Effect of chronic exercise in healthy young male adults: a metabolomic analysis

Abstract

Aims

To examine the metabolic adaptation to an 80-day exercise intervention in healthy young male adults where lifestyle factors such as diet, sleep, and physical activities are controlled.

Methods and results

This study involved cross-sectional analysis before and after an 80-day aerobic and strength exercise intervention in 52 young, adult, male, newly enlisted soldiers in 2015. Plasma metabolomic analyses were performed using liquid chromatography, tandem mass spectrometry. Data analyses were performed between March and August 2019. We analysed changes in metabolomic profiles at the end of an 80-day exercise intervention compared to baseline, and the association of metabolite changes with changes in clinical parameters. Global metabolism was dramatically shifted after the exercise training programme. Fatty acids and ketone body substrates, key fuels used by exercising muscle, were dramatically decreased in plasma in response to increased aerobic fitness. There were highly significant changes across many classes of metabolic substrates including lipids, ketone bodies, arginine metabolites, endocannabinoids, nucleotides, markers of proteolysis, products of fatty acid oxidation, microbiome-derived metabolites, markers of redox stress, and substrates of coagulation. For statistical analyses, a paired t-test was used and Bonferroni-adjusted P-value of <0.0004 was considered to be statistically significant. The metabolite dimethylguanidino valeric acid (DMGV) (recently shown to predict lack of metabolic response to exercise) tracked maladaptive metabolic changes to exercise; those with increases in DMGV levels had increases in several cardiovascular risk factors; changes in DMGV levels were significantly positively correlated with increases in body fat (P = 0.049), total and LDL cholesterol (P = 0.003 and P = 0.007), and systolic blood pressure (P = 0.006). This study was approved by the Departments of Defence and Veterans’ Affairs Human Research Ethics Committee and written informed consent was obtained from each subject.

Conclusion

For the first time, the true magnitude and extent of metabolic adaptation to chronic exercise training are revealed in this carefully designed study, which can be leveraged for novel therapeutic strategies in cardiometabolic disease. Extending the recent report of DMGV’s predictive utility in sedentary, overweight individuals, we found that it is also a useful marker of poor metabolic response to exercise in young, healthy, fit males.

Chronic exercise, Metabolic adaptation, Dimethylguanidino valeric acid

1. Introduction

Exercise provides many health benefits, including weight loss, improved lipid profiles, and improved insulin sensitivity. It is particularly relevant in the era of high-prevalence childhood and adult obesity and cardiometabolic disease. Exercise is a core tenet of all cardiovascular prevention guidelines,¹ and degree of physical fitness is a strong predictor of cardiovascular mortality.² However, the mechanisms underlying its benefits are unclear.

Metabolites are a diverse array of biochemicals that together capture an individual’s metabolic state. They are particularly useful in the investigation of cardiometabolic diseases that are now endemic in Western (and developing) societies. Furthermore, they can characterize response to both acute and chronic exercise. Several studies have revealed key changes in lipolysis, glycolysis, glycogenolysis, citric acid cycle, and amino acid metabolism after a single/acute aerobic exercise session and identified differences in metabolite substrate use between fit and unfit individuals.^3–5 However, much less work has been done with respect to metabolic changes following chronic exercise training. These studies reported increases in microbiome-derived tryptophan metabolites⁶ and acylcarnitines,^7,8 and decreases in adenine nucleotides.⁶

Most metabolomic studies of exercise have occurred in the setting of volunteers undergoing exercise interventions in an institution, then returning to their usual lifestyle. In chronic exercise training programmes, environmental exposures are typically not controlled during the course of the training programme that usually occurs over several weeks.

Many factors differ considerably in these individuals, including those deriving from diet, work environment, stress, socioeconomic status, and sleep. Therefore, we designed a study whereby these confounding environmental factors could be controlled. Consequently, we used soldiers of similar age and body mass index (BMI) (Table 1) who resided in the same domicile, ate the same food, had the same sleep routine, and performed the same daily activities. We examined 201 plasma metabolites before and after an 80-day exercise programme. A recent study identified dimethylguanidino valeric acid (DMGV) as predicting attenuated metabolic response to exercise in sedentary individuals,⁹ and we aimed to also examine the relationship of DMGV to metabolic response to exercise in our young, healthy, fit cohort. Therefore, our hypotheses were three-fold:

Table 1

Changes in clinical and fitness parameters

Variables	Baseline	Post-exercise
Age (years)	22 ± 4
Body mass index (kg/m2)	24.04 ± 3.14	23.54 ± 2.18*
Body fat (%)	15.50 ± 4.50	12.55 ± 3.11****
Systolic blood pressure (mmHg)	128.4 ± 9.73	125.8 ± 9.95
Diastolic blood pressure (mmHg)	65.81 ± 9.95	61.15 ± 7.49***
Heart rate (min−1)	63.98 ± 11.15	62.79 ± 9.69
Estimated VO2 max (mL/kg/min)	44.25 ± 4.8	49.59 ± 4.1****

Post-exercise vs. baseline: *P < 0.05, ***P < 0.001, ****P < 0.0001.

