Introduction

Previous studies have consistently reported that several weeks of very low-carbohydrate intake leads to marked increases in fat oxidation at rest and during exercise beyond that achieved with endurance training alone1indicating that maximal rates of fat oxidation during exercise are achieved only when the habitual diet is very low in carbohydrates. Peak fat oxidation rates have been shown to more than double from less than 1 g/min in athletes consuming moderate-to-high-carbohydrate diets to mean values above 1.5 g/min and approaching 2 g/min in some athletes consuming lower-carbohydrate diets2,3. There is a dearth of research focused on elucidating the molecular details surrounding the cellular machinery required to support the shift away from glucose oxidation to a high flux of fatty acid and ketone metabolism in the face of low exogenous carbohydrate intake.

Keto-adaptation refers to the global process of adapting to a very low-carbohydrate diet that achieves a sustained state of nutritional ketosis4. While enhanced fat oxidation is a hallmark of keto-adaptation, the time course and full description of genetic, metabolic and physiological alterations that take place remain relatively unexplored, particularly in humans. Aspects of nutritional ketosis, such as elevated ketogenesis and circulating beta-hydroxybutyrate (BHB) levels between 0.5 and 5.0 mmol/L4, appear within the first week of starting a ketogenic diet5. In contrast, several weeks to months appear to be required for endurance exercise performance and other features such as preservation of glycogen to manifest2,6. Until recently, there was minimal research conducted on keto-adapted athletes who maintained a ketogenic diet beyond four weeks. In the Fat-Adapted Substrate Use in Trained Elite Runners (FASTER) study, we reported metabolic differences in high-level competitive ultra-endurance athletes who habitually consumed either a high-carbohydrate, low fat (HC) or a very, high-fat (LC) eating pattern2. Despite similar competitive achievements, training regimens, and maximal aerobic capacities between groups, the keto-adapted athletes demonstrated peak fat oxidation rates 2.3-fold higher (HC: 0.67 ± 0.14 g/min vs. LC: 1.54 ± 0.18 g/min)2, which occurred at a higher percentage of VO2max. Previous research consistently describes similar metabolic shifts in keto-adapted athletes7. These works often report decreased skeletal muscle glycogen stores combined with decreased carbohydrate and increased fat oxidation rates2. One unexpected finding in FASTER was that there were no differences in skeletal muscle glycogen concentrations at any time point between groups including when compared at rest, immediately following a 180-minute run (H0), and 120 min into recovery(H2)2. In contrast from previous research, this study is one of the few to investigate metabolic shifts in ultra-endurance athletes habitually consuming a ketogenic diet while competing as opposed to more short-term dietary shifts, suggesting that long-term keto-adaptation results in a remarkable conservation of muscle glycogen, most likely by repurposing carbons derived from non-carbohydrate sources such as lactate and glycerol. Additionally, evidence suggests gene expression shifts associated with carbohydrate restriction, independent of fat content, are linked to muscle glycogen levels (i.e., the lower the muscle glycogen, the more robust the gene expression patterns associated with fat oxidation)8. Given the similar muscle glycogen patterns demonstrated between low- and high-carbohydrate athletes in our previous report on FASTER2, it might be assumed that differences in gene expression would also be attenuated.

Beyond promoting high-fat oxidation rates and decreasing reliance on glucose oxidation, keto-adaptation is associated with cellular, molecular, and tissue-specific effects extending beyond substrate utilization9,10,11with unique impacts on skeletal muscle tissue and metabolism12,13,14. Much of this body of work has been investigated in murine models. For example, Ma et al. found gene expression profiles associated with fat oxidation, ketolysis, recovery, and endurance capacity were all upregulated in mice after an 8-week ketogenic diet and consistent endurance exercise regime10. Interestingly, these changes were explicitly observed in slow-twitch muscle fibers, the fibers with the highest oxidative capacity, which are fully recruited during endurance exercise. It is important to note that energy metabolism, and by extension ketoadatpation, in murine models does not mirror nutritional ketosis in humans, particularly as it relates to oxidative capacity15, limiting this body of work’s translatability. A limited number of investigations in humans have used targeted approaches to assess underlying molecular adaptations to short-term carbohydrate restriction (2–5 days) combined with acute, endurance exercise bout16,17,18but this information does not account for the distinct and uniform phenotypes observed with longer-term, chronic carbohydrate restriction (> 6 mo.). There is one study that compared circulating and skeletal muscle glucose metabolism in trained cyclists following either a low-carbohydrate, high fat diet or a more traditional, mixed diet for a minimum of 6 months. The authors reported differences in glucose tolerance molecular metabolism between the two dietary groups19 but the underlying global gene expression signatures related to the observed differences were not explored.

To shed light on gene expression patterns in response to high and low carbohydrate habitual diets and the acute response to exercise, we performed transcriptome analysis on skeletal muscle tissue obtained from participants in FASTER to probe potential differences in underlying mechanisms observed in keto-adapted ultra-endurance athletes. The purpose of this study was to determine whether coordinated molecular pathways in skeletal muscle may be revealed by differential expression of genes20 driven by dietary pattern, exercise, and/or their interaction. In contrast to a hypothesis-driven approach, this transcriptomics exploration was a discovery, hypothesis-generating analysis to help propel the field forward.

Results

Twenty highly trained and accomplished male ultra-endurance athletes participated in this study. Participant characteristics and primary metabolic results have been previously reported2. Relevant physical and metabolic characteristics are presented again here in Table 1. Athletes were habitually following their respective dietary approach for an average of 20 months (range 9–36 months) and were matched for training styles and competition performance. The primary difference between the two groups was their habitual diet patterns, which were assessed using detailed 3-day diet records. There were no differences in total calories consumed (HC (kcal): 2756 ± 638; LC (kcal): 2721 ± 700). There were significant differences in macronutrient composition between HC and LC groups, specifically higher fat and lower carbohydrate content in the LC group (Fig. 1). As previously reported, the LC group exhibited significantly higher fat oxidation rates at rest and during submaximal exercise2 (Table 1)

Table 1 Participant physical and metabolic characteristics.
Full size table
Fig. 1
figure 1

Habitual macronutrient dietary profiles for LC and HC athletes. LC athletes had significantly higher dietary fat while consuming lower dietary carbohydrate but no difference in dietary protein intake when compared to HC (p <0.05). This figure was created using BioRender.com.

