Exercise training elicits a broad range of physiological adaptations in many tissues and organ systems (Egan & Zierath, 2013; Hawley et al., 2014). When a specific type of exercise training is emphasized, divergent exercise training phenotypes can manifest (e.g., endurance- or strength-trained athletes; Coffey & Hawley, 2017; Egan & Zierath, 2013), which include specific anthropometric and performance traits that can differ markedly in different athletic populations (Degens, 2019).
Metabolites are low molecular weight (mostly organic) chemicals that are usually the reactants, intermediates, or products of metabolic pathways (Dunn et al., 2011; Nicholson & Lindon, 2008). The metabolome represents the collective output of metabolic reactions and is arguably the most accurate representation of the phenotype of a sample at the time of measurement (Belhaj et al., 2021; Patti et al., 2012). The circulating metabolome is of interest as it represents metabolite contributions from all tissues and is an integrated snapshot of systemic metabolism (Dunn et al., 2011). Considering the extent and diversity with which exercise training can induce physiological remodeling, there is emerging interest in whether exercise training history or exercise training interventions alter the profile of the circulating metabolome (Khoramipour et al., 2022). Profiling of the circulating metabolome in a resting state using a cross-sectional design between groups with divergent histories of exercise training can infer the influence of exercise training while attenuating the confounding residual effects of recent exercise training sessions (Darragh et al., 2021).
To date, two cross-sectional studies have investigated the relationship between exercise training and the circulating metabolome at rest (Monnerat et al., 2020; Schranner et al., 2021), both of which suggest alterations associated with divergent exercise training history and/or performance characteristics. However, both studies comprised of only small sample sizes and the extent of control of dietary intake, and the last exercise training session, was unclear. Moreover, the studies acquired samples at only a single time point. When interpreting the durability of alterations to the circulating metabolome related to exercise training, these methodological issues are salient given the dynamic nature of the circulating metabolome and the evident within-subject variability in human samples (Agueusop et al., 2020; Breier et al., 2014; Floegel et al., 2011; Yin et al., 2022).
Therefore, using a targeted profile of metabolites measured in plasma samples taken at rest, the present study firstly investigated the within-subject variability in the circulating metabolome while controlling for time of day of sampling, recent dietary intake, time since last meal, and time since last exercise training session. Secondly, we investigated whether exercise training history was associated with alterations in the circulating metabolome by comparing samples from recreationally active controls and two groups of exercise-trained individuals with divergent training histories and performance characteristics.
Methods
Participants
Men (n = 38) who were endurance-trained (END; n = 13), strength-trained (STR; n = 13), and recreationally active controls (CON; n = 12) were recruited for this study (Table 1 and Supplementary Table S1 [available online]). To qualify for the respective training group, participants self-reported being able to meet the following criteria: END, at least two of the following running performances, 5,000 m < 20 min, 10,000 m < 40 min, or 16,000 m < 64 min; and STR, a one repetition maximum for at least two of the following, squat ≥ 200 kg, bench press ≥ 140 kg, or deadlift ≥ 220 kg. Three recent performances were recorded (Supplementary Table S1 [available online]), and these were verified via social media accounts (e.g., Instagram), training logs (e.g., Garmin Connect, Strava), and/or public databases (e.g., www.worldathletics.org, www.openpowerlifting.org). Based on the recently proposed Participant Classification Framework (McKay et al., 2022), END was comprised of n = 8 Tier 3/highly trained athletes and n = 5 Tier 4/elite athletes, whereas STR was comprised of n = 9 Tier 3/highly trained athletes and n = 4 Tier 4/elite athletes. The CON participants were Tier 1/recreationally active but did not participate in intensive or sport-specific training.
