Skip to main content
NCBI home page
Springer logoLink to Springer
. 2026 Sep 23;22(5):158. doi: 10.1007/s11306-026-02530-5

Urinary metabolomic signatures of muscle injury and recovery in elite football players

Guillermo Quintás 1,✉, Ricard Pruna 2,3, Manuel Wong 2,3, Sandra Mechó 2,3, Pau Madrero 2,3, Juan Daniel Sanjuán-Herráez 1, Gil Rodas 2,3,✉
PMCID: PMC13601168  PMID: 42776338

Abstract

Introduction

Muscle injuries are a leading cause of time-loss in elite football, highlighting the need for objective biomarkers capable of reflecting the athlete’s physiological state. Monitoring strategies rely predominantly on external load metrics and may not fully capture internal physiological responses to training and competition. Urinary metabolomics offers a non-invasive approach to characterize injury related physiological processes.

Objectives

To identify urinary metabolomic signatures associated with muscle injury in elite football players and to explore their potential utility for monitoring injury-related physiological stress and recovery.

Methods

This observational longitudinal study included 287 urine samples collected from 121 elite male football players across two consecutive seasons. Samples were clinically classified as muscle injury (n = 30) or control (n = 257). Targeted UPLC-MS quantified amino acids and tryptophan-related metabolites. Univariate analysis, pathway over-representation analysis, and PLS-DA were applied to identify injury-associated metabolic patterns

Results

Muscle injury was associated with alterations in amino acid turnover, purine metabolism, energetic stress, and tryptophan catabolism. Exploratory pathway analysis suggested enrichment of amino acid-related pathways, although interpretation was limited by the targeted nature of the analytical panel. PLS-DA identified an injury-associated metabolic signature with a cross-validated AUROC of 0.88 (sensitivity 73%, specificity 88%; p < 0.003). Key discriminant metabolites included β-aminoisobutyric acid, hypoxanthine, xanthurenic acid, β-alanine, and alanine. Exploratory visual assessment of individual longitudinal trajectories suggested a return toward baseline during rehabilitation.

Conclusions

Urinary metabolomics identifies coordinated metabolic alterations associated with muscle injury in elite football players, supporting its potential as a future complementary tool to external load monitoring. Further validation in larger cohorts is required before these systemic signatures can be applied to clinical monitoring or recovery assessment.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11306-026-02530-5.

Keywords: Muscle injury, Urinary metabolomics, Pathway analysis, Multivariate analysis

Introduction

The design of training programs critically depends on the effective management of external loads (e.g., running distance, speed) which are typically guided by predefined objectives or adjusted according to feedback from internal physiological responses. These include physiological parameters (e.g., heart rate recovery and blood lactate levels) and subjective indicators such as perceived exertion, fatigue, and overall wellness. However, current athlete monitoring strategies rely predominantly on external load metrics derived from electronic performance tracking systems (EPTS) (Akenhead and Nassis, 2016), which do not fully capture the individual physiological stress and biological cost of training and competition. This is indeed one of the major challenges in sports medicine because the biological response to a given training load is highly individual and influenced by a complex network of metabolic, inflammatory, and recovery-related processes. Metabolomics has emerged as a promising tool to bridge this gap by providing a snapshot of the athlete systemic biological status. Complementary to conventional monitoring systems (EPTS, fatigue, etc.), metabolomic profiling could reveal subtle biochemical alterations associated with excessive muscle stress and tissue damage, even before clinical manifestations, offering a potentially transformative approach for injury surveillance in professional athletes (Khoramipour et al., 2022; Rodas et al., 2022; González et al., 2024; Schoeny et al., 2025; Rodas et al., 2025).

Despite this careful personalized monitoring, muscle injuries linked to suboptimal physiological adaptations to external loads are one of the main causes of time-loss and performance impairment in elite football. A better characterization of the internal load and physiological stress experienced by the players could complement external load surrogates, thereby improving the personalization of training programs and optimizing adaptation and recovery to training and match demands (Arcos et al., 2017; Bourdon et al., 2017).

The metabolic demands of contracting skeletal muscles during physical activity trigger adaptations across genomic, proteomic, and metabolic levels within a complex signaling network (Hawley et al., 2014; Impellizzeri et al., 2019). Metabolomics provides a downstream functional readout of physiological adaptations integrating genomic, transcriptomic, proteomic, and environmental influences. As metabolites represent the final products of cellular regulatory processes, metabolomic profiling may offer a sensitive approach for characterizing the biological responses associated with training load, recovery, and tissue injury. In this context, metabolomics has gained interest in the field of sports physiology in recent years, as it might provide a particularly informative perspective.

Previous studies of the effects of physical exercise on metabolism across both long-term adaptations and immediate, short-term responses have reported changes key metabolic processes including ATP metabolism, glycolysis, beta-oxidation of free fatty acids and ketone bodies, as well as the association between long-term physical exercise and the upregulation of antioxidant systems (Finaud et al, 2006; Gorostiaga et al. 2012;Pechlivanis et al., 2015;Duft et al., 2017; Heaney et al., 2017; Manaf et al., 1985), Furthermore, the analysis of the association between urinary metabolomic profiles and EPTS-derived metrics identified alterations in steroid hormones, hypoxanthine-related metabolites, tyrosine, tryptophan, and riboflavin pathways as well as in amino acids, and intermediates involved in phenylalanine metabolism (Quintas et al, 2020).

Acute metabolic adaptations triggered by strenuous physical exertion involve a deep shift in skeletal muscle bioenergetics. This process is predominantly characterized by an acceleration of the ATP-phosphocreatine and anaerobic glycolytic pathways, alongside concurrent elevations in blood lactate production and purine degradation cascades (Enea et al., 2010). In adolescent male football players, exercise-induced fatigue has been demonstrated to alter interconnected biochemical networks, including nitrogen, tyrosine, and glycerophospholipid pathways, as well as the tricarboxylic acid (TCA) cycle and glycine-serine-threonine metabolism (Cao et al, 2020). Moreover, variations in cumulative training load and the implementation of distinct training regimens significantly modulate the systemic metabolome. For instance, an evaluation of metabolic variation in samples collected from football players both immediately after and 20 h following competitive matches revealed distinct metabolomic changes correlated with rating of perceived exertion (RPE) scores. These variations were localized to markers reflecting localized cellular strain (such as creatinine, creatine, and glycine) and metabolic energy flux (including pyruvate, formate, 1,3-dihydroxyacetone, and creatine) (Marinho et al., 2022). Taken together, these collective insights underscore the utility of monitoring metabolomic changes as a suitable strategy for discovering novel biomarkers indicative of workload adaptation and physiological stress during training (Neal et al., 2013).

