The effect of contact/collision sport participation without concussion on neurometabolites: A systematic review and meta‐analysis of magnetic resonance spectroscopy studies

The aim of this study was to systematically review prior research investigating the effects of contact/collision sport participation on neurometabolite levels in the absence of concussion. Four online databases were searched to identify studies that measured neurometabolite levels in contact/collision sport athletes (without concussion) using proton (1H) or phosphorus (31P) magnetic resonance spectroscopy (MRS). All study designs were acceptable for inclusion. Meta‐analytic procedures were used to quantify the effect of contact/collision sport participation on neurometabolite levels and explore the impact of specific moderating factors (where sufficient data were available). Narrative synthesis was used to describe outcomes that could not be meta‐analysed. Nine observational studies involving 300 contact/collision sport athletes were identified. Six studies (providing 112 effect estimates) employed longitudinal (cohort) designs and three (that could not be meta‐analysed) employed case–control designs. N‐acetylaspartate (NAA; g = −0.331, p = 0.013) and total creatine (tCr; creatine + phosphocreatine; g = −0.524, p = 0.029), but not glutamate–glutamine (Glx), myo‐inositol (mI) or total choline (tCho; choline‐containing compounds; p's > 0.05), decreased between the pre‐season and mid−/post‐season period. Several moderators were statistically significant, including: sex (Glx: 6 female/23 male, g = −0.549, p = 0.013), sport played (Glx: 22 American football/4 association football [soccer], g = 0.724, p = 0.031), brain region (mI: 2 corpus callosum/9 motor cortex, g = −0.804, p = 0.015), and the MRS quantification approach (mI: 18 absolute/3 tCr‐referenced, g = 0.619, p = 0.003; and tCho: 18 absolute/3 tCr‐referenced, g = 0.554, p = 0.005). In case–control studies, contact/collision sport athletes had higher levels of mI, but not NAA or tCr compared to non‐contact sport athletes and non‐athlete controls. Overall, this review suggests that contact/collision sport participation has the potential to alter neurometabolites measured via 1H MRS in the absence of concussion. However, further research employing more rigorous and consistent methodologies (e.g. interventional studies with consistent 1H MRS pulse sequences and quantifications) is required to confirm and better understand the clinical relevance of observed effects.