1. Global metabolomic profiling will capture the diverse metabolic response to exercise and will provide novel insights into the beneficial metabolic effects of exercise.

2. By mitigating confounding factors, the true magnitude and range of beneficial effects of exercise will be evident.

3. In particular, examination of metabolites associated with exercise, such as key skeletal muscle substrates (lipids¹⁰ and ketone bodies^11,12), nitric oxide metabolites (e.g. arginine¹³), and microbiome-generated metabolites,^14,15 will reveal insight into metabolic adaptation to exercise.

2. Methods

2.1 Subjects

The physical training programme as part of the Army Recruit Course was approved by the Departments of Defence and Veterans’ Affairs Human Research Ethics Committee and written informed consent was obtained from each subject. This study was performed in accordance with the Declaration of Helsinki and the project was endorsed by Forces Command through the Chief of Staff Forces Command, Director of Army Health, Commandant of Army Recruit Training Centre, and the Commanding Officer of the School of Infantry. The National Health and Medical Research Council’s National Statement on Ethical Conduct in Human Research has been adhered to in the conduct of this research.

Fifty-two newly enlisted male recruits (22 ± 4 years) from the cohort, who agreed to participate in this research, were studied before (baseline) and after an 80-day moderate-intensity mixed aerobic and strength exercise programme in 2015. Early morning fasting blood samples were taken on all participants at baseline and after the exercise programme. Bloods were collected using standard pathology collection techniques and sent to a commercial company for analysis. Anthropomorphic measurements, body fat content, plasma lipids, lipoproteins, lipoprotein subclass, apolipoproteins, fasting glucose, fasting insulin, and fitness level [20 m shuttle to estimate maximal oxygen consumption (VO₂ max)] were measured before and after the exercise programme. VO₂ max assessed pre- and post-exercise is a standard measure for cardiovascular fitness and aerobic endurance, which we used to perform power calculations and calculate required sample sizes. Based on a recent meta-analyses of changes in VO₂ max after chronic exercise regimes, with a mean VO₂ max increase of 0.3 l/min with standard deviation (SD) of 0.2 L/min, n = 7 participants would be required to detect a significant difference at a statistical power of 80% and an alpha level of 0.05.¹⁶ Therefore, our sample size of 52 individuals was more than adequate to detect a significant difference in this paired analysis. Blood samples not sent for commercial analysis were centrifuged, aliquots of plasma samples were stored at −80°C until analysis in University of New South Wales, and plasma metabolome was measured within the same batch on the same day to avoid bias caused by storage, batch, or day-to-day instrument variation.

The 80-day physical training programme as part of the Army Recruit Course consisted of a combined strength and endurance programme as well as occupational specific activities such as marching. Physical activity was converted to metabolic equivalents of task (METs) based on the 2011 Compendium of Physical activity.¹⁷ The average total physical activity time per week was 9.3 h which equated to 1.3 h per day. This equated to 7.73 MET hours per day of combined moderate- and high-intensity physical activity. Sixty-eight percent of the physical activity was of moderate-intensity (4.3–6 METs), while 32% was high-intensity (METS > 6).

The first 9 weeks of the programme predominantly consisted of 4- to 5-h long physical activity sessions, while the last ∼3 weeks culminated in several activities that consisted of long periods of physical activity.

2.2 Plasma sample extraction procedure

Plasma samples were prepared and analysed with slight modifications as described by Kimberly et al.¹⁸ and O’Sullivan et al.¹⁹ In brief, plasma samples were deproteinized with acetonitrile/methanol/formic acid (75:25:02; v/v/v) containing deuterated internal standards of 10 mM valine-d₈ (98%; Sigma-Aldrich) and 25 mM phenylalanine-d₈ (98%; Cambridge Isotope Laboratories, Inc.) for Hydrophilic Interaction Chromatography (HILIC) on an Atlantis^® HILIC column, and acetonitrile/methanol (75:25; v/v) containing 10 mM thymine-d₄, 10 mM L-phenylalanine-d₈ (98%; Cambridge Isotope Laboratories, Inc.) and 10 mM citrate-d₄ (Sigma-Aldrich) for HILIC analysis on an XBridge™ Amide column (Waters, Milford, MA, USA). After vortexing, the samples were centrifuged at 20 000 g at 4°C for 15 min, and the supernatant were transferred to HPLC-grade glass vials with inserts (Waters).