Previous studies have mainly relied on univariate statistical analysis of the association between external load with metabolic changes, and between changes in metabolic levels and pathways (García-Campos et al., 2015; Nguyen et al., 2019). However, metabolites are often co-regulated within biological networks and univariate statistics might not capture multivariate responses. To address this limitation, multivariate techniques such as Partial Least Squares (PLS) (Saccenti et al., 2014; Olivieri, 2008) has been proposed to identify biological processes. The identification of specific subsets of metabolites or pathways of interest could be used for advancing targeted metabolomics methods required for facilitating robust validation in clinical settings, improving the feasibility of clinical tests, and accelerating their translation into routine practice to support personalized monitoring and training programs.

We hypothesized that urinary metabolomic profiles may capture the biological internal load associated with muscle injury and recovery, providing complementary information to structural imaging and conventional athlete monitoring systems. This study aimed to evaluate urinary metabolomic profiling as a non-invasive tool for monitoring internal physiological load in elite football players and to identify metabolic signatures associated with muscle injury. In addition, we explored the feasibility of developing a metabolite-based non-invasive score capable of reflecting injury-related physiological stress and recovery dynamics during rehabilitation.

Materials and methods

Subjects and experimental approach

FC Barcelona carried out this observational, longitudinal study following relevant guidelines and regulations as previously described (Quintas et al., 2020). Institutional board approval for the study was obtained from the Ethics Commission of the Consell Català de l’Esport (Code 03/2019/CEICEGC, Generalitat de Catalunya, Barcelona, Spain). Urine samples and lesion status were collected from volunteers that gave written informed consent to participate in the study. To ensure confidentiality, all performance and biochemical data were pseudo-anonymized. The sample set included 287 first morning urine samples from 121 male professional football players collected throughout 2 consecutive seasons. Due to the observational nature of the study, the number of samples provided per player varied. A total of 30 injury associated samples were collected during 10 active muscle injury and rehabilitation periods from 10 distinct players, 8 of whom provided longitudinal serial samples (2 to 6 time points per lesion). All injured players were part of the 121-player cohort and also contributed non-injured control samples during healthy baseline periods. Samples were collected at a median of 22 days post-injury (IQR: 10–35 days; range: 0–58 days). 5 samples (16.7%) were collected during the acute phase (0–7 days), 9 (30.0%) during early rehabilitation (8–21 days), 12 (40.0%) during late rehabilitation (22–42 days), and 4 (13.3%) during extended rehabilitation/near RTP (> 42 days).

The remaining 257 samples were categorized as non-injured controls to capture the baseline metabolic variability in this cohort. Although the study population was imbalanced, the larger control cohort was intentionally retained to capture the substantial biological and environmental variability inherent to elite players. Urinary metabolomic profiles are influenced by physiological and behavioral factors and so, a broad representation of non-injured athletes was considered necessary to adequately characterize the reference metabolic space against which injury-associated alterations could be identified. Urine samples were collected, aliquoted and stored at −80 °C until analysis.

Muscle injury assessment

Muscle injuries were defined and recorded according to the UEFA consensus recommendations for injury surveillance in professional football. An injury was considered any physical complaint sustained during training or match play that resulted in the player being unable to fully participate in future football activities (time-loss definition) (Fuller et al., 2006). All diagnoses were made by the same medical staff throughout the study period to ensure consistency. Clinical evaluation was supported by imaging techniques, including ultrasound and/or magnetic resonance imaging (MRI), when clinically indicated. Diagnosis, prognosis, treatment, rehabilitation, and return-to-play decisions followed the FC Barcelona muscle injury management guidelines (Barça Innovation Hub, 2018). For players sustaining muscle injuries, MRI examinations were performed within 24 h after injury and subsequently repeated during the early (2–3 weeks), intermediate (4–6 weeks), and late (6–8 weeks) phases of healing to assess connective tissue repair, scar maturation, restoration of tissue tension, and edema evolution. MRI interpretation and healing assessment followed the imaging criteria and muscle healing framework previously described (Isern-Kebschull et al., 2024).

MRI examinations were performed using a 3-T scanner (Vantage Galan; Canon Medical Systems, Tochigi, Japan). The follow-up protocol for muscle injury healing included axial PDFS, coronal PDFS, and axial T1-weighted sequences. Imaging parameters were as follows: axial PDFS (TR/TE, 2900/33 ms; FOV, 23 × 23 cm; matrix, 352 × 352; slice thickness, 3.5 mm), coronal PDFS (TR/TE, 2700/44 ms; FOV, 25 × 25 cm; matrix, 256 × 304; slice thickness, 2.5 mm; gap, 0.4 mm), and axial T1-weighted imaging (TR/TE, 740/10 ms; FOV, 22 × 22 cm; matrix, 352 × 416; slice thickness, 3.5 mm).

Characteristics of injury-associated samples

Injury diagnosis and follow-up were performed according to the institutional muscle injury protocol. Muscle injuries involved different anatomical locations and severities (see Supplementary Table 1), representing the spectrum of lesions commonly observed in elite professional football. The analyzed injury-associated urine samples included grade 2b distal biceps femoris strain (11 samples), grade 3c proximal biceps femoris strain (6 samples), grade 2b proximal semimembranosus strain (6 samples), grade 2b proximal rectus femoris strain (3 samples), grade 2b proximal biceps femoris strain (2 samples), and grade 2b proximal semitendinosus strain (2 samples).