Participant Characteristics and Details of Control Measures for Visits 1 and 2
| Variable | END (n = 13) | STR (n = 13) | CON (n = 12) | Interaction F | Interaction p | Interaction | Group F | Group p | Group | Visit F | Visit p | Visit |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Age (years) | 30 ± 6 | 25 ± 5 | 26 ± 2 | 3.285 | .05 | .166 | ||||||
| Height (m) | 1.82 ± 0.06 | 1.80 ± 0.07 | 1.78 ± 0.07 | 0.876 | .426 | .052 | ||||||
| Body mass (kg) | 71 ± 7** | 94 ± 9 | 78 ± 8** | 1.634 | .211 | .093 | 27.259 | <.001 | .63 | 0.124 | .727 | .004 |
| FFM index (kg/m2) | 17 ± 1** | 23 ± 2 | 18 ± 1#,** | 2.149 | .133 | .118 | 65.943 | <.001 | .805 | 0.671 | .419 | .021 |
| Recent 5 km time (min:s) | 16:30 ± 1:22 | |||||||||||
| Recent bench press (kg) | 148 ± 30 | |||||||||||
| Recent back squat (kg) | 228 ± 30 | |||||||||||
| Recent deadlift (kg) | 253 ± 38 | |||||||||||
| Time training per week (hr) | 9 ± 10 | 9 ± 2 | 4 ± 1##,** | 51.992 | <.001 | .594 | ||||||
| Average time between visits (days) | 12 ± 10 | 9 ± 6 | 9 ± 5 | 0.02 | .98 | .001 | ||||||
| Maximum time between visits (days) | 42 | 26 | 21 | |||||||||
| Minimum time between visits (days) | 7 | 7 | 6 | |||||||||
| Time since most recent meal, Visit 1 (hr) | 12 ± 2 | 12 ± 2 | 12 ± 1 | 0.939 | .401 | .052 | 1.132 | .334 | .062 | 0.711 | .405 | .02 |
| Time since most recent meal, Visit 2 (hr) | 12 ± 1 | 12 ± 2 | 13 ± 1 | 0.939 | .401 | .052 | 1.132 | .334 | .062 | 0.711 | .405 | .02 |
| Time since most recent exercise, Visit 1 (hr) | 46 ± 23 | 66 ± 66 | 75 ± 84 | 1.945 | .159 | .103 | 0.974 | .388 | .054 | 0.376 | .544 | .052 |
| Time since most recent exercise, Visit 2 (hr) | 44 ± 12 | 42 ± 17 | 88 ± 132 | 1.945 | .159 | .103 | 0.974 | .388 | .054 | 0.376 | .544 | .052 |
| Average time of sampling, Visit 1 | 08:32 | 09:16 | 08:44 | 1.559 | .226 | .089 | 1.322 | .281 | .076 | 3.672 | .064 | .103 |
| Average time of sampling, Visit 2 | 08:17 | 09:18 | 08:32 | 1.559 | .226 | .089 | 1.322 | .281 | .076 | 3.672 | .064 | .103 |
| Maximum difference between Visit 1 and Visit 2 time of sampling (min) | 54 | 23 | 54 | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A |
| Minimum difference between Visit 1 and Visit 2 time of sampling (min) | 1 | 1 | 0 | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A | N/A |
Note. END = endurance; STR = strength; CON = control; FFM = fat-free mass; N/A = not applicable.
Significant difference from STR (**p < .01). Significant difference from END (#p < .05, ##p < .01).
Study Design
This study received ethical approval from the Research Ethics Committee of Dublin City University (DCUREC/2021/079) in accordance with the Declaration of Helsinki, and each participant provided written informed consent prior to participation. Participants visited the laboratory on two separate occasions under controlled conditions in an overnight fasted state and having not exercised for at least 24 hr beforehand (Table 1). Height was measured using a stadiometer (model 213, SECA), and body mass and body composition (fat mass, %body fat, and fat-free mass) were assessed by bioimpedance spectroscopy (SOZO, ImpediMed) (Esco et al., 2019). Participants lay supine for 10 min while answering questions to verbally confirm compliance with all previsit preparations including the timing of their most recent meal and exercise training session. A venous blood sample was then taken from a superficial forearm vein.
Dietary Control
Participants were provided with standardized meals (GourmetFuel™) for consumption on the day prior to each visit, which were delivered by a member of the research team. The meals provided 30 kcal/kg body mass with a macronutrient ratio of 50/25/25 for carbohydrate, protein, and fat, respectively. Participants were permitted to consume caffeine during the day before each visit in doses that contained negligible additional nutrients (i.e., tea or coffee without milk), but were asked to consume only water on the morning of each visit. Participants were also asked to abstain from alcohol consumption in the 24 hr preceding each visit.
Blood Sampling, Processing, and Storage
Whole venous blood was drawn by inserting a 21G butterfly needle (Greiner, Bio-One) into an antecubital vein. The initial 4 ml of blood was drawn into a generic vacutainer and discarded. Subsequently ∼50 ml of whole blood was drawn into six 9-ml blood collection tubes coated with ACD-A anticoagulant (Greiner, Bio-One). Blood samples were immediately placed on ice and centrifuged at 1500g for 15 min at 4 °C. Directly after centrifugation, plasma samples were separated into aliquots and stored at −80 °C. Samples were collected and stored between August and December 2021 and were analyzed in batch in May 2022.
Plasma Metabolomics
Plasma metabolomics analyses were performed by The Metabolomics Innovation Center using a targeted assay capable of detecting up to 172 metabolites and employing liquid chromatography and mass spectrometry methods. Further details of this method are detailed in Supplementary Material (available online) and have been described elsewhere (Zheng et al., 2020). One hundred and fifty-nine metabolites were detected by liquid chromatography and mass spectrometry methods and classified into five distinct metabolite groups (amino acids, peptides and analogs; fatty acids and fatty acid conjugates; acylcarnitines; glycerophosphocholines and phosphosphingolipids; and other) using the “ChemOnt” taxonomic classification technique (Feunang et al., 2016; Supplementary Table S2 [available online]). In addition, four metabolite sums (branched chain amino acids, gluconeogenic amino acids, essential amino acids, and total acylcarnitine) and three metabolite ratios (acylcarnitine/carnitine, C2/C0, and kynurenine/tryptophan) were calculated (Petersen et al., 2012), resulting in a total of 166 metabolite features included in the final analysis.