2.3 Metabolomics

Liquid chromatography–tandem mass spectrometry (LC–MS/MS) data were acquired on an Agilent 1260 Infinity HPLC System (Agilent Technologies, Santa Clara, CA, USA) coupled to an AB SCIEX QTRAP^® 5500 MS tandem mass spectrometer (MS/MS) triple quadrupole (QqQ) mass analyser operating in MRM scan mode in positive ion mode using an Atlantis^® HILIC column and negative ion mode using HILIC separation with ethylene bridged hybrid (XBridge™-AMIDE) phase column. In total, 201 metabolites were measured across HILIC and AMIDE methods, as previously described,²⁰ with 95 detected adequately in plasma. Samples were randomized and an internal pooled sample run every 10 samples for quality control. All the metabolites detected are curated in Supplementary material online, Table S1.

2.4 Data processing

All raw data files (Analyst software, version 1.6.2; AB Sciex, Foster City, CA, USA) were imported into Multi-Quant™ 3.0 Software for MRM Q1/Q3 peak integration. To account for any performance drift in the LC–MS/MS, the metabolite abundance in each sample was normalized to the bookended pool plasma sample (every 10 samples), deriving a ‘Normalized area (AU)’ (normalized abundance) for each metabolite per standard practice.

2.5 Statistical analysis

Graphical representations, tabular features, and statistical analysis were performed in Prism 7 for Windows (Version 7.02). Data are expressed as the mean ± SEM and were compared using paired t-test, with Bonferroni-adjusted P-values <4 × 10⁻⁴ (83 metabolites in HILIC platform, 118 metabolites in Amide platform) considered statistically significant. For multivariate statistical analysis, orthogonal partial least squares-discriminant analysis (OPLS-DA), normalization (mean-centering and division by standard deviation), and heatmap generation was performed using MetaboAnalyst 3.0. Effect sizes for each clinical trait per SD increment of change in DMGV level were obtained based on a linear regression model adjusted for age and BMI.

3. Results

The 52 individuals included in this study were military recruits who voluntarily enrolled in the study. These recruits were young males mean aged 22 ± 4 years (Table 1). At baseline, mean BMI was 24.04 ± 3.14 kg/m² and was significantly reduced to 23.54 ± 2.18 kg/m² (P = 0.02) after the exercise programme. Likewise, body fat (baseline 15.50 ± 4.50%) was significantly reduced to 12.55 ± 3.11% post-exercise (P < 0.001). Baseline systolic blood pressure had a non-significant reduction, but diastolic blood pressure significantly changed from 65.81 ± 9.95 mmHg at baseline to 61.15 ± 7.49 mmHg (P < 0.0001) post-exercise programme. Estimated VO₂ max (pre: 44.25 ± 4.8 mL/kg/min) was significantly increased by the exercise intervention (post: 49.59 ± 4.1 mL/kg/min) (P = 5.9 × 10⁻¹³). OPLS-DA revealed clear separation between pre-/post-exercise (Figure 1A), with greatest fold changes in ketone body and lipid metabolism (Figure 1B and Table 2).

Figure 1

(A) OPLS-DA score plot illustrates a dramatic shift in global metabolism post-exercise (green) compared to baseline (red). (B) Heatmap illustrating 27 significant metabolite changes induced by exercise in human cohort (n = 52 biological replicates per group) at a Bonferroni-adjusted statistical significance (P < 3 × 10⁻⁴). Rows: samples; columns: metabolites. Class (top bar): red: baseline (pre-exercise); green (post-exercise) plasma samples. Scale indicator: colour key indicates normalized data generated using MetaboAnalyst (v. 3.0) website tool (ranging from −6 to +6). Dark blue: lowest; dark red: highest.

Table 2

Plasma metabolites demonstrating significant changes before and after a 80-day exercise programme (P < 4 × 10⁻⁴)