| INTRODUC TI ON
Contact and collision sports such as association football (soccer), American football and rugby are popular forms of physical activity (AUSPLAY, 2021; U.S. Bureau of Labor Statistics, 2017;Viviers et al., 2018).Yet, participation in these sports can result in hundreds to thousands of head impacts each sporting season (Bailes et al., 2013).Head impacts that do not elicit clinical symptoms of concussion (e.g.headache, nausea, vomiting and blurred vision) are termed 'subconcussive' (Bailes et al., 2013).Despite a lack of overt symptoms, evidence of repetitive subconcussive impacts having negative effects on health is emerging (Nowinski et al., 2022).
Indeed, in a recent post-mortem analysis, ~20% of athletes and military personnel (n = 114) diagnosed with chronic traumatic encephalopathy, an incurable neurodegenerative disease related to head trauma (Nowinski et al., 2022), had no history of concussion but had been exposed to many years of subconcussive impacts (Cantu & Budson, 2019;Stein et al., 2015).Further research investigating the neural effects of repeated subconcussive head impacts in sport is, therefore, required.
Magnetic resonance imaging (MRI) provides a non-invasive approach to assess objective markers of brain structure, function and chemistry (Hennel et al., 2020).It encompasses a wide range of techniques including magnetic resonance spectroscopy (MRS), functional MRI (fMRI) and diffusion weighted imaging (DWI; Hennel et al., 2020), and therefore, has the potential to probe many different aspects of brain physiology (Bailes et al., 2013;Choe, 2016).Phosphorus ( 31 P) or proton ( 1 H) MRS can be used to measure the concentration of metabolites, such as adenosine triphosphate (ATP), N-acetylaspartate (NAA), total choline (tCho; choline-containing compounds), total creatine (tCr: creatine + phosphocreatine), myo-inositol (mI) and glutamate-glutamine (Glx) in humans (Bartnik-Olson et al., 2021).Subtle alterations to these neurometabolites are associated with various pathological states (including asymptomatic pathologies; Bartnik-Olson et al., 2021;Rae, 2014).For instance, reduced mI concentrations can indicate decreased glial cell density (Schuhmann et al., 2003), while reduced tCr concentrations can indicate neurometabolic dysfunction (Rae, 2014).Subconcussive impacts resulting from contact/ collision sport participation have been reported to generate internal compression, tension and shear/strain forces that can alter axonal integrity (Nauman et al., 2020;Koerte et al., 2023;Tayebi et al., 2021).These alterations have the potential to produce a cascade of metabolic disturbances (Giza & Hovda, 2014).Thus, MRS may assist us to better understand the neural effects of subconcussive impacts in sport.
A recent systematic and meta-analytic review investigated the effects of head trauma on neurometabolites, as assessed by 1 H MRS (Joyce et al., 2022).It included 138 studies involving >1400 cases and >1100 controls (i.e.no head trauma)-and classified head trauma cases into four sub-groups: (1) subconcussive impacts; (2) mild traumatic brain injury (TBI; inclusive of concussion); (3) moderate/severe TBI; and (4) mixed severities of TBI.The review identified a myriad of neurometabolic alterations in TBI-affected participants, including decreased NAA, increased tCho and mI (in the moderate/severe TBI group, only), and varied alterations in Glx and tCr (Joyce et al., 2022).No effect of subconcussive impacts on neurometabolites was found (Joyce et al., 2022).There were, however, a considerable amount of relevant data omitted.Indeed, while studies can investigate the effects of subconcussive impacts on neurometabolites using longitudinal (cohort) designs (i.e. by performing MRS before, during and/or after a sporting season), the review only compared cases (e.g.contact/collision sport athletes) and controls.Further, if a study had a control population and a longitudinal element, the cases and controls were only compared at one timepoint; specifically, that with the largest sample size.It should also be noted that: (1) some of these cases had a confirmed or possible history of concussion/mild TBI (i.e. the effect of subconcussion could not be isolated; Nauman et al., 2020;Schuhmann et al., 2003;Tayebi et al., 2021); and (2) the quality of the 1 H MRS data were not formally assessed.The latter point is of importance as 1 H MRS acquisition parameters can be highly variable between research groups and impact measurement quality (Near et al., 2021;Peek et al., 2020).
Accordingly, the aim of the current study was to undertake a systematic review and meta-analysis of prior research investigating the effect of contact/collision sport participation on neurometabolites (derived from 31 P or 1 H MRS) in the absence of concussion.Specifically, we investigated the: (1) change in neurometabolites levels across the sporting season (i.e.pre-season vs. mid−/post-season; in longitudinal (cohort) study designs); and (2) difference in neurometabolites between contact/collision sport and non-athlete controls.Overall, this review suggests that contact/collision sport participation has the potential to alter neurometabolites measured via 1 H MRS in the absence of concussion.However, further research employing more rigorous and consistent methodologies (e.g.interventional studies with consistent 1 H MRS pulse sequences and quantifications) is required to confirm and better understand the clinical relevance of observed effects.

K E Y W O R D S
brain chemistry, contact sport, neuroimaging, subconcussive impacts athletes and controls (i.e. in case-control study designs).The impact of assessment timing (e.g. during season, post-season), sex, type of sport, brain region assessed and the MRS quantification approach (e.g.absolute, tCr-referenced) was also explored, where possible.

| ME THODS
The methodology of this review was developed in accordance with Cochrane Handbook for Systematic Reviews of Interventions (Version 6.3;Higgins et al., 2022) and registered prospectively on Prospero (ID: CRD42022337774).