Principal Component Analysis
Prior to dimension reduction, a Kaiser–Meyer–Olkin test of sampling adequacy and Bartlett’s test of sphericity were performed on data to ensure they met the minimum standards for principal component analysis (PCA). Bartlett’s test of sphericity produced a significant result, X2(165) = 103,410; p < .01, while a Kaiser–Meyer–Olkin test of mean sampling adequacy produced a result of 0.5, which is a low, but acceptable value (Kaiser & Rice, 1974). A multilevel PCA using an orthogonal rotation with features mean zeroed and scaled for unit variance was then performed on the full set of 166 metabolites, metabolite sums, and ratios using mixOmics (version 6.14.1), a package designed for use within the statistical programming language R (Rohart et al., 2017). Multilevel PCA refers to a modified analysis applicable to repeated-measures design that involves decomposition of within-subject variation prior to dimension reduction. This approach has been demonstrated to increase model accuracy for paired-sample data (Liquet et al., 2012).
Statistical Analysis and Calculation of Intraclass Correlation Coefficients
Data are reported as mean ± SD unless otherwise stated. All analyses were performed using the base version of the programming language R (4.2.1) or the “Rstatix” package (version 0.7.0). For variables in which a difference in outcomes between visits was possibly expected (body composition, timing of each sample, time since last meal, and exercise session), data were analyzed using a two-way (Visit × Group) mixed analysis of variance. No interaction effect was observed for any variable, but given our a priori interest in between-group differences, in the presence of a significant main effect for Group, post hoc pairwise comparisons were performed using the Bonferroni correction. In circumstances where visit was not a relevant independent variable (e.g., height, age, number of days between Visit 1 and Visit 2), data were analyzed using a one-way analysis of variance with post hoc pairwise comparisons again performed using the Bonferroni correction. All analysis of variance are reported with a relevant estimate of standardized effect size (
Univariate testing of metabolite data was performed using paired t tests with the Benjamini–Hochberg method to control for the false discovery rate of multiple comparisons. This approach was first applied to the full set of identified metabolite data (166 features) within each group. When no significant differences were observed between visits for each group, Visit 1 and Visit 2 data were averaged for participants in each group, and the same series of pairwise tests were performed to compare groups (i.e., END vs. CON; STR vs. END; END vs. STR). Intraclass correlation coefficients (ICC) for each individual metabolite were calculated using the ICC2 method (Shrout & Fleiss, 1979). The reliability of circulating metabolites was interpreted using arbitrary thresholds (<0.4, “poor”; 0.4–0.5, “fair”; >0.5–0.75, “good”; and >0.75 “excellent”) that were established for analysis of psychometric test scores (Cicchetti, 1994), but have been employed to describe the reliability of circulating metabolites (Agueusop et al., 2020; Floegel et al., 2011; Li-Gao et al., 2019). All statistical tests rejected the null hypothesis at an alpha level of <.05, or a false discovery rate <0.05 for univariate tests.
Results
Descriptive and Visit Characteristics
Participants were of similar age and height (Table 1). Body mass and fat-free mass index of STR (body mass, 94.5 ± 8.8 kg; fat-free mass index, 23.0 ± 1.8 kg/m2) were greater than that of END (71.0 ± 6.8 kg; 16.9 ± 1.1 kg/m2) and CON (77.6 ± 7.7 kg; 18.1 ± 1.0 kg/m2), whereas the CON had greater body mass and fat-free mass index compared to END (all p < .05; Table 1). STR (8.5 ± 1.8 hr) and END (8.6 ± 2.2 hr) reported exercising more hours per week (both p < .05) than CON (3.8 ± 1.3 hr). The time of day that the samples were drawn was similar between groups and visits, as was the number of days between each visit between groups (Table 1).
Reliability of Metabolite Concentrations Between Visits
The majority of the 159 detected metabolites displayed fair or better reliability between visits as evidenced by 98 metabolites (∼62%) having ICCs ≥ .4, with 46 metabolites (∼29%) displaying good reliability (ICCs > .5), and 31 metabolites (∼19%) displaying excellent reliability (ICCs > .75) (Figure 1a; Supplementary Table S3 [available online]). ICCs varied between metabolite groups with glycerophosphocholines and phosphosphingolipids displaying the greatest reliability (83% of metabolites with an ICC > .5) and acylcarnitines displaying the lowest reliability (30% of metabolites with an ICC > .5) (Figure 1b).
—The ICC of detected metabolites. (a) This distribution of all metabolites by ICC interval. (b) The number of metabolites in each group (y-axis) with the percentage of individual metabolites in each specific ICC interval. ICC = intraclass correlation coefficients.
Citation: International Journal of Sport Nutrition and Exercise Metabolism 33, 3; 10.1123/ijsnem.2022-0177
Within-Subject and Between-Group Analyses
Multilevel PCA was performed to investigate whether clear spatial separations were apparent in within-subject or between-group analyses. This model produced no apparent clustering within groups or separation between groups, and different visits within participants tended to cluster closely implying the profile of a participant’s metabolome was consistent between visits (Figure 2; Supplementary Table S4 [available online]).
—A multilevel PCA of the collapsed profile of targeted metabolomes from the participants for both visits; circles represent endurance athletes, triangles represent strength athletes, and squares represent control participants. PC1/PC2 refers to the first and second principal components, that is, latent variable structures that explain the largest and second largest unique proportion of total model variance, respectively. expl.var = explained variance; PCA = principal component analysis.