Analysis of amino acids

Samples were thawed, vortexed for 15 s and centrifuged at 10000 × g for 10 min at 4 °C. 50 µL of samples were mixed with 300 µL H2O. Then, samples were derivatized (AccQTag Ultra, Waters) using the following procedure: 10 µL of diluted sample are transferred to a 1.5 mL Eppendorf vial and mixed with 70 µL borate buffer. After mixing (vortex, 15 s), 20 µL of 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate are added. The sample was vortexed (15 s) and incubated at 55 °C for 10 min. Then, the sample is transferred to a 96-well plate for UPLC-MSMS analysis within 24 h. Blanks were prepared replacing the sample by MilliQ water. Quantitative analysis of the following amino acids by UPLC-MSMS was carried out on a 1290 Infinity system from Agilent Technologies Inc. (Waldbronn, Germany), with an Agilent 6460 Triple quadrupole mass spectrometer for MRM-detection following derivatization (AccQ-Tag, Waters). Autosampler and column temperatures were set at 6 and 55 °C during sample analysis, respectively. The mobile phases were A (H2O, 0.1% v/v HCOOH) and B (CH3CN, 0.1% v/v HCOOH), and the flow rate was 500 µl/min. The initial condition (1% B) was held up to 1 min. From 1 to 2 min the proportion of solvent B was linearly increased up to 13%. Then, B increased linearly to reach 15% at 5.5 min, and then up to 95% at 6.5 min. This composition was kept for 1 min, and then it was returned to its initial conditions (1% B) in 0.1 min. The column was equilibrated for 3 min before injection of the next sample. The Multiple Reaction Monitoring (MRMs) used for aminoacid quantification by UPLC-MSMS were the following: Histidine (MRM 326.2 > 171.1), Phosphoserine (MRM 357.2 > 171.1), Hydroxyproline (MRM 302.2 > 171.1), 1,3-Methyl Histidine (MRM 340.2 > 171.1), Arginine (MRM 345.2 > 171.1), Carnosine (MRM 397.2 > 171.1), Asparagine (MRM 303.2 > 171.1), Phosphoethanolamine (MRM 312.2 > 171.1), Anserine (MRM 411.2 > 171.1), Glutamine (MRM 317.1 > 171.1), Taurine (MRM 296.2 > 171.1), Serine (MRM 276.2 > 171.1), Ethanolamine (MRM 232.2 > 171.1), Glycine (MRM 246.2 > 171.1), Aspartic acid (MRM 304.2 > 171.1), Citrulline (MRM 346.2 > 171.1), Sarcosine (MRM 260.2 > 171.1), Glutamic acid (MRM 318.2 > 171.1), Threonine (MRM 290.2 > 171.1), beta-Alanine (MRM 260.2 > 171.1), Hydroxylysine (MRM 252.1 > 171.1), Alanine (MRM 260.2 > 171.1), GammaaminoButyric Acid (GABA) (MRM 274.2 > 171.1), Aminoadipic acid (MRM 333.2 > 171.1), Ornithine (MRM 273.1 > 171.1), Proline (MRM 286.2 > 171.1) beta-Amino isobutyric acid (BAIBA) (MRM 274.2 > 171.1), Cystathionine (MRM 282.1 > 171.1), Lysine (MRM 244.1 > 171.1), Cystine (MRM 291.2 > 171.1), Alfa-AminoButyric Acid (AABA) (MRM 274.2 > 171.1), Tyrosine (MRM 352.2 > 171.1), Methionine (MRM 320.2 > 171.1), Valine (MRM 288.2 > 171.1), Homocystine (MRM 305.2 > 171.1), Isoleucine (MRM 302.2 > 171.1), Leucine (MRM 302.2 > 171.1), Phenylalanine (MRM 336.2 > 171.1), Tryptophan (MRM 375.2 > 171.1). The following internal standards were included in the analysis: Histidine-13C6-15N3 (MRM 335.2 > 171.1) (Histidine), Serine-13C3-15N (MRM 280.2 > 171.1) (Phosphoserine, Taurine, Serine), Proline-13C5-15N (MRM 292.2 > 171.1) (Hydroxyproline), Histidine-13C6-15N3 (MRM 335.2 > 171.1) (Methylhistidine), Arginine-13C6-15N4 (MRM 355.2 > 171.1) (Arginine, Carnosine, Asparagine, Anserine), Glycine-13C2-15N (MRM 249.2 > 171.1) (Phosphoethanolamine, Ethanolamine, Glycine, gamma-Aminobutyric acid, beta-Aminoisobutyric acid, alpha-Aminoisobutyric acid), Glutamic acid-13C5-15N (MRM 324.2 > 171.1) (Glutamine, Glutamic acid, Cystathionine), Aspartic acid-13C4-15N (MRM 309.2 > 171.1) (Aspartic acid, Citrulline, Aminoadipic acid), Alanine-13C3-15N (MRM 264.2 > 171.1) (Sarcosine, b-Alanine, Alanine), Threonine-13C4-15N (MRM 295.2 > 171.1) (Threonine), Lysine-13C6-15N2 (MRM 248.1 > 171.1) (Hydroxylysine, Ornithine, Lysine), Proline-13C5-15N (MRM 292.2 > 171.1) (Proline), Cystine-13C6-15N2 (MRM 295.1 > 171.1) (Cystine, Homocystine), Tyrosine-13C9-15N (MRM 362.2 > 171.1) (Tyrosine), Methionine-13C5-15N (MRM 326.2 > 171.1) (Methionine), Valine-13C5-15N (MRM 294.2 > 171.1) (Valine), Isoleucine-13C6-15N (MRM 309.2 > 171.1) (Isoleucine), Leucine—13C6-15N (MRM 309.2 > 171.1) (Leucine), Phenylalanine-13C9-15N (MRM 346.2 > 171.1) (Phenylalanine, Tryptophan).

Quantification of the compounds in samples and blanks was performed with MassHunter Quantitative Analysis QQQ (Agilent Technologies Inc.). Standards were used to construct linear regression curves using the ratio of analyte area/internal standard area against the corresponding concentrations of the calibrators. MRM transition specificity, retention time matching within ± 0.05 min relative to authentic standards, and calibration curve linearity (R2 > 0.95) were monitored across batches. Pooled biological QC samples were not included. Analytical performance was instead monitored using calibration standards, procedural blanks, retention-time matching, calibration-curve linearity, and stable isotope-labelled internal standards added to each sample.

Analysis of tryptophan-related metabolites

Samples were thawed at room temperature, vortexed for 15 s and centrifuged at 10000 × g for 15 min at 4 °C. 50 µL of samples were placed into 1.5 mL Eppendorf tubes and mixed with 50 µL of an internal standard solution of CH3CN (0.1% v/v HCOOH) containing hydroxytryptophan-D4, L-Kynurenine-D4, indole-D5−3-acetamide, 6-hydroxymelatonin-D4, kynurenic acid-D5, PAGN-D5, serotonin-D4, tryptamine-D4, tryptophan-D5, xanthurenic acid-D4 and phenylalanine-D5 (900 nM each). Samples are mixed (vortex, 15 s) and transferred to a 96-well plate for UPLC-MSMS analysis. Blanks were prepared replacing the sample by MilliQ water. The quantification of the set of metabolites was carried out on an Acquity-Xevo TQS (Waters) triple quadrupole system equipped with an electrospray ionization source using a previously developed method (Lario et al., 2017). Samples were analysed using an Acquity HSS T3 C18 (100 × 2.1 mm, 1.8 μm) column. Mobile phases were H2O (0.1% v/v HCOOH) (A) and (0.1% v/v HCOOH) CH3CN (B). The gradient elution was as follows: phase B was held 2% from 0 to 0.5 min, then increased linearly to 45% over the following 5 min. Then phase B was increased to 90% in 0.2 min followed by a fast return to initial conditions between 5.7 and 6 min, which were held for 1.5 min for column re-equilibration. Injection volume, flow rate and column temperature were set at 3 µL, 550 µL/min and 55 °C, respectively. Autosampler temperature was set at 6 °C during sample analysis. Electrospray ionization was carried out using the following conditions: capillary 2.9 kV, cone 25 V, source temperature 120 °C, desolvation temperature 395 °C, N2 cone and desolvation gas flow rates were 150 and 800 L/h, respectively. Data acquisition and manual integration of standards and internal standards was carried out using MassLynx (Waters) software. The MRMs in the analysis were the following: 3-Hydroxykynurenine (225.1 > 110), 5-Hydroxytryptophan (221.1 > 162.2), 5-Hydroxytryptophan-D4 (225 > 208), 6-hydroxymelatonin-D4 (253 > 194), Aminophenol (110.14 > 92), Anthranilic acid (137.89 > 120), Guanosine (284 > 152), Guanine (298.9 > 136), Hypoxanthine (137 > 110), Indole-3-acetamide (175.1 > 103.4), indole-D5−3-acetamide (180 > 133), Indolelactic acid (204 > 116), Kynurenic acid (190 > 89), Kynurenic acid-D5 (195 > 149), Kynurenine (209 > 94), Kynurenine-D4 (213 > 98), N-Formylkynurenine (237.1 > 136), Phenylacetylglutamine (263 > 127), Phenylacetylglutamine-D5 (268 > 145), S-adenosylhomocysteine (385.1 > 134), S-adenosylmethionine (298.9 > 136), Serotonine (177 > 115), Serotonine-D4 (181 > 164), Tryptamine (161 > 122), Tryptamine-D4 (165 > 148), Tryptophan (205 > 118), Tryptophan-D5 (210 > 193), Xanthurenic Acid (206.1 > 132), Xanthurenic acid-D4 (210 > 164). Quantification of the compounds in samples and blanks was performed with MassLynx (Waters). Standards were used to construct linear regression curves using the ratio of analyte area/internal standard area against the corresponding concentrations of the calibrators. MRM transition specificity, retention time matching within ± 0.05 min relative to authentic standards, and calibration curve linearity (R2 > 0.95) were monitored across batches. Pooled biological QC samples were not included. Instrumental stability and signal response were monitored across analytical runs using stable isotope-labeled internal standards.