| Literature search
Potential research studies were identified by searching the online databases PubMed (MEDLINE), Web of Science (via Clarivate Analytics), SPORTDiscus (EBSCO) and Scopus from inception until 9 May 2023 using the Boolean expression described in Table 1.Where applicable, thesaurus terms (i.e.'MeSH terms' for PubMed, 'Category' for Web of Science, 'Subjects' for SportDiscus) were included in combination with the defined search terms (Table 1).All search terms were applied to the 'Title', 'Abstract' and 'Keyword' fields.Where an eligible study contained multiple 'arms' (e.g. two or more groups of contact/collision sport athletes), the separate arms were treated as discrete studies, termed 'trials', identifiable by the additional letters (e.g.a-d) in the citation.Further, as single trials often took multiple measurements (e.g. over the sporting season, in multiple brain regions), each one could contribute multiple effect estimates to the review (note: multilevel models were used to account for dependency of effect estimates in statistical analyses -refer to 2.6 Quantitative Analysis).In the event that results were not adequately reported, and the study was published within the previous 10 years (2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021)(2022)(2023), an attempt was made to contact the corresponding author to retrieve the relevant missing data.If the author failed to respond, and graphical results were available in the paper, PlotDigitizer (https:// plotd igiti zer.com/ ) was used to extract numerical values.

TA B L E 1
Search terms used for literature search.exposure (3-items).'Stars' are awarded to each item deemed 'low risk' (Wells et al., n.d.).The NOS has previously been adapted to assess RoB in studies of subconcussive impacts and neuroimaging (Tarnutzer et al., 2017;Walter et al., 2022).The criteria were further adapted for this review to better suit the broad range of sports accepted and provide further clarity of the definitions (File S2).Study quality was rated according to established thresholds (Tarnutzer et al., 2017;Walter et al., 2022): 'good' (≥7 stars), 'fair' (3-6 stars), 'poor' (≤2 stars).

| MRS quality assessment
The quality of the MRS acquisition in included studies was evaluated by two independent assessors (N.D. & A.P.) using the MRS-Quality (MRS-Q) tool (Peek et al., 2020).The MRS-Q is categorised into three parts with 11 criteria: (1) parameters (four criteria); (2) utilisation of quality checks (three criteria); and (3) study design/post-processing (four criteria; Peek et al., 2020).Each study was characterised according to whether it met each criterion: Y (met criterion), N (did not meet criterion) or < i > (provided insufficient information).The 'parameters' category was used to determine the quality of acquisition (High or Low) as suggested by Peek et al. (2020).

| Data synthesis
Meta-analyses were performed to quantify the change in neurometabolite levels in longitudinal studies.Where insufficient data were available for meta-analysis (i.e.only one effect estimate could be derived), a narrative synthesis of the results was provided.Studies reporting glutamate (Glu) signal were included in the analysis of Glx (unless Glx was also reported), as the Glx signal is a combination of Glu and glutamine (Gln), and Glu dominates the Glx signal (Joyce et al., 2022).If neurometabolites were reported with multiple quantification approaches (e.g. absolute concentration and in ratio to another neurometabolite), only the absolute concentration(s) were included.The pre-season period (i.e.comparator or 'baseline') was defined as the period prior to the commencement of contact/collision sport training.Any other assessment period (e.g.mid-season or post-season) was defined as the time of assessment (ToA), which was the basis of comparisons.
Meta-analyses quantifying the difference in neurometabolite levels between contact/collision sport athletes and controls could not be performed as only three relevant case-control studies were identified and these had heterogeneous outcome measures.
Therefore, a narrative synthesis of these results was provided.

| Quantitative analysis
The quantitative analysis was performed in R Version 4.2 using the metafor-package (Viechtbauer, 2010) and syntax adapted from Assink and Wibbelink (Assink & Wibbelink, 2016).A series of four-level meta-analyses and four-level moderator analyses with multiple random effects was conducted.A two-level analysis is equivalent to a traditional meta-analysis involving only one random effect.We added random effects at two additional levels to account for dependency among effect estimates derived from: (1) the same studies; and (2) the same trials (defined above).The four sources of variance modelled were therefore: (Level 1) the sampling variance for the observed effect estimates; (Level 2) the variance between effect estimates derived from the same trials; (Level 3) the variance between effect estimates derived from the same studies, and (Level 4) the variance between studies.