Citation: International Journal of Sport Nutrition and Exercise Metabolism 33, 3; 10.1123/ijsnem.2022-0177
The results of univariate within-subject analyses demonstrated that no metabolites displayed significantly different abundances between visits within any group (Figure 3). Univariate tests comparing the resting concentration of each metabolite between groups (END vs. CON; STR vs. CON; and END vs. STR) revealed a combined total of 44 metabolites with differences between groups (false discovery rate <0.05; Figure 4; Supplementary Table S5 [available online]). Boxplots for each of these 44 metabolites are presented in Supplementary Figure S1 (available online). The largest number of differentially abundant metabolites (42) was between END and CON, with 16 metabolites demonstrating lower abundance and 26 metabolites demonstrating higher abundance in END (Figure 4a). Of the 16 metabolites with lower abundance, four metabolites were in the acylcarnitine group, four metabolites were in the amino acid, peptide and analog group, six metabolites were in the fatty acid and fatty acid conjugate group, one metabolite was in the glycerophosphocholines and phosphosphingolipids, and one metabolite was in the others groups (Supplementary Table S5 [available online]). Of the 26 metabolites with higher abundance, six metabolites were in the acylcarnitine group, eight metabolites were in the fatty acid and fatty acid conjugate group, and 12 metabolites were in the glycerophosphocholines and phosphosphingolipids group (Supplementary Table S5 [available online]).
—Univariate analysis of metabolites within each group. (a) Endurance athletes, (b) strength athletes, (c) recreationally active controls. Shapes represent metabolites in specific metabolite groups with the following coding; squares = acylcarnitines; circles = amino acids, peptides, and analogs; triangles = fatty acids and conjugates; diamond = glycerophosphocholines and phosphosphingolipids; inverted triangle = others.
Citation: International Journal of Sport Nutrition and Exercise Metabolism 33, 3; 10.1123/ijsnem.2022-0177
—Univariate analysis of metabolites between each groups. (a) Endurance athletes versus control, (b) strength athletes versus control, and (c) endurance athletes versus strength athletes. Annotated metabolites are differentially abundant (false discovery rate <0.05). (d) A Venn diagram showing the overlap of differentially expressed metabolites between individual group analysis. Shapes represent metabolites in specific metabolite groups with the following coding; squares = acylcarnitines; circles = amino acids, peptides, and analogs; triangles = fatty acids and conjugates; diamond = glycerophosphocholines and phosphosphingolipids; inverted triangle = others.
Citation: International Journal of Sport Nutrition and Exercise Metabolism 33, 3; 10.1123/ijsnem.2022-0177
Comparing STR with CON (Figure 4b), 10 metabolites were differentially abundant, with five metabolites demonstrating higher abundance (four metabolites in the fatty acid and fatty acid conjugate group, and one metabolite in the Others group), and five metabolites displaying lower abundance (all in the fatty acid and fatty acid conjugate group) in STR (Supplementary Table S5 [available online]). Comparing END to STR (Figure 4c), five metabolites were differentially abundant with two metabolites displaying lower abundance (one in the amino acid, peptide and analog group, and one in the others group) and three metabolites displaying higher abundance (all in the glycerophosphocholines and phosphosphingolipids group) in END (Supplementary Table S5 [available online]).
Discussion
The present study employed a targeted profile of the circulating metabolome in resting plasma samples, first to investigate reliability when measured on two separate days. Under controlled conditions, namely time of day of sampling, recent dietary intake, time since last meal, and time since last exercise training session, the reliability of plasma metabolite concentrations varied largely at the level of individual metabolites. Specifically, ∼48% of metabolites displaying good-to-excellent reliability in resting samples, whereas the remaining ∼52% of metabolites displayed fair-to-poor reliability. Second, investigating whether divergent histories of exercise training were associated with alterations in the circulating metabolome revealed that the abundance of ∼28% (44/159) of the metabolites detected in the targeted metabolite profile were altered in plasma between endurance- or strength-trained men compared with recreationally active controls.
Using ICC as an assessment of reliability, the median ICC across metabolites was .49, and ∼62% of metabolites displayed an ICC ≥ .4. These data suggest that the majority of metabolites displayed fair or better reliability between visits, with ∼48% displaying good-to-excellent reliability. However, the median ICC observed in the present study is lower than previously reported in other studies in blood samples taken from overnight fasted humans on multiple days, including both untargeted (median ICC of .65 for 1,438 metabolites identified in serum, Agueusop et al., 2020; median ICC of .66 for 148 metabolites identified in plasma, Li-Gao et al., 2019) and targeted (median ICC of .57 for 163 metabolites identified in serum, Floegel et al., 2011; median ICC of .62 for 138 metabolites identified in plasma, Yin et al., 2022).