Statistical analysis

Target metabolites were initially quantified as absolute concentrations (µM) using standard calibration curves and isotope-labeled internal standards. For downstream statistical modeling, absolute concentrations were transformed into relative normalized variables to account for urinary dilution (56 metabolite-to-creatinine ratios) and specific pathway turnover dynamics (15 metabolite-to-tryptophan ratios of Aminophenol, Anthranilic acid, Hydroxyanthranilic acid, Tryptamine, Indole-3-acetamide, Serotonine, Kynurenic acid, Xanthurenic acid, Kynurenine, 5-Hydroxytryptophan, 3-Hydroxykynurenoine, N-Formylkynurenine, Phenylacetylglutamine, Indolelactic acid, and 3-Indolelactic acid)). Univariate and multivariate analysis was carried out in MATLAB (Mathworks Inc., Natick, MA, USA). Principal Component Analysis (PCA) and PLS regression models were developed using PLS Toolbox 9.5 (Eigenvector Research Inc., Wenatchee, WA, USA) and MATLAB 2023b scripts. The number of latent variables of the PLS models was chosen based on the classification error estimated using cross validation (CV). Given the limited number of injury samples available (n = 30), leave-one-out cross-validation (LOOCV) was used to maximize the amount of information available for model training while still providing an internal estimate of discrimination performance. Under LOOCV, each sample is excluded once and predicted using a model developed from all remaining samples, thereby minimizing the loss of training data associated with more restrictive partitioning schemes such as leave-one player-out or random k-fold CV. This consideration was particularly relevant in the present study because muscle injury samples represented a relatively small fraction of the dataset. The primary objective of the multivariate analysis was to identify metabolomic signatures associated with the injury state rather than to evaluate generalization to entirely unseen athletes. The statistical significance of figures of merit estimated by CV was assessed using permutation testing. The permutation test involved the random permutation of the vector of labels (control, injury), the subsequent development of a PLS-DA model using the permuted class labels, and the estimation of the mean classification error of CV (CVperm) of the PLS model (Rubingh et al., 2006). This step was repeated 300 times, to obtain a null distribution of CVperm performance values that was used to compute the p-value for the performance estimation obtained using the true class labels (CVinit), as the fraction of permuted statistics CVperm < CVinit. Permutation testing was performed by randomly permuting sample class labels. Although multiple samples were available from some athletes, player-level permutation strategies were not considered appropriate because muscle injury represented a transient condition and most players contributed only non-injury samples, which would substantially restrict the number of valid subject-level permutations and result in an uninformative null distribution.

Backward stepwise elimination was used as a feature selection algorithm by excluding iteratively individual features one by one to maximize model performance. First, the search algorithm builds a PLS-DA model where the number of latent variables is selected based on the estimated classification error as described before by LOOCV. Then, the Variable Importance in the Projection (VIP) score vector is calculated, and the variable showing the lowest VIP score is identified and excluded from the selected variable subset. The process is iterated until the size of the selected variable subset is lower than the maximum number of latent variables evaluated (3 in this study). Metabolic functional analysis was carried out using the mummichog algorithm (Li et al., 2013; Pang et al., 2022) available in MetaboAnalyst 6.0 (Pang et al., 2024).

Results and discussion

Overview of the data set

A total of 287 urine samples collected from 121 elite male football players across two consecutive seasons were analyzed using targeted UPLC-MS metabolomics for the quantification of 56 metabolic ratios after normalization by creatinine concentration and 15 after normalization by Trp concentrations. Samples were categorized into two clinical conditions: muscle injury (n = 30) and non-injured controls (n = 257).

Prior to supervised statistical analysis, principal component analysis (PCA) was performed to assess the global structure of the dataset, identify potential analytical outliers, and evaluate the major sources of variance. The PCA model showed the overlap between samples obtained during muscle injury periods and samples collected during non-injured control conditions (Fig. 1). No clustering according to injury status was observed in the two first principal components, indicating that muscle injury was not among the main sources of variance within the metabolic dataset. No analytical outliers were detected, supporting the robustness of the sample collection and analytical workflows. In addition, the distribution of samples according to season or year of collection did not reveal systematic batch-related separation, suggesting that inter-season variability had a limited influence on the data structure.

Fig. 1.

Fig. 1

Open in a new tab

PCA of the full data set. Scores plot obtained from PCA analysis of urinary metabolomic profiles collected from elite football players during muscle injury and non-injured periods. Each point represents an individual urine sample

From a systems perspective, the urinary metabolome reflects the integrated outcome of external load exposure, internal physiological strain, and injury-recovery status, among other factors. The observed overlap between injured and non-injured states, indicating that muscle injury responses do not dominate global metabolic variance is consistent with training load theory, where injury-related perturbations are expected to be subtle compared with the broader physiological variability observed in football players.

Discriminant analysis

To identify metabolites associated with muscle injury, univariate analyses combining Welch’s test, fold-change assessment, and Hedge’s g effect size estimation were performed using winsorized metabolite data (Fig. 2). Metabolites showing significant differences were subsequently subjected to pathway over-representation analysis (ORA) using the KEGG database.

Fig. 2.