| Effect estimates
All studies included in the meta-analyses used within-subject (i.e.repeated measures) experimental designs.Therefore, Hedges' g effect estimates were calculated by standardising the mean difference between the baseline and ToA measurement against the SD of this difference (SD Δ ; corrected for correlation) and correcting for bias because of small sample size (Borenstein et al., 2021).Where non-parametric statistics were reported, means and SDs were imputed using methods recommended by Luo et al. (2018) and Wan et al. (2014), respectively.The magnitude of effect was defined in accordance with Cohen (Cohen, 1988), where Hedges' g values of 0.2, 0.5 and 0.8 are considered 'small', 'medium' and 'large', respectively.
Negative effect estimates were used to signify a reduction in the level of a given neurometabolite compared to baseline.
The SD Δ was determined using either raw data, the SD Δ from published data, or a p-value (or t-statistic) derived from a paired t-test.Where a t-statistic derived from a paired t-test was reported, the following formula was used to calculate the SD Δ (Higgins et al., 2022): where n is the sample size.If a t-statistic was not provided, but a pvalue (i.e.p = x or p < x) derived from a paired t-test was reported, it was used to derive a t-statistic.
If none of this information was available, the SD Δ was estimated using the following formula (Higgins et al., 2022: where R is the correlation coefficient calculated for each neurometab- (1) (2) Effect estimates were weighted as described elsewhere (Viechtbauer, 2010), with results considered statistically significant if the 95% confidence interval did not include zero.
Heterogeneity was assessed using Cochran's Q, the I 2 -index and the within-and between-cluster variance components (i.e.σ 2 ).
Significant heterogeneity was indicated by a p-value <0.05 for Cochran's Q (Borenstein et al., 2021).heterogeneity where (a) significant heterogeneity was identified (Viechtbauer, 2007); and (b) sufficient data were available (i.e. more than one effect estimate).The influence of the following period.In both cases, only the first measurements were included in the moderator analyses.The latter study (Bari et al., 2019) also reported taking their measures during the 'first and second halves' of the season rather than in the number of weeks played (Bari et al., 2019)
Notably, none of the studies using mixed designs contained eligible case-control data (i.e.contained eligible longitudinal data, only) as one or more of their participants had a confirmed or possible history of concussion (but had not experienced a concussion during the period of observation).Overall, n = 300 contact/collision sport athletes (65% male), n = 50 non-contact sport athletic controls (64% male) and n = 24 non-athletic controls (50% male) were included in this review.Studies included 11 trials and provided a combined total of 130 effect estimates.Two studies (Bari et al., 2019;Churchill et al., 2017) had some missing data (unable to be retrieved).The data that were available in these publications were included in this review.

| Study characteristics
The characteristics of the included studies are detailed in Table 2.

| MRS quality assessment
Results from the MRS quality assessment are provided in File S4.
According to the MRS-Q, 8/9 studies were assessed as having highquality MRS acquisition based on questions relating to 'adequate parameters' and 'adequate quality' of the MRS-Q.Utilisation of data quality checking procedures were missing from several studies.
Significant heterogeneity was present in this analysis (Q = 68.5, p < 0.001), thus moderators were explored.Sex and Sport Played were significant moderating factors (File S8a).
A significantly lower effect was found in males than females (6F/23M; g = −0.549,p = 0.013).Soccer players had a significantly higher mean effect than American Football players (22 American Football/4 Soccer; g = 0.724, p = 0.031).No other significant differences between Sport Played and other moderators were observed (all p's > 0.05).

F I G U R E 2
The effect of contact/collision sport participation on Glx concentration in longitudinal studies (baseline vs. the ToA).CI, confidence interval; ES, effect size; M, mean; RE, random effects; SD, standard deviation; ToA, Time of Assessment.