Pathway	Biochemical name	KEGG	HMDB	FC	P-value
Protein catabolism	Threonine (THR)	C00188	HMDB00167	1.44	5.0 × 10−07
	α-Ketobutyrate (α-KB)	C00109	HMDB00005	−6.25	1.0 × 10−15
	Homoserine (HSE)	C00263	HMDB00719	1.31	3.0 × 10−04
	2-Oxoadipate	–	HMDB00225	−2.71	1.0 × 10−15
	Phenylpyruvate	C00166	HMDB00205	1.41	1.0 × 10−10
	Tyrosine	C00082	HMDB00158	1.42	1.6 × 10−07
	Prephenate	–	HMDB12283	1.28	1.0 × 10−05
	Hydroxyisocaproate (HICA)	C03264	HMDB00746	−1.49	4.6 × 10−09
Microbial metabolism	Indole-3-propionic acid	–	HMDB02302	1.76	4.6 × 10−05
	Taurodeoxycholate (TDCA)	C05463	HMDB00896	−2.44	2.9 × 10−03
Arginine metabolism	Arginine (ARG)	C00062	HMDB00517	1.54	2.4 × 10−05
	Ornithine (ORN)	C00077	HMDB03374	1.40	6.2 × 10−06
	α-Keto-δ-(NG,NG-dimethylguanidino)valeric acid (DMGV)	–	HMDB0240212	2.31	6.6 × 10−06
	NG-monomethyl-l-arginine (L-NMMA)	–	HMDB29416	1.54	1.6 × 10−05
	Guanidinoacetate (GAA)	C00581	HMDB00128	−2.78	<1.0 × 10−15
Carbohydrate metabolism	Glyceraldehyde	–	HMDB01051	1.16	5.2 × 10−07
Lipid metabolism	Malonate	C00383	HMDB00691	−9.10	4.7 × 10−13
	Arachidonate (20:4n6) (ARA)	C00219	HMDB01043	−1.39	6.6 × 10−08
	3-Hydroxybutyrate (BHBA)	C01089	HMDB00357	−9.10	3.2 × 10−14
	2-Arachidonoylglycerol (20:4) (2-AG)	C13856	HMDB04666	−1.11	1.9 × 10−03
	α-Ketocaproate	C00902	HMDB01864	−1.33	8.5 × 10−09
	2-Methylbutyrylcarnitine (C5)	–	HMDB00378	−1.33	5.8 × 10−04
Nucleotide metabolism	Allantoin	C02350	HMDB00462	1.28	5.1 × 10−06
	Adenosine 3′,5′-cyclic monophosphate (cAMP)	C00575	HMDB00058	−1.20	8.0 × 10−07
	UDP-N-acetyl-glucosamine (UDP-GlcNAc)	C00043	–	1.14	1.2 × 10−03
	Thymidine	C00214	HMDB00273	−1.96	2.6 × 10−8
Coagulation/haemostasis	Phytonadione	C05850	HMDB15157	−2.17	8.1 × 10−09

FC >1 indicate an increase in metabolite abundance and those with FC <1 indicate decrease in metabolite abundance.

FC, fold change (comparing before and after 80-day exercise programme); HMDB, Human Metabolome DataBase; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Tryptophan microbiome-derived metabolite indole-3-propionate (I3P) (FC = 1.76, P = 4.6 × 10⁻⁵) (Figure 2A) was significantly increased after the exercise programme. In contrast, another microbiome-derived secondary bile acid, taurodeoxycholate (TDCA), was significantly decreased by 2.44-fold (P = 2.9 × 10⁻³) post-exercise (Figure 2B).

Figure 2

Human cohort (A–D). *P < 0.05, **P < 0.01, ****P < 0.0001. (A) Indole-3-propionate, a microbiome-derived tryptophan metabolite, significantly increased post-exercise (P < 0.0001). (B) Lipid metabolites, including malonate (P < 0.0001), arachidonate (ARA, P < 0.0001), ketone body (BHBA, P < 0.0001), 2-arachidonylglycerol (2-AG, P < 0.01), and secondary bile acid taurodeoxycholic acid (TDCA, P < 0.01), significantly decreased post-exercise. (C) Arginine-related metabolites arginine (ARG), ornithine (ORN), dimethylguanidino valeric acid (DMGV), and N^G-monomethyl-l-arginine (L-NMMA) all significantly increased post-exercise (all, P < 0.0001). (D) Plasma levels of threonine (THR) and homoserine (HSE) significantly increased (both P < 0.0001), and α-ketobutyrate (α-KB) significantly decreased in human exercise cohort (P < 0.0001). Data are expressed as the mean (± SEM) with N = 52 biological replicates in each group. All data are analysed using paired t-test.

We found highly significant decreases in many intermediates of lipid metabolism post-exercise training (Figure 2B), such as lipid synthesis intermediate malonate (9.1-fold decrease, P = 4.7 × 10⁻¹³), polyunsaturated fatty acid and cell membrane constituent arachidonate (ARA) (1.39-fold decrease, P = 6.6 × 10⁻⁸), ketone body beta-hydroxybutyrate (BHBA) (9.1-fold decrease, P = 3.2 × 10⁻¹⁴), endocannabinoid receptor ligand 2-arachidonylglycerol (2-AG) (1.11-fold decrease, P = 1.9 × 10⁻³), keto-acid α-ketocaproate (1.33-fold decrease, P = 8.5 × 10⁻⁹), which is a potent insulin secretagogue,²¹ and fatty acid derivative 2-methylbutyrylcarnitine (1.33-fold decrease, P = 5.8 × 10⁻⁴).