Fig. 2

Open in a new tab

Differential urinary metabolites associated with muscle injury. A volcano plot showing differential metabolite abundance between injury and control groups. The x-axis represents log2(Fold change) (FC; Injury/Control), and the y-axis represents -log10(p-value Welch’s test). Each point corresponds to an individual metabolite. Metabolites significantly increased in the injury group are shown in green, while significantly decreased metabolites are shown in red; non-significant metabolites are shown in gray. B Forest plot showing Hedge’s g effect sizes and confidence intervals for metabolites with p < 0.05. Positive values indicate higher abundance in the injury group, whereas negative values indicate lower abundance relative to controls. Larger absolute Hedge’s g values reflect greater separation between groups;C  KEGG pathway over-representation analysis (ORA) of significant metabolites. Dot size represents enrichment ratio and color represents pathway significance (p-value)

Several metabolites showed statistically significant associations with muscle injury, involving coordinated alterations in pathways related to energetic metabolism, amino acid turnover, tryptophan metabolism, and cellular stress responses. In addition to statistical significance (Fig. 2A), effect sizes were evaluated using Hedge’s g to quantify the magnitude of group differences (Fig. 2B). Most significant metabolites displayed moderate-to-large effect sizes, supporting the biological relevance of the observed alterations. The largest positive effect sizes were observed for metabolites related to tryptophan metabolism, including xanthurenic acid, kynurenic acid, and tryptamine/tryptophan ratio, as well as S-adenosylmethionine, indicating substantially higher levels in injured players. Conversely, aminophenol, proline, ornithine, and lysine exhibited negative effect sizes, reflecting lower concentrations in the injury group. The overall concordance between Welch’s test significance, fold-change analysis, and Hedge’s g increases the confidence that these metabolites represent meaningful injury-associated metabolic alterations rather than isolated statistical findings.

A metabolic change associated with muscle injury was the alteration of metabolites linked to purine degradation, particularly hypoxanthine, together with changes in methylation-related intermediate such as S-adenosylmethionine. This pattern could indicate increased ATP turnover and energetic stress at the cellular level. Purine catabolism products such as hypoxanthine are well-established markers of energetic stress, reflecting the breakdown of ATP during periods of increased metabolic demand. In the context of muscle injury, these changes could reflect the joint effect of localized energy depletion within damaged muscle fibers and systemic metabolic adjustments required to sustain tissue repair processes. As mentioned before, a second major component of the injury-associated metabolic changes involved alterations in amino acids and amino acid–related metabolites, including alanine, aminoadipic acid, glutamic acid, lysine, ornithine, proline, taurine, β-alanine, and phenylacetylglutamine. Again, rather than representing independent metabolic perturbations, collectively these changes could be linked to coordinated protein turnover and tissue remodeling. Several of these metabolites are directly implicated in skeletal muscle structural remodeling. For instance, proline and ornithine are closely linked to collagen synthesis and connective tissue repair. The elevation of BAIBA further could be linked to changes in the activity of metabolic pathways associated with muscle stress adaptation and mitochondrial remodeling. A third component involved significant alterations in tryptophan metabolism, particularly through the kynurenine pathway and related indole derivatives. Metabolites belonging to the kynurenine metabolism displayed both significant p-values and among the largest effect sizes, suggesting that perturbation of this pathway represents one of the most relevant metabolic responses associated with muscle injury. Increased levels of kynurenic acid and xanthurenic acid, together with changes in indole-3-acetic acid, indole-3-acetamide, and tryptamine, indicate a broad reprogramming of tryptophan metabolism during muscle injury. The kynurenine pathway is strongly regulated by immune signaling and inflammatory mediators and has been widely implicated in exercise-induced physiological adaptation and inflammatory regulation (Cervenka et al., 2017; dos et al., 2025). Furthermore, indolic metabolites likely reflect interactions between host metabolism and gut microbial activity.

Although these metabolic changes can be categorized into distinct biochemical components, their simultaneous change suggests a coordinated systemic adaptation to muscle injury. Thus, this integrated response likely reflects the combined demands of tissue damage repair, inflammatory regulation, and metabolic demands in elite players exposed to high external and internal loads. These results support the potential utility of urinary metabolomics as a non-invasive tool for monitoring physiological responses to injury.

The metabolites showing significant Welch’s test results and moderate-to-large Hedge’s g values were subsequently used for pathway over-representation analysis (ORA) (Fig. 2C). This analysis showed a significant enrichment of amino acid metabolic pathways, particularly arginine and proline metabolism, indicating that the individual metabolite alterations converge on processes that might be associated with tissue repair and remodeling. Nonetheless, this study relies on a targeted UPLC-MS metabolomics panel, and so pathway ORA provides an exploratory overview of biological enrichment rather than definitive mechanistic evidence. The observed enrichment in arginine and proline metabolism should be interpreted as part of a broader systemic response that includes purine degradation and tryptophan catabolism.

Multivariate discriminant analysis

PLS-DA was performed to identify coordinated metabolic patterns associated with muscle injury. Despite the significant imbalance between injury and control samples, the PLS-DA model showed statistically significant classification performance (Fig. 3). Cross-validated predictions correctly classified 73% of injury samples and 88% of non-injured samples (Fig. 3C), yielding an overall AUROC(CV) of 0.88. The positive predictive value was 41% (95% CI 32–50%) (Fig. 3D), and the negative predictive value 97% (95% CI 94–98%). The F1-score and the Matthews correlation coefficient (MCC) were 0.52 and 0.48, respectively. The F1-score indicated a moderate discrimination between injury detection and false positive mitigation under highly imbalanced conditions (~ 1:9 ratio in this study), while the MCC value confirms that the model’s performance is statistically significant with a performance above random classification. Furthermore, permutation testing supported the statistical significance of the model (AUROC-CV and classification error p-value < 0.003) (Fig. 3E). To ensure that the LOOCV approach did not introduce bias or data leakage due to the multi-sample contribution of individual athletes, a sensitivity analysis was performed by iteratively excluding entire player trajectories (leave-one-player-out, LOPO-CV). A substantial reduction in model performance was observed (AUROC-CV = 0.72, F1-score = 0.32, MCC = 0.22) which was attributed to the effect of LOPO in a highly imbalanced set (30 injury vs. 257 control samples) in which the 30 injury samples originated from a pool of 10 distinct players. As expected, performance decreased under LOPO-CV (AUROC 0.72), highlighting the contribution of inter-individual metabolic variability and the need for independent player-level validation.

Fig. 3.

Fig. 3

Open in a new tab

PLS-DA model for discrimination between muscle injury and non-injured control samples. A VIP scores and PLS-DA regression vector values; B Predicted class probabilities obtained from the PLS-DA model; C Cross-validated PLSDA predicted y values; D AUROC curves showing fitted and CV-results obtained; E Results from a permutation test to assess the statistical significance of the PLSDA CV-figures of merit classification error and AUROC. The green circle represents the value estimated using real class labels and the histogram summarizes results obtained using randomly permuted class labels

The regression vector and variable importance in projection (VIP) analysis identified BAIBA, hypoxanthine, xanthurenic acid, β-alanine, and alanine among the metabolites contributing most strongly to sample discrimination (Fig. 3A). The metabolites identified through multivariate modeling were largely consistent with the results obtained from univariate analyses and pathway enrichment approaches, supporting the robustness and biological coherence of the identified metabolic signatures.