Full size image

Skeletal muscle gene expression patterns

A total of 25,282 genes were reliably sequenced across all muscle samples. A group-by-time analysis of variance revealed main effects for diet in 653 (2.6%) transcripts (p < 0.01, FDR < 0.05) and for exercise in 8,413 (33.3%) transcripts (p < 0.01, FDR < 0.05). Diet-by-exercise interaction was seen for the expression levels of 312 transcripts (p < 0.01, FDR < 0.05). Specific comparisons of interest were investigated in follow-up post-hoc analysis using Benjamini-Hochberg adjustment for multiple comparisons and are described below.

HC and LC athletes demonstrated divergent gene expression patterns at Rest

To determine differences in skeletal muscle gene expression between athletes adapted to the two diets at rest (i.e., prior to exercise), differentially expressed genes (DEG) were identified using ANOVA, with post-hoc tests focused on differential expression between LC and HC at baseline. Baseline skeletal muscle gene expression patterns differed between HC and LC (Fig. 2A). Seven genes showed lower expression, and 37 genes had higher expression in LC vs. HC. Of the seven genes with lower expression, 4 had a log2FD of < −1, and of the 37 genes with higher expression, 11 had log2FD > + 1. No DEGs more lowly expressed in LC annotated to known protein-coding genes. However, several DEGs higher in LC vs. HC were associated with fatty acid beta-oxidation via overrepresentation analysis (Fig. 2B, nominal P < 0.0001). These included acetyl-CoA acyltransferase 2 (ACAA2, + 1.6-fold higher in LC-adapted muscle), enoyl-CoA hydratase 1 (ECH1, + 1.8-fold in LC), and hydroxyacyl-CoA dehydrogenase trifunctional multienzyme complex subunit alpha (HADHA, + 1.7-fold in LC). Additionally, 3-hydroxymethylglutaryl-CoA synthase 2 (HMGCS2), involved in ketogenesis, demonstrated the highest fold difference between LC and HC athletes (+ 4.3-fold). This difference held up at the immediate post-exercise (H0) time point (+ 4.9fold higher in LC-adapted muscle) but was diminished 2 h post-exercise (H2) (Table 2).

Fig. 2
figure 2

Volcano plot showing differential gene expression in skeletal muscle of elite ultra-endurance athletes adapted to low-carbohydrate (LC) or high-carbohydrate (HC) diet. Color denotes whether individual gene reached false-discovery rate (FDR)-level significance, log2fold-difference between groups, both, or neither. 3B: gene ontology for processes associated with genes more highly expressed in muscle of individuals adapted to LC diet.

Full size image
Table 2 Differentially expressed genes in skeletal muscle of low-carb diet adapted athletes vs. high-carb adapted athletes at each study time point.
Full size table

Exercise resulted in a dynamic response in Gene expression

Exercise-responsive DEGs were identified based on significant change from baseline within a diet group at each postexercise timepoint. Overlapping patterns in the time course of diet-dependent gene expression are shown in Fig. 3A. Of the 44 DEGs at baseline, 17 remained more highly expressed in LC than HC at the H0 time point, whereas 19 additional DEGs were identified (16 upregulated, 3 downregulated). Only 1 gene [PPP1R1A (protein phosphatase 1 regulatory inhibitor subunit 1 A] was higher in LC muscle 2 h post-exercise; this gene was different across all time points between diets (Fig. 3B). Table 2 lists all DEGs between diets at each time point based on FDR < 0.05 and |log2FC| >1.

Fig. 3
figure 3

A UpsetR plot showing numbers of genes differentially expressed between diet groups at each time point, with intersections (overlaps) denoted by lines connecting rows of the bottom panel. Downregulated genes are shown in blue and upregulated in red. B: DiVenn diagram showing differentially expressed genes between diets at each time point. Each cluster corresponds to a bar on the chart in panel A. HC: high-carbohydrate, LC: low-carbohydrate; BL: baseline; H0: immediately postexercise, H2: 120-min postexercise (2 h recovery).

Full size image

Figure 4A and Fig. 4B represent numbers of unique and shared DEGs within each time point relative to the baseline for each diet group. For example, 330 genes were differentially expressed vs. baseline in HC only. These are primarily annotated to biological pathways including osteoblast differentiation, angiogenesis, and bone morphogenic protein (BMP) signaling. Relevant DEGs included SRY-box transcription factor 8 (SOX8), homeobox A2 (HOXA2), bone morphogenic protein 2 (BMP2), and melanoma cell adhesion molecule (MCAM). A total of 161 genes were different at H0 vs. baseline in LC.

Fig. 4
figure 4

UpsetR plot showing numbers of genes differentially expressed relative to baseline for each of the two diet groups, with intersections (overlaps) denoted by lines connecting rows of the bottom panel. Downregulated genes are shown in blue and upregulated in red. DiVenn diagram showing differentially expressed genes between diets at each time point. Each cluster corresponds to a bar on the chart in panel A. HC: high-carbohydrate, LC: low-carbohydrate; BL: baseline; H0: immediately postexercise, H2: 120-min postexercise (2h recovery).