There were significant elevations in several metabolites related to arginine metabolism post-exercise (Figure 2C): arginine (ARG) (FC = 1.54, P = 2.4 × 10⁻⁵), N^G-monomethyl-l-arginine (L-NMMA) (FC = 1.54, P = 1.6 × 10⁻⁵), ornithine (ORN) (FC = 1.4, P = 6.2 × 10⁻⁶), and DMGV (FC = 2.31, P = 6.6 × 10⁻⁶). DMGV had the greatest increase of all metabolites. However, a metabolite called guanidinoacetate (GAA), which is a product of arginine and glycine metabolism and a precursor in the synthesis of creatine, was significantly decreased by 2.78-fold post-exercise (P < 1 × 10⁻¹⁵).

Glyceraldehyde (FC = 1.16, P = 5.2 × 10⁻⁷) was increased post-exercise. As a product of glyceraldehyde-3-phosphate, this may represent increased abundance of this glycolytic intermediate in the plasma post-exercise.

There were several metabolite changes suggestive of increased protein catabolism. There was a dramatic decrease in plasma levels of α-ketobutyrate (α-KB) (6.25-fold decrease, P = 1 × 10⁻¹⁵) (Figure 2D), which is a metabolite of threonine (THR) and a substrate of isoleucine and other branched-chain amino acids (BCAAs). Another leucine metabolite, α-hydroxyisocaproate (HICA) was also significantly decreased by 1.49-fold (P = 4.6 × 10⁻⁹) in plasma post-exercise. However, two-related metabolites, the essential amino acid THR (FC = 1.44, P = 5.0 × 10⁻⁷) and its intermediate homoserine (HSE) (FC = 1.31, P = 3.0 × 10⁻⁴), were significantly elevated in the plasma post-exercise (Figure 2D). 2-Oxoadipate, a substrate of mitochondrial 2-oxoadipate dehydrogenase, was dramatically decreased by 2.71-fold (P = 1 × 10⁻¹⁵) post-exercise.

Phenylpyruvate, a keto-acid and metabolite of essential amino acid phenylalanine, was elevated in plasma (FC = 1.41, P = 1 × 10⁻¹⁰). Another essential amino acid, tyrosine (FC = 1.42, P = 1.6 × 10⁻⁷), and prephenate (a metabolite of both phenylalanine and tyrosine) (FC = 1.28, P = 1 × 10⁻⁵), were also elevated in plasma.

Purine metabolite, allantoin (FC = 1.28, P = 5.1 × 10⁻⁶) was significantly elevated post-exercise. Likewise, nucleotide product UDP-N-acetyl-glucosamine (UDP-GlcNAc) (FC = 1.14, P = 1.2 × 10⁻³) was significantly elevated. Conversely, adenine purine metabolite cyclic adenine monophosphate (cAMP) was significantly decreased by 1.2-fold (P = 8 × 10⁻⁷), as was plasma thymidine level (thymine-containing pyrimidine), with a 1.96-fold decrease (P = 2.6 × 10⁻⁸) post-exercise.

Reflecting haemostatic changes induced by exercise, phytonadione (2.17-fold decrease, P = 8.1 × 10⁻⁹), a co-factor used by the liver in the formation of coagulation factors, II, VII, IX, and X,²² was significantly decreased post-exercise.

As metabolite DMGV was recently reported to be a predictor of poor metabolic adaptation to exercise in a cohort of sedentary inviduals⁹ (and had the greatest increase of all our metabolites), we performed further analyses to determine if it had similar relationships to cardiovascular risk factors in this younger, fitter cohort. There was no significant association of baseline DMGV levels with changes in cardiovascular risk factors. However, we determined effect sizes for each clinical trait per SD increment of DMGV level change post-/pre-exercise (Figure 3), adjusted for age and BMI, and found a significant positive association with increased body fat (β = 0.3, P = 0.05), total cholesterol (β = 0.4, P = 0.003), LDL cholesterol (β = 0.4, P = 0.007), and systolic blood pressure (β = 0.4, P = 0.007), consistent with previous findings in an older, sedentary cohort.⁹ That is, DMGV levels increased in those with maladaptive metabolic changes to exercise (increases in these parameters).

Figure 3

Effect sizes for each clinical trait are reported per SD increment of DMGV level post-/pre-exercise after adjustment for age and BMI, with statistical significance was determined at P-value < 0.05. Changes in DMGV levels were positively correlated with changes in body fat (P = 0.05), total cholesterol (P = 0.0031), low-density lipoprotein level (LDL) cholesterol (P = 0.007), apolipoprotein A1 (P = 0.048), and systolic blood pressure (P = 0.0068).