When analyzed by metabolite group in the present study, the glycerophosphocholines and phosphosphingolipids group displayed the greatest reliability (median ICC of .78, with 63% of metabolites displaying ICCs > .75, and only 14% displaying a ICCs < .4), and the acylcarnitine group displaying the poorest reliability (median ICC of .28, with 60% of metabolites displaying ICCs < .4 and no metabolites displaying an ICC > .75). In similar studies, phosphocholines and sphingolipids have likewise displayed good-to-excellent reliability (median ICCs ∼.6 to ∼.8; Agueusop et al., 2020; Floegel et al., 2011; Yin et al., 2022), but our observation of poor reliability of the acylcarnitine group is in contrast to previous studies in overnight fasted humans (median ICCs ∼.6 to ∼.9; Agueusop et al., 2020; Breier et al., 2014; Floegel et al., 2011; Yin et al., 2022). The reliability of acylcarnitines may vary in a manner that is dependent on the length of the acylcarnitine chain with short-medium acylcarnitines displaying higher ICCs compared with longer chains (Breier et al., 2014; Floegel et al., 2011). A difference may also exist between plasma and serum as evidenced by acylcarnitines identified in serum having higher ICCs than those in plasma (Breier et al., 2014). The present study analyzed a mixed panel of 42 acylcarnitines ranging from C1 to C18, and measured metabolite concentrations in plasma, not serum, and at the level of specific acylcarnitines, some of the reliability data are similar to other studies (Floegel et al., 2011; Yin et al., 2022). For example, C4, which we observed to have an ICC of .70, has previously been reported to have excellent reliability with an ICC of .79 in plasma (Yin et al., 2022) and .81 in serum (Floegel et al., 2011). Possible explanations for the lower reliability observed in the acylcarnitines group in the present study may be related to the collection of plasma rather than serum, and to using a general classification of acylcarnitines rather than classifying acylcarnitines into separate categories of short or long, respectively. Overall these findings suggest, in agreement with others (Agueusop et al., 2020; Breier et al., 2014; Floegel et al., 2011), that the reliability of circulating metabolite concentrations can vary considerably within metabolite groups, even when measured under controlled conditions. Establishing the reliability of individual metabolites and specific metabolite groups is important within analyses that aim to investigate differences between groups or time points by providing an indication of whether potential differences in abundance are likely to be robust.
For the analysis of between-group differences, multilevel PCA revealed no separation between participant groups, implying that the decomposed metabolomic profile was not different between groups. However, both visits from individual participants tended to cluster closely together, suggesting that the metabolomes from individual participants projected similarly between each visit, which agrees with the reliability data described above. The observation of no clear separation between groups using a multivariate method is in contrast to that reported by two other cross-sectional studies investigating the influence of exercise training history and performance characteristics on the circulating metabolome (Monnerat et al., 2020; Schranner et al., 2021). The first study compared elite long-distance runners with “low” (<65 ml·kg−1·min−1; n = 7) or “high” (>75 ml·kg−1·min−1; n = 7) values for maximal oxygen consumption (Monnerat et al., 2020), whereas the second study compared a control group (n = 4) and endurance (n = 6), sprint (n = 5), and bodybuilding (n = 4) athletes (Schranner et al., 2021). While both studies report clear separations between their athlete groups (Monnerat et al., 2020; Schranner et al., 2021), both studies also employed a partial-least squares discriminant analysis (PLSDA) modeling approach in contrast to the PCA approach employed in the present study. In contrast to PCA, PLSDA is a “supervised” method, wherein the model algorithm attempts to produce linear components that maximize the separation between predefined class structures (e.g., exercise training groups; Ruiz-Perez et al., 2020). This technical difference between methods is important because it means that PLSDA models will produce interclass separation, even in circumstances where no true class structure exists in the data (Ruiz-Perez et al., 2020). Therefore, the utility of a PLSDA model is based on a quality assessment of the model parameters (Szymańska et al., 2012). While neither model from the previous exercise training studies are reported as producing model overfitting (i.e., the training model over predicting the cumulative variance produced by the full data set; Monnerat et al., 2020; Schranner et al., 2021), both models did have evidence of underfitting (Q2 values greater than R2 of half of the full datasets across multiple permutations), which implies poor predictive power. In addition, the model produced by Schranner et al. (2021) explained a low cumulative variance (R2 = .5). Therefore, the contrasting results between the present study and these prior studies (Monnerat et al., 2020; Schranner et al., 2021) in terms of broad between-group differences may largely be due to differences in the selection of model algorithm(s).
When pairwise univariate testing was performed on our data, no individual metabolite differed between visits within each group, again implying that metabolites detected in the targeted metabolite panel were consistent between visits. However, univariate comparisons between groups, that is, END versus CON; STR versus CON; and END versus STR identified a combined total of 44 metabolites with differential abundance between groups. The largest number of metabolites with differential abundance was between END and CON (42 metabolites), with a smaller number different between STR and CON (10 metabolites), and END and STR (five metabolites). The mechanisms that explain these between-group differences remain to be elucidated, and discussion at present is speculative. For example, increased abundance of numerous lysophospholipids was observed between END and both other groups (10 lysophospholipids higher vs. CON; three higher vs. STR). In plasma, lysophospholipids can reflect the activity of lecithin–cholesterol acyltransferase, an enzyme that is abundant in plasma and serves to facilitate the transport of cholesterol esters between high-density lipoprotein particles and hepatic tissue (Glomset, 1968; Tan et al., 2020). Aerobic exercise training can increase the presence of circulating high-density lipoprotein (Kodama et al., 2007). In addition, cross-sectional studies involving athletes with endurance training backgrounds have noted increased lecithin–cholesterol acyltransferase activity at rest (Gupta et al., 1993; Tsopanakis et al., 1988) and this activity is significant in the context of reverse cholesterol transport (Leaf, 2003). Therefore, one speculation would be that the increased abundance of lysophospholipids observed in END may indeed be related to this adaptation to aerobic exercise training.