Full size image

Irrespective of diet (both HC and LC combined), 198 genes were differentially expressed at the immediate postexercise timepoint (H0) and were primarily affiliated with angiogenesis and related processes, e.g., NFE2-Like BZIP transcription factor 2 (NFE2L2), phosphoinositide-3-kinase regulatory subunit 3 (PIK3R3), and hypoxia-inducible factor 3 subunit alpha (HIF3A). The 198 shared genes were also associated with a range of processes including but not limited to negative regulation of muscle adaptation [e.g., ERBB receptor feedback inhibitor 1 (ERRFI1), forkhead box O1 (FOXO1), and nitric oxide synthase (NOS3)], vasculature development [e.g., histone deacetylase 9 (HDAC9), nuclear factor of activated T-cells 2 (NFATC2) and Nodal growth differentiation factor (NODAL)], and the cellular response to starvation [e.g., cytoplasmic polyadenylation element binding protein 4 (CPEB4) and cyclin-dependent kinase inhibitor 1 A (CDKN1)]. Genes that overlapped between diet and/or timepoint were assessed for ontological relationships; when available, these are shown in Supplementary Table 1 for each of the 15 intersections (overlaps) plotted in Fig. 4A.

Visual representation of overrepresented biological processes via ClusterProfileR (Fig. 5) shows the overlapping and distinct biological pathways associated with DEG sets. Immediately after the cessation of running exercise relative to baseline, both groups demonstrated differential expression of genes related to transcription coregulator activity, nuclear hormone receptor binding, and RNA-polymerase II-specific DNA-binding transcription factor binding. At the same time, the HC group demonstrated differential expression of unique gene clusters including actin binding and ubiquitin protein ligase binding, whereas LC demonstrated unique gene clusters annotating to processes such as catalytic activity on RNA and histone acetyltransferase activity.

Fig. 5
figure 5

ClusterProfiler representation of top pathways associated with differentially expressed genes in each diet group at each time point. Color indicates Q-value (FDR-adjusted P-value) and size indicates the proportion of genes in the input gene set that are enriched in a given pathway. HC: high-carbohydrate, LC: low-carbohydrate; BL: baseline; H0: immediately postexercise, H2: 120-min postexercise (2 h recovery).

Full size image

The Early Recovery Period led to a Dynamic Response in Gene Expression, but Revealed Divergences in Diet.

Recovery-responsive DEGs were identified based on significant change from BL within a diet group at the post two hour-exercise timepoint (H2). At H2 vs. BL, the HC group showed differential expression of 222 different genes, (Fig. 3A) for which no significant prominent biological ontological association was available. A separate group of 193 genes were differently expressed based in the LC group two hours into recovery. Overrepresentation analysis shows associations with immune response and neutrophil activation, among others. As shown in Fig. 3, there were 100 DEGs shared between diets and associated with pathways related to metal ion handling and homeostasis (e.g., metallothioneins B, F, X, and H).

Figure 5 shows the top biological processes associated with DEGs at each postexercise time point by diet group. Noticeably, HC and LC were different when considering the H2 time point, indicating that recovery from exercise may be distinct as a function of diet. In HC, pathways such as symporter and nuclear receptor activity were differentially affected, but LC demonstrated differential expression of genes associated with inflammation and immunity, e.g., RAGE receptor signaling and general immune receptor signaling, among others (Fig. 5). Compared to the H0 time point, no significant biological pathways were identified from DEG in LC, whereas HC had differential expression of genes that annotated to DNA binding and nuclear receptor activity. Thus, while exercise immediately minimizes the gene expression differences between groups, recovery with divergent nutrient provisions may partially reestablish them.

Gene Expression Signatures of Metabolic Characteristics of Interest

A linear model testing the association between changes in substrate use and baseline gene expression was performed for both major fuel sources. Baseline carbohydrate oxidation rate was related to the baseline expression of 24,531 genes at FDR < 0.05. The change in carbohydrate oxidation from BL to H0 was related to baseline expression of 4,021 genes. However, no genes were related to the change in carbohydrate oxidation from baseline to H2 postexercise. The top genes are shown in Table 3 and elaborated in Supplementary Table 2. Baseline fat oxidation rate was related to the baseline expression of 1,957 genes. Change in fat oxidation from BL to H0 exercise was related to baseline expression of 3,704 genes, whereas 7,545 were related to the change from BL to H2. The top ten genes in each gene set are shown in Table 4 and elaborated on in Supplementary Table 2.

Similarly, genes associated with baseline glycogen availability, as well as genes that might predict the tendency to deplete glycogen during exercise and/or recovery, we used a continuous linear model to assess all genes for relatedness to glycogen levels in skeletal muscle samples collected at the same biopsy time points. Baseline skeletal muscle glycogen content was correlated to the expression of 1,680 genes. Furthermore, 2,372 genes had baseline expression patterns related to the change in glycogen from BL to H0, and 9,121 genes were associated with the change from baseline to H2. Table 5shows the top 10 based on unit-wise correlation and denotes those that have been previously associated with isolated skeletal muscle fiber types, where applicable21. Supplementary Table 3 shows the results for each time point (absolute value at baseline and change from baseline for each postexercise time point).

Table 3 Top ten genes related to change in carbohydrate oxidation following exercise (H0 vs. baseline) across all participants (n=20).
Full size table
Table 4 Top ten genes related to change in total fat oxidation during exercise and recovery across all participants (n=20).
Full size table
Table 5 Top ten genes related to change in skeletal muscle glycogen during exercise and recovery (H2 vs. baseline) across all participants (n=20).
Full size table

Discussion

For the first time, this investigation targeted skeletal muscle gene expression in elite ultra-endurance athletes adapted to vastly different dietary approaches surrounding an acute bout of exercise. Habitual dietary carbohydrate restriction combined with higher fat intake is associated with a series of consistently observed phenotypes, such as higher fat oxidation levels and reduced carbohydrate oxidation rates. Despite these well-documented effects, the extent to which skeletal muscle gene expression is altered in human muscle responses to keto-adaptation has been examined but there is a lack of information and the full extent to muscle adaptations have not been fully examined. This knowledge gap has implications for multiple relevant phenomena in exercise, including early recovery, repeated exposures, interference, and response heterogeneity. To date, the gene expression patterns in skeletal muscle amongst LC athletes and how this may reflect the keto-adapted phenotype in humans remain unknown. The current study aimed to address these research gaps.