| myo-Inositol (mI)
Four longitudinal studies (n = 151 participants; 60% male) providing 21 effect estimates were included in the analysis investigating the effect = 0.019, Figure 5).The magnitude and non-significance of this effect was stable during sensitivity analyses using different R values (R = 0.5: g = −0.275,p = 0.070; R = 0.9: g = −0.123,p = 0.069; File S5) -but not during one-out analyses (File S6d).Indeed, when the effect estimate from Chamard et al. (2012); E1) was removed, the overall weighted mean effect indicated a significant reduction in mI concentration.The funnel plot and Egger's regression test did not identify significant publication bias (File S7; p = 0.146).
Significant heterogeneity was present in this analysis (Q = 47.4,p < 0.001), thus moderators were explored.MRS Quantification Approach and Brain Region were significant moderating factors (File S8d).A significantly higher mean effect was observed for values in-ratio to tCr than water-reference absolute values (18 Absolute/3 tCr; g = 0.619, p = 0.044).A significantly lower mean effect was identified for the dominant motor cortex (M1) than the corpus callosum (CC; 2 CC/9 M1; g = −0.804,p = 0.046).

| Total creatine (tCr)
Two longitudinal studies (n = 76 participants; 100% male) providing 18 effect estimates were included in the analysis investigating the effect of contact/collision sport participation on tCr (Poole et al., 2014;Vike et al., 2022).The overall weighted mean effect indicated a significant reduction in tCr concentration across the The effect of contact/collision sport participation on mI concentration in longitudinal studies (baseline vs. the ToA).CI, confidence interval; ES, effect size; M, mean; RE, random effects; SD, standard deviation; ToA, Time of Assessment.

| Glutamine (Gln)
One study (n = 48 participants; 100% female) providing a single effect estimate was reviewed (Schranz et al., 2018).This study measured absolute Gln concentration using 3 T MRI with a long TE (135 ms) to reduce the error associated with quantification of macromolecules.The scans were taken 0-4 weeks 'in-season' and during the 'off-season'.Overall, Gln concentration was significantly higher 'in-season' compared to 'off-season' (taken as baseline; in-season: 0.66 ± 0.37, off-season, 0.52 ± 0.33, p = 0.01), with a small-medium effect size (g = 0.447).

| Neurometabolites in Case-Control studies
Three studies compared the neurometabolites of contact/collision sport athletes (with no history of concussion) to controls (Churchill et al., 2017;Koerte et al., 2015;Lefebvre et al., 2018).Demographic information for these studies can be found in Table 2. Briefly, one study (Lefebvre et al., 2018) compared current rugby and soccer athletes (n = 24) to athlete controls (n = 24; swimmers) and non-athlete controls (n = 24).Another (Churchill et al., 2017) compared current collision (n = 12; rugby, ice hockey, lacrosse males and American football) and contact (n = 11; soccer, field hockey, basketball, lacrosse females and water polo) sport players to non-contact sport athletes (n = 12; volleyball).The third study (Koerte et al., 2015) compared retired professional soccer athletes (n = 11) to retired professional non-contact sport athletes (n = 14; table tennis, running or ballroom dancing).
Less data, or less consistent effects, were noted for other metabolites.tCho was significantly higher in contact sport athletes (p = 0.04), but only one study examined this neurometabolite (Koerte et al., 2015).GABA was significantly lower (p = 0.017) in contact sport athletes, and Glx was also lower (although only in one (Lefebvre et al., 2018) of the two (Koerte et al., 2015) studies for Glx).GSH was not significantly altered in the one study that measured it (Koerte et al., 2015).

| DISCUSS ION
This systematic review and meta-analysis investigated the effect of contact/collision sport participation on neurometabolite levels in the absence of overt concussion, but in the likely presence of subconcussive impacts.Nine observational studies were reviewed.Overall, we identified: (1) significant changes in 1 H MRS neurometabolite levels across the sporting season; specifically, a small reduction in NAA levels and a medium reduction in tCr levels from the pre-season to the mid−/post-season period (i.e.baseline vs. ToA); (2) significant (but inconsistent) moderating effects of sex, sport played, brain region, and MRS quantification approach; and (3) very few studies comparing the 1 H neurometabolite levels of contact/collision sport athletes, without a history of overt concussion, to controls.