Other notable results include the observation that of all metabolites identified to be differentially abundant, nine (2-methylbutyric acid; 2-methylhexanonic acid; alpha linolenic acid; cis,8,11,14 eicosatrienoic acid; gamma linolenic acid; homovanillic acid; propionic acid; and tridecyclic acid) were common to both athlete groups compared with CON in addition to these differences being directionally the same in both END and STR. Interestingly, one metabolite—Homovanillic acid, an endpoint of dopamine metabolism and marker of metabolic stress (Amin et al., 1992)—was also observed to be lower in END and STR compared with CON, but lower again in END versus STR. Ultimately, elucidating the importance of these differentially abundant metabolites will require further work, as would establishing whether these differences are definitively a consequence of exercise training given that a cross-sectional design can only establish association, rather than causation.
In athletes or exercise contexts, the present study includes the largest sample size profiled to date and is the first study to our knowledge to report reliability data for individual metabolites but is not without limitations. Exercise training history and current performance status of participants was established largely through self-reported methods, and thus, objective measures of the physical fitness of participants (e.g., VO2max) are lacking. In addition, while our sample size is indeed the largest currently reported, some of our model parameters (e.g., the Kaiser–Meyer–Olkin score) are suggestive of us having low statistical power for multivariate modeling, a fact that may influence the results of the PCA. An important limitation that precludes the broader applicability of the results is that the groups comprised of only male participants, which is important given the influence of endogenous and exogenous sex hormones on metabolism, and that the responses of several genetic, metabolic, and physiological parameters to exercise differs between males and females (Ansdell et al., 2020; D’Eon et al., 2002; Fu et al., 2009; Landen et al., 2019; Maher et al., 2010). Lastly, we employed a targeted metabolomic approach; and therefore, our findings are restricted to only this subfraction of the circulating metabolome.
In conclusion, the present study found that when sampled under controlled measurement conditions, the resting plasma concentration of 166 metabolites and calculated metabolite sums and ratios did not differ overall between two visits, yet reliability was variable as evidenced by a large range in the average ICC within- and between-specific metabolite groups. In addition, divergent histories of exercise training were associated with alterations in the circulating metabolome at rest, but future work will be required to determine the importance of these differences, and whether these differences are definitively a consequence of adaptations to exercise training.
Acknowledgments
All authors declare no conflict of interest. None of the governing bodies providing funding to Darragh, O’Driscoll, or Egan have reviewed or provided input to the manuscript. Author Contributions: Conceptualization, methodology, formal analysis, investigation, writing – original draft: Darragh. Writing – review & editing, supervision, funding acquisition: O’Driscoll. Conceptualization, methodology, writing – review & editing; supervision, project administration, funding acquisition: Egan. Author Funding: This work was supported by funding from The Irish Research Council through both the Government of Ireland Postgraduate Scholarship Programme to IAJD and BE (grant number: GOIPG/2020/162), and an Irish Research Council Advanced Laureate Award to LOD (grant number: IRCLA/2019/49). The Irish Research Council is an associated agency of the Department of Education and Skills and operates under the aegis of the Higher Education Authority of Ireland.
References
Agueusop, I., Musholt, P.B., Klaus, B., Hightower, K., & Kannt, A. (2020). Short-term variability of the human serum metabolome depending on nutritional and metabolic health status. Scientific Reports, 10, Article 16310. https://doi.org/10.1038/s41598-020-72914-7
Amin, F., Davidson, M., & Davis, K.L. (1992). Homovanillic acid measurement in clinical research: A review of methodology. Schizophrenia Bulletin, 18, 123–148. https://doi.org/10.1093/schbul/18.1.123
Ansdell, P., Thomas, K., Hicks, K.M., Hunter, S.K., Howatson, G., & Goodall, S. (2020). Physiological sex differences affect the integrative response to exercise: Acute and chronic implications. Experimental Physiology, 105, 2007–2021. https://doi.org/10.1113/EP088548