A primary finding of this investigation is that distinct dietary patterns varying primarily in carbohydrate and fat in male elite ultra-endurance athletes were associated with diet-specific divergent skeletal muscle gene expression patterns at rest, that became more uniform between diet groups during the immediate recovery from an acute bout of submaximal exercise yet began to emerge again two hours post exercise. A secondary finding was that the gene expression patterns observed in LC athletes reflected the metabolic characteristics associated with keto-adapted athletes such as increased fat oxidation, fatty acid catabolic processes, ketone metabolism, and beta-oxidation2. An unexpected finding was that PPP1R1A, a gene sensitive to carbohydrate levels and involved in glycogen synthesis, was a prominent gene differentially expressed between groups at every time point and remained higher in LC athletes at all time points in this study.

Baseline (i.e., at rest, before exercise) provided the greatest number of between-group differences in LC vs. HC. Several gene expression patterns in the LC group supported the well-established findings that endurance exercise training combined with low carbohydrate and high fat intake optimizes lipid metabolism pathways within skeletal muscle22. While several genes involved in these processes were identified, some with the highest fold differences between groups included ACAA2, ECH1, and HADHA. All of these genes are closely involved with beta-oxidation in various tissues throughout the body, and we identified them specifically in skeletal muscle. For example, ACAA2is specifically involved in the final steps of mitochondrial beta-oxidation and has also been linked to adipocyte differentiation in in vitro models23. ECH1, encoding for a peroxisomal beta-oxidation enzyme, is part of a bifunctional enzyme with HADHA24. ECH1has been associated with improved metabolic health and normalized dyslipidemia in in vivo rodent models25. To date, human studies have associated ECH1with fatty acid oxidation and as a primary player in lipid metabolism25,26. HADHAencodes the alpha subunit of the mitochondrial trifunctional protein (MTP)27. In addition to fat oxidation, HADHA levels coincided with circulating lipolysis biomarkers, such as glycerol and BHB. Increases in HADHA imply more MTP production that supports the augmented free fatty acid oxidation observed in LC athletes.

HMGCS2, a rate-limiting step in ketogenesis28, had the highest fold difference between groups at BL and H0, with higher levels observed in LC when compared to HC athletes. This gene encodes the mitochondrial enzyme that catalyzes the first reaction in ketogenesis. Most research available for HMGCS2 focuses on hepatic tissue, the primary location of ketogenesis29. While it has been detected in skeletal muscle, there is limited information on its exact role and how it relates to ketogenesis. Interestingly, HMGCS2upregulation mirrored the significant increase in ketone BHB serum levels we observed at BL and immediately post exercise in this group of LC athletes2. In vitro studies have shown HMGCS2increases in the fasted state and decreases in the fed state in Hep-G2 cell lines. When increased, it induces fatty acid beta-oxidation and ketogenesis and regulates the metabolic adaptations to fasting30. In the fasted state, HMGCS2 is regulated through Sirtuin 3 (Sirt3) activity. Sirt3, found in the mitochondria, is part of a highly conserved protein family and most recently has been studied in several comparative genomic models for its role in longevity31. Calorie restriction and fasting induce Sirt3, both of which also result in increases in beta-oxidation and ketogenesis. It has also been linked with increased mitochondrial respiration while simultaneously decreasing reactive oxygen species production through its induction of UCP-132, a mitochondrial gene responsive to fasting, as well as a carbohydrate restricted, ketogenic diet33. Shimazu et al. observed HMGCS2 enzyme activity, along with BHB production, as being dependent on Sirt3levels34. This research has further expanded and now directly links increased HMGCS2activity to ketone body production and longevity in mice35. The stark difference in HMGCS2 responses in skeletal muscle between groups observed in this study warrants a more targeted investigation, primarily focused on the relationship between carbohydrate restriction, HMGCS2, oxidative stress, exercise training adaptations, and recovery processes.

Despite the incomplete overlap in DE genes between the diets, we did not detect major differences between the two groups’ biological processes associated with muscle gene expression following 3 h treadmill running. This highlights the robust effect of acute exercise on gene expression. While some differences were observed, the two groups had more uniform gene responses with the primary biological processes. For example, both had an increase in gene expression pathways associated with metabolic reactions such as gluconeogenesis and lipid metabolism (FOXO1), nitric oxide and reactive oxygen species (NOX3), and immune function (NFATC2). FOXO1has been shown to increase after a bout of endurance exercise as well as higher fat (but not necessarily lower carbohydrate) meal36. Previous works have suggested that FOXO1 is sensitive and increases in response to changes in free fatty acid availability and may be a regulator of skeletal muscle carbohydrate oxidation36by potentially inhibiting pyruvate dehydrogenase complex. Others have shown that a short-term (5 d) low-carbohydrate, high-fat diet led to decreased pyruvate dehydrogenase activity in skeletal muscle resulting in increased fat oxidation and decreased carbohydrate oxidation rates during cycling bouts37. While short-term and long-term carbohydrate restriction have been found to have distinct metabolic molecular signatures19, this phenomenon may be observed in both settings37. That is to say, FOXO1 may be the upstream regulator of this observed molecular shift during reduced dietary carbohydrate availability, regardless of whether it is a short-term diet change or habitual dietary approach. In our study, both groups saw an increased expression of FOXO1at H0, and both HC and LC groups decreased glycogen to the same level H0 despite significantly different carbohydrate and lipid oxidation rates during exercise. These gene expression patterns match the physiological data previously published2 and continue to emphasize the potentially unique carbohydrate and glycogen metabolism signatures that keto-adapted athletes may present with and also emphasize how powerful a bout of exercise is on skeletal muscle gene expression.