The predicted probability distributions obtained by CV showed partial overlap between injury and control samples. Nevertheless, the statistically significant classification performance indicates that coordinated metabolic information contained within urinary metabolomic profiles can capture physiologically relevant signatures associated with muscle injury.

Feature selection

To identify a reduced subset of metabolites with maximal discrimination relevance, backward stepwise feature selection was applied to the PLS-DA model. Variables with lower VIP scores were progressively removed values while monitoring model performance during CV. Feature selection resulted in a restricted subset of metabolites that preserved classification performance (AUROC = 0.81) while substantially simplifying the model structure (Fig. 4). The selected metabolic set included metabolites associated with tryptophan metabolism, purine metabolism, amino acid turnover, and energetic stress pathways.

Fig. 4.

Fig. 4

Open in a new tab

Backward stepwise feature selection and injury-associated metabolic score. A Number of selected PLSDA latent variables and classification performance based on the classification error and AUROC estimated by CV of the PLS-DA model as a function of the number of retained metabolites during backward feature elimination. The red dot indicates the selected model; B AUROC estimated in the selected model; C Cross-validated PLSDA predicted y values after feature selection

The identification of a limited panel of discriminant metabolites is particularly relevant from a translational perspective. Compared to large-scale quantitative or untargeted metabolomics, reduced targeted panels may provide improved analytical robustness and facilitate the development and validation of quantitative assays for routine athlete monitoring. Using the selected subset of metabolites, a metabolite-based score reflecting an injury-associated metabolic signature was constructed. Although validation in independent cohorts would be required, these findings support the feasibility of developing targeted metabolomic tools for individualized athlete monitoring and recovery assessment.

Longitudinal profiles

Longitudinal metabolomic profiles were available for several players with serial urine samples collected throughout injury and recovery periods. Overall, a visual exploratory analysis of the longitudinal trajectories of the metabolic injury scores suggested the progressive normalization of the metabolomic profiles during rehabilitation up to the return-to-play (RTP) milestone. Figure 5 presents individual case trajectories displaying the cross-validated PLS-DA predicted class probabilities as the metabolite-based injury score for individual players from whom serial longitudinal urine samples were collected. Each individual panel corresponds to a distinct player trajectory across time. Metabolic profiles altered during the acute phase of injury progressively approached baseline control values as physiological recovery progressed in some individual trajectories (Fig. 5). These descriptive trajectories suggest that the metabolite-based injury score may change during rehabilitation; however, the limited number of players and non-standardized sampling intervals preclude inference regarding recovery kinetics. Although considerable inter-individual variability was observed, descriptive analysis revealed progressive normalization in some individual trajectories, including players who sustained no muscle injuries throughout the study period, pointing toward potential baseline stability in the score. These longitudinal trajectories represent descriptive trends within a limited cohort. While visual inspection indicates a progressive return toward baseline, formal longitudinal statistical modeling in larger cohorts will be necessary to establish validated recovery kinetics.

Fig. 5.

Fig. 5

Open in a new tab

Longitudinal trajectories of the metabolite-based injury score during muscle injury, rehabilitation, and recovery. Each panel displays serial predicted injury scores (cross-validated PLS-DA predicted probabilities) calculated from longitudinal urine samples of an individual player over time. The dashed line indicates the decision threshold separating non-injured baseline states from muscle injury signatures. Red circles denote samples collected during clinically confirmed muscle injury and rehabilitation periods, while green circles denote non-injured control samples. Panels illustrate individual variability in temporal responses: progressive normalization and recovery dynamics (A, B, F), distinct acute injury signatures (C, E, I, J) in injured players, or baseline stability in healthy control profiles (D, H, K, L)

As an illustrative example of structural recovery assessment, Fig. 6 shows serial MRI examinations (axial and coronal Proton Density Fat-Suppressed [PDFS] sequences) of a 24-year-old professional football player who sustained a high-speed running injury involving the biceps femoris. MRI obtained within 24 h of injury demonstrates a complete tear of the proximal common hamstring aponeurosis (arrow) located 10.5 cm distal to its ischial tuberosity origin, with distal retraction, loss of connective tissue tension, interstitial edema, and intermuscular edema (MLG-R grade IPp3r0 BAMIC grade 3 C) (Fig. 6A). Three weeks after injury, the tendon gap (arrow) is bridged by a hypertrophic reparative scar (phase 2a), with restoration of connective tissue tension and reduction of interstitial and intermuscular edema (Fig. 6B). Six weeks after injury, the scar had progressed to a phase 2b–3a appearance (arrow), characterized by a more uniform hypointense peripheral rim and decreasing central hyperintensity, reflecting ongoing scar maturation. Interstitial edema has resolved, with only mild adaptive edema and minimal residual intermuscular edema remaining (Fig. 6C). At eight weeks after injury, the scar was predominantly hypointense (arrow), consistent with advanced remodeling and connective tissue maturation, with only minimal adaptive edema persisting (Fig. 6D).

Fig. 6.

Fig. 6

Open in a new tab

Longitudinal MRI assessment of muscle healing following a biceps femoris injury. Serial axial fat-suppressed T2-weighted MRI examinations acquired from the acute phase of injury to return-to-play. Images demonstrate the progressive resolution of intramuscular edema, reduction of the connective tissue gap, restoration of muscle architecture, and maturation of the repair tissue over time. MRI follow-up was performed every 2–3 weeks as part of the FC Barcelona muscle injury management protocol. These imaging findings illustrate the dynamic biological healing process and were used in parallel with urinary metabolomic monitoring to investigate the relationship between tissue repair and systemic metabolic adaptations during recovery

Future studies could align these structural trajectories with longitudinal metabolomic profiles. While conventional clinical milestones rely heavily on the macro-structural restoration of muscle architecture seen on MRI (as shown in Fig. 6), macro-structural adaptation might not always reflect the resolution of local energetic or systemic inflammatory stress. Thus, the mapping of metabolomic trajectories relative to structural lesion evolution could support the assessment of the recovery of the athlete’s cellular homeostasis, or if persistent metabolic alterations such as elevated purine degradation or altered tryptophan catabolism remain despite normal structural imaging findings.

Conclusions and outlook

This study demonstrates that urinary metabolomic profiling can detect coordinated metabolic alterations associated with muscle injury in elite football players. While injury-related changes were subtle, the combination of univariate analysis, pathway enrichment, and multivariate classification identified biological signatures primarily involving amino acid turnover, tryptophan catabolism, and energetic stress (purine metabolism) associated with muscle injury. These findings suggest that non-invasive urinary metabolomic profiling captures systemic physiological stress associated with muscle injury. However, urinary profiles reflect systemic metabolic balance rather than local tissue remodeling, and these signatures cannot currently be used to guide return-to-play decisions. Future multicenter studies integrating longitudinal metabolomics with structural imaging are needed to determine whether this multimodal approach provides clinically meaningful value beyond standard return-to-play assessments.