4. Discussion

Our study is unique in that it is the first of its kind, to our knowledge, to examine the metabolic effects of chronic exercise training whereby confounders such as differences in diet, stress, sleep patterns, and work environment were carefully controlled. Addressing our first hypothesis, we captured the diverse metabolic response to exercise providing novel insights into the beneficial effects of exercise, as discussed further below. Furthermore, as per our second hypothesis, we report metabolite fold changes that were far greater than previously reported (e.g. short chain keto acid alpha-ketobutyrate; ketone body 3-hydroxybutyrate; dicarboxylic acid malonate), highlighting the value of controlling for mitigating environmental factors.

Two other studies also controlled for environmental confounding factors using a military-based cohort; however, the type of intervention and goals of these studies were very different, e.g. metabolic effects of a 4-day, 51-km ski-march²³ or high-intensity 6-week combat training incorporating both physical and psychological stress.²⁴ Therefore, to our knowledge, no studies have yet been performed that adequately control for confounding environmental factors in studies of chronic moderate-intensity exercise intervention in young adult males that is broadly relevant to the general population.

Recent work by Brennan et al.⁶ examined metabolite changes in three different 24-week exercise protocols (high-amount, high-intensity; high-amount, low-intensity; and low-amount, low-intensity). Several metabolites changed pre- to post-exercise within groups at the Bonferroni threshold including an increase in microbiome-derived tryptophan metabolite indole-3-lactic acid in the high-amount, high-intensity group. However, in this study, there were no significant differences in metabolite changes between the exercise and control groups after 24 weeks at a Bonferroni-adjusted statistical significance, again highlighting the need for appropriate control of environmental confounders.

In our study, numerous metabolites were significantly different after exercise at the Bonferroni threshold, across many metabolic pathways, and at significance levels not previously seen. The strongest changes were the dramatic reduction of plasma fatty acid and ketone body intermediates, two substrate classes consumed more by trained, energy-efficient skeletal muscle, and the first time this profound change has been captured. Addressing our third hypothesis, we found significant changes in skeletal muscle substrates such as ketone bodies, nitric oxide-related metabolites such as ARG and ORN, and microbiome-derived metabolites such as I3P, discussed in more detail below. We also determined unexpected changes in other metabolites, explored in more detail below.

It was recently reported that both primary and secondary plasma bile acids change (some increase, some decrease) in response to exercise.²⁵ In our study, plasma levels of secondary bile acid, TDCA, decreased in plasma after exercise training, which may result from reduced cholesterol availability (due to decreased cholesterol levels post-exercise) for bile acid synthesis that can be conjugated to taurine to form TDCA in the liver. It was previously reported⁶ that a xenometabolite derived from microbiome metabolism of tryptophan, indole-3-lactate, was significantly increased by chronic exercise, consistent with our report of increased levels of another microbiome-derived tryptophan metabolite, I3P (Figure 2A). The mechanism of increase of these gut-derived tryptophan metabolites after exercise is unclear, but they are reported to enhance the intestinal endothelial barrier and to be anti-inflammatory.²⁶ Increased plasma xenometabolites may underpin some of the anti-inflammatory and immune benefits of exercise.²⁷

Malonate is involved in the synthesis of fatty acids, and as endurance training substantially increases fatty acid oxidation capacity, decreased malonate may represent increased consumption of fatty acids by oxidation. Interestingly, of all metabolite changes, malonate had the joint-greatest magnitude of change (9.1-fold reduction), suggesting a highly robust response to exercise. Another polyunsaturated fatty acid, ARA (20:4n-6), was significantly reduced in plasma post-exercise regimen, as previously reported in exercising humans and rats.^28,29 In these previous reports, and our results, it is unclear why there is a reduction of this lipid, which is known to be associated with peripheral (skeletal muscle) insulin sensitivity. The reduced level of short-chain ACs such as 2-methylbutyrylcarnitine in the plasma may be reporting on increased efficiency of skeletal muscle beta-oxidation. Incomplete fatty acid beta-oxidation leads to accumulation of ACs; exercise training and improved fitness leads to more efficient beta-oxidation and reduced AC levels.¹⁰ Increased fatty acid beta-oxidation may also contribute to the improved lipid profiles associated with exercise.^30,31 Therefore, the reduced ACs we saw in plasma post-exercise likely represents the increased capacity of skeletal muscle to metabolize them, which is the same process that enhances the oxidation of fatty acids, thereby improving lipid profiles.