While not as distinct as at rest, gene expression at H2 demonstrated differences between diets in several critical biological processes beyond those observed at H0. While H2 vs. H0 would also have shed light on interesting physiological responses, the rationale for comparing BL to H2 was that the diet differences were most stark at baseline and acutely minimized by exercise; therefore, the clearest indication of recovery would be a return to resting status within each diet group. The biological pathways uniquely associated with DE genes in LC athletes included pathways related to immunity and inflammation. Ketones influence inflammation and immune health through a variety of cellular pathways38. Shaw et al. (2020) explored how a short-term ketogenic diet (i.e., 31 days) compared to a habitual, moderate-carbohydrate diet impacted immune markers before and after a run to exhaustion at 70% VO2max in trained male endurance athletes39. The authors found that the ketogenic diet induced differing immune profiles compared to the athletes’ habitual diet that, included increases in IL-10 in combination with increases in other cytokines such as IL-2, IL-1β, IL-8, and IFN-γ. Kim et al. (2022) found IL-1β, and TNF-α decreased after healthy adults consumed a very short-term, isocaloric ketogenic diet (i.e. 3 days). The authors concluded that the impact of a ketogenic diet on the inflammasome could be detected within as little as 3-days, which has been observed in other dietary approaches such as extended fasting or very low-calorie diets40. It is unknown what the immune and inflammation impacts in skeletal muscle post-exercise translate to in this study. Thus, it is of enormous value to follow up on the present work to establish whether transient changes in genes in the present study led to chronic differences in fatigability and suitability for repeated performance.

PPP1R1A was the only gene significantly upregulated in the LC group at H2 and was differentially expressed in the LC group at all three time points compared to the HC group. PPP1R1A encodes for a potent inhibitor of protein-phosphatase 1 (PP1). PP1, which increases with insulin, can dephosphorylate glycogen synthase to promote glycogen synthesis when glucose levels are high34. Similarly, PP1is deactivated when glucose levels are low to allow for activation of glycogen degradation pathways. Very little is known about the mechanisms that regulate PP1 and its inhibitors, such as PPP1R1A70, but it is highly responsive to dietary carbohydrate levels. A consistent phenotype during carbohydrate-restriction is a reduction in carbohydrate oxidation at rest and during submaximal exercise2. In contrast, intramuscular glycogen levels did not differ between groups at any point during this study, despite the LC group consuming significantly lower habitual and day-of dietary carbohydrates, indicating glycogen was degraded and resynthesized uniformly across all athletes, irrespective of diet. It is unclear why glycogen was degraded to the level it was in LC athletes, given their efficiency and evidence of fat oxidation for fuel and glycogen sparing6. We previously speculated glycogen might be utilized as an alternative fuel source within the pentose phosphate pathway. There is still much that is unknown, but it is clear that habitual carbohydrate restriction results in unique adaptations in glycogen metabolism. This phenotype may depend on the length of time the athlete restricts carbohydrate intake (i.e., weeks vs. months). Furthermore, while substrate levels in this study did not differ between groups, signaling factors related to glycogen metabolism did, warranting further investigation to elucidate this mechanism fully.

Future directions

This novel study lays crucial groundwork for many future directions exploring the role of a carbohydrate restricted, high fat diet on skeletal muscle health and performance. Ketones have been found to impact gene expression and skeletal muscle physiology in several ways, such as reduced oxidative stress and enhanced mitochondrial function41. Similarly, a rise in ketones has been associated with reduced muscle damage and improved recovery after a bout of exercise in animal models42. However, its primary role in skeletal muscle is to act as a fuel source during exercise. This has been especially prominent in exercise metabolism in highly-trained athletes, regardless of habitual diet41, but may be even more so in keto-adapted athletes. Furthermore, ketones seem to impact different muscle fiber types uniquely41, and these changes depend on whether the increase in ketones occurred due to a ketogenic diet al.one or a ketogenic diet combined with exercise training43. To date, much of this work has been conducted in murine models. Our study is the most extensive analysis to date to focus on habitual nutritional ketosis and human muscle tissue. Still, it seems to have similar findings of increased gene expression profiles related to ketogenesis and lipid metabolism in keto-adapted athletes. This study also lays the foundation for investigating the impact of ketogenic diets on exercise metabolism in the broader context of the athletes participating in these sports. Our study was limited to male participants; findings should not be extrapolated to female athletes, untrained men, or individuals taking a ketogenic diet to manage an existing metabolic disorder. Given the sex balance in participation in ultra-endurance sports44and purported higher reliance on fat resources during intensity-matched exercise in females, it is important to establish whether similar biological patterns underlie women’s responses to exercise and recovery44,45. Future opportunities could explore how gene expression profiles differ by fiber type in athletic, keto-adapted environments and how this may translate to fuel metabolism, muscle physiology, and more translational aspects related to exercise modality and training, performance, and recovery.

An area for future investigation and discovery is how these results translate to other metabolically active tissues in humans. Skeletal muscle represents a key communicative and locomotive organ system, and these roles are often dependent on physical location. For instance, the m. vastus lateralis (VL)sampled in the present study is a highly-profiled mixed-fiber type muscle with key roles in movement but may be less relied upon for treadmill running than other candidate muscles, such as the soleus or gastrocnemius46. Future opportunities to investigate the transcriptomic responses of other such muscles are warranted. Presently, selecting to profile the VL enables our findings to be compared against the multitude of existing data sets that sample this muscle. While the VL may not be the predominant locomotor muscle for treadmill running, it is substantially activated and contributes to overall systemic metabolism. This may partially explain the present similarity in glycogen utilization between the two diet groups. Additionally, the VL’s mixed fiber type presents an opportunity to investigate glycogen-dependent metabolism, as the type IIa fiber population preferentially uses glycogen47. Our findings support the idea that glycogen utilization is related to the baseline expression of type IIa-fiber marker genes identified in healthy, highly trained male VL muscle21. The impact of a ketogenic diet on gene expression in adipose tissue has been studied to some extent in in vivo models. However, this body of literature is still in its infancy. Ma et al. (2018) studied how an 8-week ketogenic diet impacted circulating markers related to energy metabolism and skeletal muscle and lipid gene expression profiles in aerobically exercising mice10. After 8 weeks, the authors found exercise capacity increased due to increased lipid and ketone metabolism. Lipid gene expression profiles matched observed patterns in circulating markers, and authors identified an increase in key lipolytic and fat oxidation genes that may be associated with enhanced exercise capacity10. Given our finding of increased fat and ketone metabolism in skeletal muscle, a future direction might explore how adipose tissue is affected by nutritional ketosis and how this relates to other phenotypes observed in keto-adapted athletes, such as increased fat oxidation, lipolysis, and ketogenesis.