Several limitations should be considered when interpreting the results of this study. First, Nutritional records were not available in this study and will be incorporated in future studies. Nutritional intake and other potential confounding factors such as acute training workload or medication which might influence the urinary metabolomic profile were not recorded. Second, although a large longitudinal database of urine samples was analyzed, the number of clinically confirmed muscle injury events remained limited and restricted to a single-center cohort of 10 distinct players. Third, urinary metabolomics reflects systemic metabolic responses and may therefore capture physiological processes not exclusively related to skeletal muscle injury. Finally, the observational nature of the study limits causal interpretation of the identified associations.

Future multicenter studies should focus on validating these urinary signatures in larger independent cohorts. Integrating longitudinal metabolomics with structural imaging datasets tracking lesion evolution, biochemical markers, and clinical assessments may support the analysis of the synchronous relationship between tissue remodeling and metabolic normalization. Future studies should determine whether this approach provides clinically meaningful information beyond current return-to-play assessments.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1 (54.8KB, docx)

Acknowledgements

The authors acknowledge the dedicated support from the FCB volunteers, FCB Medical Service, and Barça Innovation Hub (https://barcainnovationhub.com/).

Author contributions

GQ: methodology, performed research, analyzed data; RP: resources, analyzed data; MW: resources, analyzed data; SM: resources, analyzed data; PM: resources, analyzed data; JDS: analysis; GR: conceptualization, methodology, analysis, resources. All authors wrote and reviewed the manuscript.

Data availability

Due to the highly specific nature of the cohort and the inclusion of longitudinal clinical, performance, and injury-related temporal information, the dataset carries a non-negligible risk of participant re-identification even after pseudonymisation. In accordance with the conditions of the institutional ethics approval and applicable data protection regulations, the raw dataset is therefore not publicly available.

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Guillermo Quintás, Email: gquintas@leitat.org.

Gil Rodas, Email: gil.rodas@fcbarcelona.cat.