The ketone body BHBA was the other metabolite with a 9.1-fold reduction in levels post-exercise. Ketone bodies as a source of fuel in athletes have received much attention lately.^11,12 Ketone bodies are oxidized as a fuel source during exercise, are markedly elevated during post‐exercise recovery, and the ability to utilize ketone bodies is higher in exercise‐trained skeletal muscle.¹² Therefore, exercise-trained skeletal muscle will utilize ketone bodies to a greater capacity than the untrained muscle, explaining its reduction in plasma.

ARG, a substrate of nitric oxide synthase, and related metabolites were elevated after the exercise intervention. ARG plays an important role in multiple physiological phenomena, including the production of creatine, ammonia, urea, protein, and nitric oxide, which leads to vasodilation and greater blood flow, important in exercising muscle.¹³ ORN, a non-proteinogenic amino acid reported to regulate growth hormone/insulin-like growth factor-1/insulin-like growth factor-binding protein 3 complex in muscle tissue, was also elevated.³² ARG and ORN supplementation stimulate growth hormone secretion, increasing lipolysis and lipid oxidation, thereby enhancing energy expenditure after heavy resistance exercise.³² Therefore, our results suggest that exercise-induced increases in circulating ARG and ORN may contribute to the improved vascular function and fat loss associated with exercise.³³

α-KB is a metabolite of THR, substrate of valine and isoleucine synthesis, and substrate of enzyme branched-chain alpha-keto dehydrogenase (BCKDH) complex. The dramatic decrease in levels of α-ketobutyrate in our study is likely reporting on the activation of muscle BCKDH complex by exercise, resulting in enhanced BCAA catabolism and consumption of BCAA substrates. 2-Oxoadipate, whose oxidation generates superoxide via the electron transport chain, was dramatically reduced, likely reflecting increased muscular oxidative efficiency.³⁴

PHE and TYR are essential amino acids in our diet. Our samples were taken in the fasting state, when PHE entry via proteolysis is matched by protein synthesis and metabolic disposal to TYR. PHE is a good marker of proteolysis and both PHE and TYR can reflect protein turnover.³⁵ This may explain increased plasma phenylpyruvate (produced from PHE by PHE transaminase), TYR, and prephenate (metabolite of PHE and TYR) as plasma markers of exercise-induced proteolysis in exercising individuals.

It has been consistently shown that the endocannabinoid AEA is significantly increased after a single episode of exercise,^36–39 whilst less is known about the relationship of endocannabinoid 2-AG with any type of exercise. Heyman et al.³⁸ reported no change in circulating 2-AG levels after cycling. However, we found 2-AG to be significantly decreased post-exercise training (P < 0.01). The biological implications of this are unclear, as 2-AG and endocannabinoid signalling control various central functions, such as nociception, feeding, energy homeostasis, mood, learning, memory, growth, development, and reward processes.⁴⁰ The decrease in this endocannabinoid after chronic exercise compared to earlier reports of increased endocannabinoids after acute exercise^36–39 may represent adaptation and fitness (as compared to acute perturbations).

Allantoin was significantly elevated in plasma post-exercise, consistent with previous reports of accumulation of products of adenine purine nucleotide catabolism, such as uric acid, hypoxanthine, and allantoin in plasma post-acute exercise in horses⁴¹ and humans.⁴² As allantoin is produced from the non-enzymatic oxidation of urate, it has been suggested to be a useful indicator of oxidative stress caused by intensive exercise⁴² and a biomarker of oxidative stress more generally.⁴³ In the context of chronic exercise, increased allantoin most likely represents increased oxidative stress during exercise that causes increased oxidation of urate to allantoin.⁴⁴

Conversely, two nucleotides—cAMP and thymidine—were significantly decreased post-exercise. Intriguingly, cAMP was previously shown to increase post-acute exercise in humans.⁴⁵ However, rats post-chronic exercise training over a 5-week period had decreased tissue cAMP compared to untrained rats,⁴⁶ consistent with our human results, and suggesting an adaptive nucleotide response to chronic exercise training that is distinct to acute effects.

Exercise leads to activation of both coagulation and fibrinolytic cascades, and in the post-exercise period fibrinolysis returns to normal, whereas coagulation was shown to remain elevated up to 1-day post-exercise.⁴⁷ The decreased plasma phytonadione (used in the formation of clotting factors) we observed may well be as a result of sustained activation of the clotting system on the day following the exercise training programme ended. This has important implications for the exercise regimes in cardiovascular disease patients.