While it may be argued that the exercise stimulus investigated was not exhaustively challenging for these elite athletes, it is useful to consider this bout characteristic of a typical training regimen. To mimic this further, individuals slightly modulated training to accommodate study testing but did not completely abstain from exercise altogether in the days preceding the testing bout. The employed 3 h submaximal intensity run presents a metabolic and mechanical stress that may be compounded by subsequent training sessions in the same or other modalities (e.g., “brick” training in triathlons, outdoor/trail running). Examining the effect of diet on muscle metabolism following repeated exposures would provide valuable insight into optimal training practices, race preparedness, safe recovery methods, and potentially even injury avoidance. Future studies should examine the muscle gene response to longer or repeated bouts to gain insight into the role of a ketogenic diet in protection against accumulated fatigue. Furthermore, a more detailed time course with later biospecimen collections could be very useful, as others have shown muscle damage following endurance running to last for several days in highly-trained men48. Similarly, it is important to note that, while it was a small amount of calories compared to the energy expenditure observed during the protocol, the athletes were not fasted and consumed a protein shake immediately before and after the run. This small amount of dietary intake reflects real-world settings where individuals often consume some level of caloric intake before training and for this study, may have impacted substrate oxidation and glycogen utilization to some capacity. Future investigations exploring how fasting versus specific dietary consumption surrounding exercise have a high opportunity for translation to practical application with a focus on better understanding how immediate dietary decisions impact exercise metabolism.

Conclusion

Low-carbohydrate, high fat, ketogenic diets consistently and uniformly lead to increased fat oxidation and fatty acid metabolism rates even beyond those achieved through endurance training adaptations. Here we establish for the first time the effects of a habitual low-carbohydrate, high fat diet on human skeletal muscle transcriptomic response to accustomed acute exercise. These results indicate that a habitual ketogenic diet leads to differences in constitutive skeletal muscle gene expression at rest and surrounding a bout of accustomed exercise. These gene expression patterns support the keto-adapted phenotype of increased capacity for fat oxidation and accelerated ketogenesis. Presently, the exercise stimulus had a profound influence on gene expression, but differences due to habitual diet could still be detected at every time point. Supporting observed phenotypes associated with keto-adaptation, a LC diet substantially increased genes involved in fat oxidation, lipid metabolism, and ketogenesis. One distinctive discovery was the consistently elevated expression of the glycogen inhibitor PPP1R1A in LC athletes. This unexpected differential gene expression did not match substrate outcomes, which found that glycogen breakdown and resynthesis patterns were the same between the two groups at all time points. Future research can further elucidate the connection between diet and gene expression patterns and questions such as the metabolic outcomes, how training status and exercise modality impact gene expression, and how gene expression patterns reflect signatures observed in diet-specific phenotypes. Together, this knowledge can build an understanding of the broad reach of diet manipulation, particularly ketogenic diets, and their impact on skeletal muscle health and athletic performance, with clear ramifications for metabolism in the general population and metabolic disease.

Methods

Experimental approach

This study aims to provide mechanistic context into the molecular shifts that occur at the skeletal muscle level using biospecimens collected as part of the previously published FASTER study2. FASTER recruited twenty, highly competitive, elite ultra-endurance male athletes for a cross-sectional comparison of athletes who habitually consumed a high-carbohydrate, low-fat diet (HC, n = 10) versus those who habitually consumed a low-carbohydrate, high-fat diet (LC, n = 10). Athletes were actively competing at the time of the study and competed in primarily either ≥ 50 km running competitions and/or a minimum of half-length Ironman triathlons (113.15 km). Collectively, athletes held several athletic accomplishments that included sponsorships, course, state, and/or national-level record-holdings, and national and international appearances for Team USA.

Athletes were matched for age, training status, competition distance, and ultra-endurance performance (Table 1). Primary inclusion criteria centered on athletes who consumed either a very high-carbohydrate, low-fat diet (CHO: >55%, PRO: 20%, FAT: <25%) or a very low-carbohydrate, high-fat diet (CHO: <20%, PRO: 20%, FAT: >60%) for a minimum of 6 months. Interested athletes were asked to submit a detailed, 3-day (2 weekdays, 1 weekend day) diet record to assess inclusion criteria which was reviewed and entered into a commercial nutrient analysis software (Nutritionist Pro, Axxya Systems, Stafford, TX) by a registered dietician. Eligible athletes attended a screening/informational session where they were fully informed of the protocol design and possible risks of the investigation prior to signing an informed consent. Athletes also provided written, informed consent for their biospecimens to be used in future analysis. Athletes were asked to maintain their training level and dietary profile leading up to testing. The experimental protocol was approved by the University of Connecticut Institutional Review Board and all methods were performed in accordance with the Declaration of Helsinki.

A full description of the experimental approach was published previously2. Briefly, athletes completed a 180-minute run on a treadmill at 64% of maximal oxygen consumption. About 90 min before and immediately after the run, athletes consumed a shake (5 kcal/kg LBM) with similar ingredients but with a macronutrient composition representative of their respective diets (HC (%C: F:P):50:36:14; LC (%C: F:P): 5:81:14). Shakes were prepared fresh each testing day using whole food sources (i.e. whey protein, strawberries, heavy cream, olive oil, walnut oil, and the HC athletes had additional bananas and agave syrup). Skeletal muscle biopsies were obtained from the m. vastus lateralis before (BL), immediately post (H0), and 120 min after the exercise protocol (H2) (Fig. 6). Samples were then processed for mRNA-sequencing analysis.