References

  1. Akenhead, R., & Nassis, G. P. (2016). Training load and player monitoring in high-level football: Current practice and perceptions. International Journal of Sports Physiology and Performance,11, 587–593. 10.1123/ijspp.2015-0331 [DOI] [PubMed] [Google Scholar]
  2. Arcos, A. L., Mendez-Villanueva, A., & Martínez-Santos, R. (2017). In-season training periodization of professional soccer players. Biology of Sport,34, 149–155. 10.5114/biolsport.2017.64588 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Barça Innovation Hub. (2018). Muscle injury guide: Prevention of and return to play from muscle injuries.
  4. Bourdon, P. C., Cardinale, M., Murray, A., Gastin, P., Kellmann, M., Varley, M. C., Gabbett, T. J., Coutts, A. J., Burgess, D. J., Gregson, W., & Cable, N. T. (2017). Monitoring athlete training loads: Consensus statement. International Journal of Sports Physiology and Performance,12, S2161–S2170. 10.1123/IJSPP.2017-0208 [DOI] [PubMed] [Google Scholar]
  5. Cao, B., Liu, S., Yang, L., & Chi, A. (2020). Changes of differential urinary metabolites after high-intensive training in teenage football players. BioMed Research International,2020, 2073803. 10.1155/2020/2073803 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Cervenka, I., Agudelo, L. Z., & Ruas, J. L. (2017). Kynurenines: Tryptophan’s metabolites in exercise, inflammation, and mental health. Science,357, Article eaaf9794. 10.1126/science.aaf9794 [DOI] [PubMed] [Google Scholar]
  7. dos S. Rangel, M. V., Lopes, K. G., Qin, X., & Borges, J. P. (2025). Exercise-induced adaptations in the kynurenine pathway: Implications for health and disease management. Frontiers in Sports and Active Living,7, Article 1535152. 10.3389/fspor.2025.1535152 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Duft, R. G., Castro, A., Chacon-Mikahil, M. P. T., Cavaglieri, C. R., Duft, R. G., Castro, A., Chacon-Mikahil, M. P. T., & Cavaglieri, C. R. (2017). Metabolomics and exercise: possibilities and perspectives. Motriz: Revista de Educação Física. 10.1590/s1980-6574201700020010 [DOI] [Google Scholar]
  9. Enea, C., Seguin, F., Petitpas-Mulliez, J., Boildieu, N., Boisseau, N., Delpech, N., Diaz, V., Eugène, M., & Dugué, B. (2010). (1)H NMR-based metabolomics approach for exploring urinary metabolome modifications after acute and chronic physical exercise. Analytical and Bioanalytical Chemistry,396, 1167–1176. 10.1007/s00216-009-3289-4 [DOI] [PubMed] [Google Scholar]
  10. Finaud, J., Lac, G., & Filaire, E. (2006). Oxidative stress: Relationship with exercise and training. Sports Medicine,36, 327–358. [DOI] [PubMed] [Google Scholar]
  11. Fuller, C. W., Ekstrand, J., Junge, A., Andersen, T. E., Bahr, R., Dvorak, J., Hägglund, M., McCrory, P., & Meeuwisse, W. H. (2006). Consensus statement on injury definitions and data collection procedures in studies of football (soccer) injuries. British Journal of Sports Medicine,40, 193–201. 10.1136/bjsm.2005.025270 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. García-Campos, M. A., Espinal-Enríquez, J., & Hernández-Lemus, E. (2015). Pathway analysis: State of the art. Frontiers in Physiology,6, Article 383. 10.3389/fphys.2015.00383 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. González, J. R., Cáceres, A., Ferrer, E., Balagué-Dobón, L., Escribà-Montagut, X., Sarrat-González, D., Quintás, G., & Rodas, G. (2024). Predicting injuries in elite female football players with global-positioning-system and multiomics data. International Journal of Sports Physiology and Performance,19, 661–669. 10.1123/ijspp.2023-0184 [DOI] [PubMed] [Google Scholar]
  14. Gorostiaga, E. M., Navarro-Amézqueta, I., Calbet, J. A. L., Hellsten, Y., Cusso, R., Guerrero, M., Granados, C., González-Izal, M., Ibañez, J., & Izquierdo, M. (2012). Energy metabolism during repeated sets of leg press exercise leading to failure or not. PLoS One,7, Article e40621. 10.1371/journal.pone.0040621 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Hawley, J. A., Hargreaves, M., Joyner, M. J., & Zierath, J. R. (2014). Integrative biology of exercise. Cell,159, 738–749. 10.1016/j.cell.2014.10.029 [DOI] [PubMed] [Google Scholar]
  16. Heaney, L. M., Deighton, K., & Suzuki, T. (2017). Non-targeted metabolomics in sport and exercise science. Journal of Sports Sciences,0, 1–9. 10.1080/02640414.2017.1305122 [DOI] [PubMed] [Google Scholar]
  17. Impellizzeri, F. M., Marcora, S. M., & Coutts, A. J. (2019). Internal and external training load: 15 years on. International Journal of Sports Physiology and Performance,14, 270–273. 10.1123/ijspp.2018-0935 [DOI] [PubMed] [Google Scholar]
  18. Isern-Kebschull, J., Mechó, S., Pedret, C., Pruna, R., Alomar, X., Kassarjian, A., Luna, A., Martínez, J., Tomas, X., & Rodas, G. (2024). Muscle healing in sports injuries: MRI findings and proposed classification based on a single institutional experience and clinical observation. RadioGraphics,44, Article e230147. 10.1148/rg.230147 [DOI] [PubMed] [Google Scholar]
  19. 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 (Auckland, N. Z.), 52, 547–583. 10.1007/s40279-021-01582-y [DOI] [PubMed] [Google Scholar]
  20. Lario, S., Ramírez-Lázaro, M. J., Sanjuan-Herráez, D., Brunet-Vega, A., Pericay, C., Gombau, L., Junquera, F., Quintás, G., & Calvet, X. (2017). Plasma sample based analysis of gastric cancer progression using targeted metabolomics. Scientific Reports,7, Article 17774. 10.1038/s41598-017-17921-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Li, S., Park, Y., Duraisingham, S., Strobel, F. H., Khan, N., Soltow, Q. A., Jones, D. P., & Pulendran, B. (2013). Predicting network activity from high throughput metabolomics. PLoS Computational Biology,9, Article e1003123. 10.1371/journal.pcbi.1003123 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Manaf, F. A., Lawler, N. G., Peiffer, J. J., Maker, G. L., Boyce, M. C., Fairchild, T. J., & Broadhurst, D. (2018). Characterizing the plasma metabolome during and following a maximal exercise cycling test. Journal of applied physiology (Bethesda, Md. : 1985),125, 1193–1203. 10.1152/japplphysiol.00499.2018 [DOI] [PubMed] [Google Scholar]
  23. Marinho, A. H., Sousa, F., Vilela, R., Balikian, P., Bento, Ed. S., Aquino, Td. M., Crispim, A., Ataide-Silva, T., & Araujo, G. (2022). The rating of perceived exertion is able to differentiate the post-matches metabolomic profile of elite U-20 soccer players. European Journal of Applied Physiology,122, 371–382. 10.1007/s00421-021-04838-7 [DOI] [PubMed] [Google Scholar]
  24. Neal, C. M., Hunter, A. M., Brennan, L., O’Sullivan, A., Hamilton, D. L., De Vito, G., & Galloway, S. D. R. (2013). Six weeks of a polarized training-intensity distribution leads to greater physiological and performance adaptations than a threshold model in trained cyclists. Journal of applied physiology Bethesda Md 1985,114, 461–471. 10.1152/japplphysiol.00652.2012 [DOI] [PubMed] [Google Scholar]
  25. Nguyen, T.-M., Shafi, A., Nguyen, T., & Draghici, S. (2019). Identifying significantly impacted pathways: A comprehensive review and assessment. Genome Biology,20, Article 203. 10.1186/s13059-019-1790-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Olivieri, A. C. (2008). Analytical Advantages of Multivariate Data Processing. One, Two, Three, Infinity? Analytical Chemistry, 80, 5713–5720. 10.1021/ac800692c [DOI] [PubMed] [Google Scholar]
  27. Pang, Z., Lu, Y., Zhou, G., Hui, F., Xu, L., Viau, C., Spigelman, A. F., MacDonald, P. E., Wishart, D. S., Li, S., & Xia, J. (2024). MetaboAnalyst 6.0: Towards a unified platform for metabolomics data processing, analysis and interpretation. Nucleic Acids Research. 10.1093/nar/gkae253 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Pang, Z., Zhou, G., Ewald, J., Chang, L., Hacariz, O., Basu, N., & Xia, J. (2022). Using MetaboAnalyst 5.0 for LC–HRMS spectra processing, multi-omics integration and covariate adjustment of global metabolomics data. Nature Protocols,17, 1735–1761. 10.1038/s41596-022-00710-w [DOI] [PubMed] [Google Scholar]
  29. Pechlivanis, A., Papaioannou, K. G., Tsalis, G., Saraslanidis, P., Mougios, V., & Theodoridis, G. A. (2015). Monitoring the Response of the Human Urinary Metabolome to Brief Maximal Exercise by a Combination of RP-UPLC-MS and (1)H NMR Spectroscopy. Journal Of Proteome Research, 14, 4610–4622. 10.1021/acs.jproteome.5b00470 [DOI] [PubMed] [Google Scholar]
  30. Quintas, G., Reche, X., Sanjuan-Herráez, J. D., Martínez, H., Herrero, M., Valle, X., Masa, M., & Rodas, G. (2020). Urine metabolomic analysis for monitoring internal load in professional football players. Metabolomics,16, Article 45. 10.1007/s11306-020-01668-0 [DOI] [PubMed] [Google Scholar]
  31. Rodas, G., Ferrer, E., Reche, X., Sanjuan-Herráez, J. D., McCall, A., & Quintás, G. (2022). A targeted metabolic analysis of football players and its association to player load: Comparison between women and men profiles. Frontiers in Physiology. 10.3389/fphys.2022.923608 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Rodas, G., Ferrer, E., Sanjuan, J. D., & Quintás, G. (2025). UPLC-MS and multivariate analysis reveal metabolic pathway adaptations to training in professional football players. Talanta,291, Article 127893. 10.1016/j.talanta.2025.127893 [DOI] [PubMed] [Google Scholar]
  33. Rubingh, C. M., Bijlsma, S., Derks, E. P. P. A., Bobeldijk, I., Verheij, E. R., Kochhar, S., & Smilde, A. K. (2006). Assessing the performance of statistical validation tools for megavariate metabolomics data. Metabolomics,2, 53–61. 10.1007/s11306-006-0022-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Saccenti, E., Hoefsloot, H. C. J., Smilde, A. K., Westerhuis, J. A., & Hendriks, M. M. W. B. (2014). Reflections on univariate and multivariate analysis of metabolomics data. Metabolomics, 10, 361–374. 10.1007/s11306-013-0598-6 [DOI] [Google Scholar]
  35. Schoeny, H., Stelzer, B., Hofbauer, T., Reisenbauer, F., Abiead, Y. E., Scharhag, J., & Koellensperger, G. (2025). Metabolome trajectories in male and female athletes. Journal of Sport and Health Science,14, Article 101065. 10.1016/j.jshs.2025.101065 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1 (54.8KB, docx)

Data Availability Statement

Due to the highly specific nature of the cohort and the inclusion of longitudinal clinical, performance, and injury-related temporal information, the dataset carries a non-negligible risk of participant re-identification even after pseudonymisation. In accordance with the conditions of the institutional ethics approval and applicable data protection regulations, the raw dataset is therefore not publicly available.

Articles from Metabolomics are provided here courtesy of Springer

RESOURCES