Belhaj, M.R., Lawler, N.G., & Hoffman, N.J. (2021). Metabolomics and lipidomics: Expanding the molecular landscape of exercise. Biology, 11, Article 151. https://doi.org/10.3390/metabo11030151
Breier, M., Wahl, S., Prehn, C., Fugmann, M., Ferrari, U., Weise, M., Banning, F., Seissler, J., Grallert, H., Adamski, J., & Lechner, A. (2014). Targeted metabolomics identifies reliable and stable metabolites in human serum and plasma samples. PLoS One, 9, Article e89728. https://doi.org/10.1371/journal.pone.0089728
Cicchetti, D.V. (1994). Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychological Assessment, 6, 284–290. https://doi.org/10.1037/1040-3590.6.4.284
Coffey, V.G., & Hawley, J.A. (2017). Concurrent exercise training: Do opposites distract? Journal of Physiology, 595, 2883–2896. https://doi.org/10.1113/JP272270
D’Eon, T.M., Sharoff, C., Chipkin, S.R., Grow, D., Ruby, B.C., & Braun, B. (2002). Regulation of exercise carbohydrate metabolism by estrogen and progesterone in women. American Journal of Physiology—Endocrinology and Metabolism, 283, E1046–E1055. https://doi.org/10.1152/ajpendo.00271.2002
Darragh, I.A.J., O’Driscoll, L., & Egan, B. (2021). Exercise training and circulating small extracellular vesicles: Appraisal of methodological approaches and current knowledge. Frontiers in Physiology, 12, Article 1894. https://doi.org/10.3389/fphys.2021.738333
Degens, H. (2019). Physiological comparison between non-athletes, endurance, power and team athletes. European Journal of Applied Physiology, 119, 1377–1386. https://doi.org/10.1007/s00421-019-04128-3
Dunn, W.B., Broadhurst, D.I., Atherton, H.J., Goodacre, R., & Griffin, J.L. (2011). Systems level studies of mammalian metabolomes: The roles of mass spectrometry and nuclear magnetic resonance spectroscopy. Chemical Society Reviews, 40, 387–426. https://doi.org/10.1039/B906712B
Egan, B., & Zierath, J.R. (2013). Exercise metabolism and the molecular regulation of skeletal muscle adaptation. Cell Metabolism, 17, 162–184. https://doi.org/10.1016/j.cmet.2012.12.012
Esco, M.R., Fedewa, M.V., Freeborn, T.J., Moon, J.R., Wingo, J.E., Cicone, Z., Holmes, C.J., Hornikel, B., & Welborn, B. (2019). Agreement between supine and standing bioimpedance spectroscopy devices and dual-energy X-ray absorptiometry for body composition determination. Clinical Physiology and Functional Imaging, 39, 355–361. https://doi.org/10.1111/cpf.12585
Feunang, Y.D., Eisner, R., Knox, C., Chepelev, L., Hastings, J., Owen, G., Fahy, E., Steinbeck, C., Subramanian, S., Bolton, E., Greiner, R., & Wishart, D.S. (2016). ClassyFire: Automated chemical classification with a comprehensive, computable taxonomy. Journal of Cheminformatics, 8, Article 61. https://doi.org/10.1186/s13321-016-0174-y
Floegel, A., Drogan, D., Wang-Sattler, R., Prehn, C., Illig, T., Adamski, J., Joost, H.-G., Boeing, H., & Pischon, T. (2011). Reliability of serum metabolite concentrations over a 4-month period using a targeted metabolomic approach. PLoS One, 6, Article e21103. https://doi.org/10.1371/journal.pone.0021103
Fu, M.H., Maher, A.C., Hamadeh, M.J., Ye, C., & Tarnopolsky, M.A. (2009). Exercise, sex, menstrual cycle phase, and 17β-estradiol influence metabolism-related genes in human skeletal muscle. Physiological Genomics, 40, 34–47. https://doi.org/10.1152/physiolgenomics.00115.2009
Glomset, J.A. (1968). The plasma lecithins: Cholesterol acyltransferase reaction. Journal of Lipid Research, 9, 155–167.
Gupta, A.K., Ross, E.A., Myers, J.N., & Kashyap, M.L. (1993). Increased reverse cholesterol transport in athletes. Metabolism, 42, 684–690. https://doi.org/10.1016/0026-0495(93)90233-E
Hawley, J.A., Hargreaves, M., Joyner, M.J., & Zierath, J.R. (2014). Integrative biology of exercise. Cell, 159, 738–749. https://doi.org/10.1016/j.cell.2014.10.029
Kaiser, H.F., & Rice, J. (1974). Little Jiffy Mark IV. Educational and Psychological Measurement, 34, 111–117. https://doi.org/10.1177/001316447403400115
Khoramipour, K., Sandbakk, Ø., Keshteli, A.H., Gaeini, A.A., Wishart, D.S., & Chamari, K. (2022). Metabolomics in exercise and sports: A systematic review. Sports Medicine, 52, 547–583. https://doi.org/10.1007/s40279-021-01582-y
Kodama, S., Tanaka, S., Saito, K., Shu, M., Sone, Y., Onitake, F., Suzuki, E., Shimano, H., Yamamoto, S., Kondo, K., Ohashi, Y., Yamada, N., & Sone, H. (2007). Effect of aerobic exercise training on serum levels of high-density lipoprotein cholesterol: A meta-analysis. Archives of Internal Medicine, 167, 999–1008. https://doi.org/10.1001/archinte.167.10.999
Landen, S., Voisin, S., Craig, J.M., McGee, S.L., Lamon, S., & Eynon, N. (2019). Genetic and epigenetic sex-specific adaptations to endurance exercise. Epigenetics, 14, 523–535. https://doi.org/10.1080/15592294.2019.1603961