Fig. 6
figure 6

Experimental Timeline. Athletes ran for 180-minute on a treadmill at 64% of maximal oxygen consumption. About 90 min before and immediately after the run, athletes consumed a shake (5 kcal/kg LBM) with similar ingredients but with a macronutrient composition representative of their respective diets (HC (%C: F:P):50:36:14; LC (%C: F:P): 5:81:14). Skeletal muscle biopsies were obtained from the m. vastus lateralis before (BL), immediately post (H0), and 120 min after the exercise protocol (H2).

Full size image

180-minute treadmill exercise bout

Whole body substrate use, muscle glycogen, and circulating metabolites from this cohort were recently published2. Briefly, each subject spent 3 days visiting the laboratory, which was composed of the arrival day, day of testing, and the departure day. Participants were instructed to maintain their habitual diet leading up to testing. To ensure habitual diet matched the athlete’s respective group, each athlete recorded a food log for 48 h prior to the testing day. Participants were flown to the laboratory and arrived between 1400 and 1600 h the day before testing. Participants were instructed to minimize running and/or major races for 7 days prior to testing.

The morning of testing, participants arrived at the laboratory at 0600 h following a 10-hr overnight fast. Participants were instructed to drink 0.5 L of water the previous night and another 0.5 L of water upon waking up to ensure a euhydrated state for testing. Immediately upon arrival to the laboratory, urine specific gravity (USG) (Model A300CL, Spartan, Japan) was tested to check hydration status. Athletes were required to have a USG ≤ 1.025 to continue testing procedures. Next, anthropometrical measurements were collected. Height was measured to the nearest 0.1 cm, and body weight was recorded to the nearest 0.1 kg (OHAUS Corp, Fordham Park, NJ). Body composition was assessed via Dual-energy X-ray absorptiometry (Prodigy, Lunar Corporation, Madison, WI). Baseline resting energy expenditure and blood draw samples were collected. Immediately after, baseline muscle biopsy samples were collected and processed as described below.

Athletes were then provided a nutritional shake (5 kcal/ kg) designed to be representative of each dietary group. On average, the HC was 332 kcal shake with a breakdown of CHO: 50%, PRO: 14%, FAT: 36%. Similarly, the LC was an average of 343 kcal shake with a breakdown of CHO: 5%, PRO: 14%, FAT: 81%. All shakes were made with heavy cream, olive oil, whey protein, walnut oil, and strawberries. HC shakes had additional ingredients to increase carbohydrate content by including bananas and agave syrup. Athletes consumed the shake within 10 min after which each rested quietly in a seated position for 90 min.

Indirect calorimetry measurements were collected immediately before athletes began the exercise protocol. Athletes then ran on a treadmill at a uniform grade and a consistent pace equivalent to 64% VO2max for 180 min. Breath-by-breath gas exchange measurements of VO2 and VCO2 were collected and recovered every 30 s for a total of 10 min at the start of, 60 min into, and 150 min into the run. Oxygen uptake, minute ventilation, respiratory exchange ratio, and carbohydrate and fat oxidation rates were measured at each time point. Participants were able to drink water ad libitum throughout the run, but no other calories were ingested. Fifteen minutes after completing the run, a second muscle biopsy was taken after which athletes then consumed a shake identical to the first one. Indirect calorimetry measurements were collected immediately post, 30 (30+), 60 (60+), and 120 (H2) minutes into recovery. At H2 the final skeletal muscle biopsy was obtained and processed.

Skeletal muscle biopsy procedure and processing

All three muscle biopsies from the vastus lateralis were obtained using the modified Bergstrom technique49 with added suction described by Evans et al.50. Samples were obtained from alternate legs with the final muscle biopsy occurring 3 cm from the baseline sample site. A trained physician sterilized the biopsy location first with alcohol and then with betadine. The aseptic site was numbed using 8–10 cc of local anesthetic (2% lidocaine hydrochloride) injected at a 450 proximal and a 450 distal angle to the injection site in a cone-like formation. A #11 surgical scalpel was used to create a small incision (~ 1 cm) through the skin and muscle fascia. The muscle sample was obtained using a sterile biopsy needle, 5 mm in diameter (Surgical Instruments Engineering, Midlothian, United Kingdom). It was inserted ~ 2 cm into the incision site where a double-chop method with suction was used to collect sample. The needle was immediately removed and replaced with gauze to minimize bleeding from the incision site. A single suture was used to close the incision site.

Muscle samples were removed from the needle, cleaned to remove any excess blood or connective tissue, divided into multiple pieces, and processed accordingly. To ensure high quality RNA, immediately following the biopsy, one piece was placed in 1 ml TRIzol Reagent (Ambion, Life Technologies, Grand Island, NY), positioned on ice and immediately homogenized in 4 × 30 s set homogenizing bursts using a rotor-stator handheld homogenizer (D1000, Benchmark Scientific, Edison, NJ). Samples were then immediately frozen in liquid nitrogen and stored in −800 C for subsequent RNA extraction.

RNA extraction and sequencing

Total RNA was extracted using the traditional TRIzol method. Homogenized samples were thawed on ice followed by the addition of 200µL of chloroform and thorough mixing. Samples were left to incubate at room temperature for 5 min and then centrifuged for 15 min at 4ºC at a speed of 12,000 g. The aqueous layer containing nucleic acid was removed and nucleic acid was precipitated with 100% cold ethanol (EtOH) to reach total 35% EtOH. Samples were incubated for 30 min at −20ºC and centrifuged for 15 min at 4ºC, 16,000 g. RNA pelleted together at the bottom of the tube and supernatant was discarded. 75% ethanol was used to wash the RNA pellet twice, with supernatant being removed each time. RNA pellet was left to air-dry after which it was resuspended in 20µL of nuclease-free, ultra-pure diethlpyrocarbonate-treated water.