The recent observation that DMGV predicts poor metabolic response to exercise⁹ suggests even further utility of this novel metabolite, as we previously discovered its role as a marker of non-alcoholic fatty liver disease and a predictor of Type 2 diabetes mellitus over a decade in advance in cohorts of different ethnicities.¹⁹ Our result in this cohort suggests that even in young, healthy, and fit adult males, DMGV’s relationship to poor metabolic response to exercise persists. This is consistent with the recent report of DMGV’s association with visceral adiposity, decreased insulin sensitivity, and dyslipidaemia amongst a young (mean age 26 years) subsample of the HERITAGE participants with normal weight (median BMI 24 kg/m²).⁹ Therefore, our data buttresses recent findings of the value of DMGV as a marker of subclinical metabolic dysfunction even in low risk and apparently healthy individuals.

Since we discovered the relationship of DMGV to metabolic disease,¹⁹ little has been published on the mechanistic underpinnings of this relationship. The two subsequent papers that have been published examined the relationship of this metabolite to exercise⁹ and coronary artery disease/lifestyle factors.⁴⁸ Robbins et al.⁹ paper determined that baseline DMGV levels predicted those who did not get a beneficial metabolic response to exercise. The Ottosson et al.⁴⁸ paper determined that plasma DMGV was an independent predictor of incident coronary artery disease and cardiovascular mortality, that there was a positive association of DMGV with ingestion of sugar-sweetened beverages, and a negative association of DMGV with vegetable intake and exercise. It also replicated the original association with Type 2 diabetes mellitus that we found in our original paper.¹⁹ We have also recently determined a positive association between DMGV and sugar-sweetened beverage intake and a negative association with dietary fibre intake in a large human cohort (unpublished data). In addition, we demonstrated that fructose activates the enzyme that produces DMGV in terms of mRNA expression, protein abundance, and enzymatic activity (unpublished data). However, much further work needs to be done to definitively determine mechanism.

4.1 Limitations

There were some limitations to our study. Our human cohort consisted of only males, so relevance for females is unclear. Furthermore, our cohort was young and healthy, with a relatively high baseline fitness. Therefore, this study needs to be repeated in cohorts of different ages, female gender, different BMIs, and with different baseline fitness levels. Nevertheless, control of extraneous environmental factors remains highly important regardless of the cohort.

5. Conclusion

Utilizing a before/after comparison in a human exercise cohort in which extraneous environmental factors were carefully controlled has allowed a determination of the true magnitude and extent of metabolic adaptation to exercise training. For the first time, therefore, we were able to provide an accurate report of the degree of increased consumption of fatty acid and ketone body substrates by trained, energy-efficient muscle. We also captured heretofore unseen, in terms of scale and scope, shifts in metabolism across many different substrates including lipids, ketone bodies, arginine metabolites, endocannabinoids, nucleotides, markers of proteolysis, products of fatty acid oxidation, microbiome-derived metabolites, markers of redox stress, and substrates of coagulation. These findings have important implications in cardiovascular disease prevention and risk reduction regimes. Finally, our data further supports DMGV’s efficacy as an early maker of subclinical metabolic dysfunction in apparently healthy and fit young individuals.

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Authors’ contributions

Y.C.K. and J.F.O.S. had full access to the data and take full responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: K.S. and D.S.C. Acquisition, analysis, or interpretation of the data: Y.C.K., M.L., J.Y., and J.F.O.S. Drafting of the manuscript: D.S.C., Y.C.K., and J.F.O.S. Critical revision of the manuscript for important intellectual content: D.S.C., Y.C.K., J.Y., and J.F.O.S. Statistical analysis: M.L., Y.C.K., and J.Y. Obtained funding: D.S.C., K.S., and J.F.O.S. Administrative, technical, or material support: D.S.C. and J.F.O.S. Supervision: D.S.C., J.Y., and J.F.O.S.

Conflict of interest: none declared.

Funding

This work was supported via a Program Grant from the National Health and Medical Research Council Australia (to D.S.C.), Heart Research Institute (to J.F.O.S. and Y.C.K.), Sydney Medical School Foundation Chapman Fellowship (to J.F.O.S.), and NSW Health Early-Mid Career Fellowship (to J.F.O.S.).

Translational perspective

• For the first time, the true extent of metabolic adaptation to chronic exercise was revealed, by eliminating extraneous environmental confounders.

• As exercise is a core tenet of all cardiovascular prevention guidelines, it is critical that the full extent of metabolic adaptation to exercise is known.

• Knowing indicators of response and non-response can help tailor exercise programmes at an individual level for maximum benefit.

• It may also reveal novel therapeutic strategies to replicate the beneficial effects of exercise. This could be important for individuals who are less able, or unable, to exercise.

• Levels of a metabolite called dimethylguanidino valeric acid, known to be influenced by lifestyle factors, may identify individuals yielding less benefit from exercise and who may need additional lifestyle interventions to improve their health.