Leaf, D.A. (2003). The effect of physical exercise on reverse cholesterol transport. Metabolism, 52, 950–957. https://doi.org/10.1016/S0026-0495(03)00147-1
Li-Gao, R., Hughes, D.A., le Cessie, S., de Mutsert, R., den Heijer, M., Rosendaal, F.R., Willems van Dijk, K., Timpson, N.J., & Mook-Kanamori, D.O. (2019). Assessment of reproducibility and biological variability of fasting and postprandial plasma metabolite concentrations using 1H NMR spectroscopy. PLoS One, 14, Article e0218549. https://doi.org/10.1371/journal.pone.0218549
Liquet, B., Cao, K.-A.L., Hocini, H., & Thiébaut, R. (2012). A novel approach for biomarker selection and the integration of repeated measures experiments from two assays. BMC Bioinformatics, 13, Article 325. https://doi.org/10.1186/1471-2105-13-325
Maher, A.C., Akhtar, M., & Tarnopolsky, M.A. (2010). Men supplemented with 17β-estradiol have increased beta-oxidation capacity in skeletal muscle. Physiological Genomics, 42, 342–347. https://doi.org/10.1152/physiolgenomics.00016.2010
McKay, A.K.A., Stellingwerff, T., Smith, E.S., Martin, D.T., Mujika, I., Goosey-Tolfrey, V.L., Sheppard, J., & Burke, L.M. (2022). Defining training and performance caliber: A participant classification framework. International Journal of Sports Physiology and Performance, 17, 317–331. https://doi.org/10.1123/ijspp.2021-0451
Monnerat, G., Sánchez, C.A.R., Santos, C.G.M., Paulucio, D., Velasque, R., Evaristo, G.P.C., Evaristo, J.A.M., Nogueira, F.C.S., Domont, G.B., Serrato, M., Lima, A.S., Bishop, D., Campos de Carvalho, A.C., & Pompeu, F.A.M.S. (2020). Different signatures of high cardiorespiratory capacity revealed with metabolomic profiling in elite athletes. International Journal of Sports Physiology and Performance, 15, 1156–1167. https://doi.org/10.1123/ijspp.2019-0267
Nicholson, J.K., & Lindon, J.C. (2008). Metabonomics. Nature, 455, 1054–1056. https://doi.org/10.1038/4551054a
Patti, G.J., Yanes, O., & Siuzdak, G. (2012). Metabolomics: The apogee of the omics trilogy. Nature Reviews Molecular Cell Biology, 13, 263–269. https://doi.org/10.1038/nrm3314
Petersen, A.-K., Krumsiek, J., Wägele, B., Theis, F.J., Wichmann, H.-E., Gieger, C., & Suhre, K. (2012). On the hypothesis-free testing of metabolite ratios in genome-wide and metabolome-wide association studies. BMC Bioinformatics, 13, Article 120. https://doi.org/10.1186/1471-2105-13-120
Rohart, F., Gautier, B., Singh, A., & Cao, K.-A.L. (2017). mixOmics: An R package for ‘omics feature selection and multiple data integration. PLoS Computational Biology, 13, Article e1005752. https://doi.org/10.1371/journal.pcbi.1005752
Ruiz-Perez, D., Guan, H., Madhivanan, P., Mathee, K., & Narasimhan, G. (2020). So you think you can PLS-DA? BMC Bioinformatics, 21, Article 2. https://doi.org/10.1186/s12859-019-3310-7
Schranner, D., Schönfelder, M., Römisch-Margl, W., Scherr, J., Schlegel, J., Zelger, O., Riermeier, A., Kaps, S., Prehn, C., Adamski, J., Söhnlein, Q., Stöcker, F., Kreuzpointner, F., Halle, M., Kastenmüller, G., & Wackerhage, H. (2021). Physiological extremes of the human blood metabolome: A metabolomics analysis of highly glycolytic, oxidative, and anabolic athletes. Physiological Reports, 9, Article e14885. https://doi.org/10.14814/phy2.14885
Shrout, P.E., & Fleiss, J.L. (1979). Intraclass correlations: Uses in assessing rater reliability. Psychological Bulletin, 86, 420–428. https://doi.org/10.1037//0033-2909.86.2.420
Szymańska, E., Saccenti, E., Smilde, A.K., & Westerhuis, J.A. (2012). Double-check: Validation of diagnostic statistics for PLS-DA models in metabolomics studies. Metabolomics, 8, 3–16. https://doi.org/10.1007/s11306-011-0330-3
Tan, S.T., Ramesh, T., Toh, X.R., & Nguyen, L.N. (2020). Emerging roles of lysophospholipids in health and disease. Progress in Lipid Research, 80, Article 101068. https://doi.org/10.1016/j.plipres.2020.101068
Tsopanakis, C., Kotsarellis, D., & Tsopanakis, A. (1988). Plasma lecithin: Cholesterol acyltransferase activity in elite athletes from selected sports. European Journal of Applied Physiology, 58, 262–265. https://doi.org/10.1007/BF00417260
Yin, X., Prendiville, O., McNamara, A.E., & Brennan, L. (2022). Targeted metabolomic approach to assess the reproducibility of plasma metabolites over a four month period in a free-living population. Journal of Proteome Research, 21, 683–690. https://doi.org/10.1021/acs.jproteome.1c00440
Zheng, J., Zhang, L., Johnson, M., Mandal, R., & Wishart, D.S. (2020). Comprehensive targeted metabolomic assay for urine analysis. Analytical Chemistry, 92, 10627–10634. https://doi.org/10.1021/acs.analchem.0c01682