Samples were further treated with DNAse I (Ambion Life Technologies, Grand Island, NY). First, 1µL of RNAse I and 2 uL of 10X DNAse I buffer were added. Samples were then incubated for 30 min at 37ºC. Immediately following, 2.3µL DNase Inactivation Reagent was added. Samples were incubated at room temperature for 2 min and then centrifuged for 2 min at 4ºC at a speed of 16,000 g. Total RNA supernatant was removed and transferred to an RNAse and DNAse-free Eppendorf tube.

Extracted total RNA samples were checked for quality and concentration with a Nanodrop (NanoDrop 2000, Thermo Scientific, Wilmington, DE). One µL of sample was used to measure 260/280 ratios and 260/230 ratios. An additional cleanup was performed using an ethanol precipitation with sodium acetate to further purify each sample. One µL of 3 M sodium acetate (pH between 5 and 5.5) was added to the RNA sample and mixed well. 25µL of 100% cold ethanol was added and mixed thoroughly. Samples were incubated for 30 min at −20ºC and centrifuged for 15 min at 4ºC at a speed of 16,000 g to allow RNA to pellet. Supernatant was discarded and 100µL of 75% cold ethanol was added to the pelleted RNA and inverted several times to mix. Samples were centrifuged for 10 min, and the previous step was then repeated. After the second ethanol wash, all supernatant was removed, and RNA pellet was dried at room temperature for 5 min. 10µL of RNase-free water was added to suspend the RNA pellet. A small amount was aliquoted out to be reserved for the quality control check and the majority of the sample was stored in −80ºC for subsequent analysis.

RNA quality and concentration was determined on a Nanodrop (NanoDrop 2000, Thermo Scientific, Wilmington, DE) using 1µL of sample. Afterwards, samples were further assessed for total RNA quality and integrity on the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA) with the use of a microfluidic chip (Agilent RNA 6000 Nano Kit, Agilent Technologies, Santa Clara, CA). Chip-technology allows for a more sensitive and accurate portrayal of quality and integrity of RNA samples. The technology assesses quality through a patented RNA Integrity Number (RIN) with a custom minimum RIN threshold needed for specific downstream analyses. RIN score for sequencing was set at seven for all samples.

mRNA-sequencing

Library preparation and sequencing for all 60 samples of extracted total RNA was completed by Ocean Ridge Biosciences (Palm Beach Gardens, FL). Libraries were created using a non-strand specific library of polyA RNA (Illumina TruSeq RNA Sample Prep Kit v2, Illumina, San Diego, CA). Further quality control analysis of each DNA library included a repeat chip analysis using a BioAnalyzer and sample quantification using real time PCR.

Samples were sequenced on the Illumina HiSeq 2500 (Illumina, San Diego, CA); all samples met a minimum of 48 million paired-end 100 nucleotide reads (24 million per direction) passing sequencing filters. Mean Phred quality score was at least 30 for 94% of passed-filter reads and averaged 36.5 ± 0.2 across all reads. Across all samples, reads aligned with 69.13 ± 0.01% efficiency to the reference genome Ensembl H. sapiens (Hg19), the most up-to-date genome build at the time of the study, as described below.

Statistical approach and bioinformatics

Independent t-tests were used to compare physical characteristics, dietary profiles, and exercise data between the HC and the LC groups. Metabolic characteristics were analyzed with a two-way repeated measures analysis of variance for between (HC versus LC) and within (time-dependent) factors. Significantly different main or interaction effects were further analyzed using Fisher’s least significant difference post-hoc test. These data are presented as means ± SD. Statistical significance was defined as p ≤ 0.05.

Ocean Ridge Biosciences performed the sequencing and preliminary bioinformatic analyses. Alignment was done using both sequenced strands. FASTQ files from the sequencing run were aligned to Ensembl H. sapiens (Hg19), the Ensembl human reference genome, using TopHat 1.4.150. Files were used to generate BAM alignment files and Samtools 0.1.1851 was used to generate BAM index files. Normalization counts of reads were done by mapping reads per kilobase of transcript per million (RPKM) to annotated Ensembl genes using the open-source software Bioconductor easyRNASeq v1.6.0 package on R version 3.0 (Lucent Technologies, Inc., Murray Hill, NJ). RPKM values were filtered where the Reliable Quantification Threshold was set at 50 mapped reads. Anything under 10 reads per gene was replaced with the average RPKM value equivalent to 10 reads per gene across all samples. Filtered and adjusted gene RPKM values were used for statistical analysis.

A mixed-effect model was performed that focused on the effects of diet and exercise on gene expression. Type I Analysis of variance via the Companion to Applied Regression (car) package in R (v 3.0) was used to compare differential gene expression due to diet, exercise, and the interaction thereof. When appropriate, Tukey tests were used for post-hoc analyses. False discovery rates (FDR) were calculated from p-values using the Benjamini Hochberg method. Post-hoc tests of interest included comparing expression between both diet groups at each exercise time point (BL, H0, and H2), as well as changes in expression across time within a diet group. Gene expression was declared significantly different where FDR < 0.05 and absolute log2fold-change (log2FC) > 1, corresponding to overall expression being halved or doubled relative to the reference condition. Within each diet, DEGs that met FDR and log2FC criteria vs. baseline and compared qualitatively between diets using UpsetR analysis, as described by our group51, and graphically represented for directionality using DiVenn52. When possible, each intersection (overlapping list) of genes was assessed for biological pathways in the gene ontology database using over-representation analysis via WebGestaltR. ClusterProfileR was used to further investigate the biological similarities and differences for gene sets passing FDR and log2FC vs. baseline filters.

Finally, as an exploratory direction, baseline gene expression was assessed for correlation to baseline metabolic characteristics including skeletal muscle glycogen concentration as well as carbohydrate and fat oxidation rates. Genes were tested for their predictive capacity via relatedness to the change in glycogen (absolute difference) and carbohydrate and fat oxidation rates (fold-change) from baseline to either H0 or H2. This was performed using linear models in the limma package in R v. 4.2.2. All genes with FDR < 0.05 were pursued for interpretation.