants. Both HIIT and RT improved metabolic capacity, especially indicated by BCAA genes, but mitochondrial adaptations only occurred after HIIT and CT. Lipidomic analysis revealed a complex relationship between muscle lipids and cardiometabolic parameters, including SI. After exercise training, although HIIT, RT, and CT all improved SI and other important cardiometabolic parameters, lipid species abundances were differentially regulated by each exercise training mode.

Acknowledgments. The authors acknowledge the effort and skilled assistance of Melissa Aakre.

Funding. This research was supported by National Institutes of Health Center for Scientific Review grants R01AG062859 and UL1 TR000135 (K.S.N.) and T32 DK007352 (M.W.P. and M.M.R.). Additional support was provided by the Mayo Foundation and the Dr. Emslander Professorship (K.S.N.). This work was also supported by the Mayo Clinic Metabolomics Core Facility, Mayo Clinic Clinical Research and Trials Unit, and Georgia Institute of Technology Systems Mass Spectrometry Core Facility.

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

Author Contributions. M.W.P., A.P.K., D.A.G., S.G.M., M.M.R., K.A.K., and K.S.N. conducted the experiments and/or acquired the data. M.W.P., S.D., and A.A.K. analyzed the data. M.W.P. and K.S.N. designed the study and wrote the manuscript. All authors reviewed the manuscript. K.S.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.