| 1 H MRS Neurometabolites in longitudinal studies
The observed reductions in NAA and tCr levels from the pre-season to the mid-/post-season period (i.e.baseline vs. ToA) contrast the conclusions of a previous meta-analysis (Joyce et al., 2022) which found no effect of subconcussive impacts on 1 H MRS neurometabolites.Importantly, the previous meta-analysis compared contact/collision sport athletes to controls, rather than at various timepoints across the sporting season.It also incorporated fewer studies.
Nonetheless, our findings should be interpreted with some caution as: (1) neither the NAA nor Cr effect was truly stable (i.e.'held up' during the one-out analyses); (2) the tCr analysis was based on only two studies (Poole et al., 2014;Vike et al., 2022); and (3) both neurometabolites (i.e.NAA or tCr) were missing (non-significant) effect estimates (from (Schranz et al., 2018)).
Interestingly, the observed changes in NAA and tCr levels appear to align more closely with research investigating the effects of TBI on 1 H MRS neurometabolites (Joyce et al., 2022).Indeed, the abovementioned meta-analysis (which investigated the effects of both subconcussive impacts and TBI) found that TBI resulted in a moderate reduction in NAA and small reduction in tCr (>3 months post-TBI only; Joyce et al., 2022).Alterations to tCr are often considered a marker of changed energy homeostasis (Rae, 2014), while NAA is traditionally considered an indicator of neuronal integrity (Nadler & Cooper, 1972;Parry et al., 2004).However, additional research is required to determine the clinical significance, including the short-(e.g.risk of more severe concussion symptoms) and long-term (e.g.risk of neurodegenerative diseases) consequences, of such effects.
There was no discernible change in Glx levels across the sporting season.However, Sex and the Sport Played appeared to be moderating factors, with the change in Glx being significantly lower for males (g = −0.549)than females, and higher for soccer players (g = 0.724) than American Football players.This result is difficult to interpret as all the females in these studies were soccer players (i.e. the effects of Sex and Sport Played cannot be disentangled).Nevertheless, head impact kinematics (Bari et al., 2019;Nevins et al., 2019;Poole et al., 2015;Stojsih et al., 2010) could potentially explain these differences.Head impact kinematics such as linear and rotational force can be measured through wearable accelerometers (Kieffer et al., 2020).This technology has been utilised in multiple contact/ collision sports, with males typically incurring a larger number of subconcussive impacts, and higher strength (i.e.linear/rotational force) subconcussive impacts than females (Nevins et al., 2019;Stojsih et al., 2010).For example, male American Football players have been shown to accumulate more ≥50 g linear force impacts than female soccer players (Bari et al., 2019).This is potentially important given that a negative correlation between the number of high force subconcussive impacts (≥60 g linear force) and Glx concentration has been demonstrated in American Football athletes (Poole et al., 2015).Future studies examining Glx concentration in contact/collision sport athletes should consider analysing males and females separately, and including head impact kinematics where possible.
There was no discernible change in tCho or mI levels across the sporting season.However, the one-out analyses revealed a significant reduction in mI levels across the sporting season when the effect estimate from Chamard et al. (Chamard et al., 2012;E1) was removed.In addition, MRS Quantification Approach (tCho; g = 0.554, mI; g = 0.619) and Brain Region (mI only; g = −0.804)were found to be moderating factors.Firstly, the moderating effect of MRS Quantification Approach may be explained by the finding that contact/collision sport participation affected (i.e.decreased) the reference metabolite (i.e.tCr).While the MRS Quantification Approach was not significant for other neurometabolites, this could be because of a low proportion of tCr-referenced values (i.e.NAA: 2/21 studies; Glx: 3/29 studies) contributing to these effects, rather than suggesting there is no effect.These results, combined with the finding that tCr was reduced in the main analysis, provide further evidence for researchers to avoid the use of internal metabolite references (e.g.tCr) when analysing data related to subconcussive impacts.Instead, researchers should utilise data referenced to the internal water signal as an absolute concentration (molarity or molality) or in ratio (institutional units), as described in a recent expert consensus paper (Near et al., 2021).
Secondly, regional differences in mI may be explained by the superficial location of the M1 relative to the CC.Indeed, the M1 is located in the precentral gyrus on the medial surface of the brain, whereas the CC is a deep grey and white matter interface structure (Crossman & Neary, 2018).More severe head impacts (i.e.TBI) can result in diffuse axonal injury and swelling in the CC (Schuhmann et al., 2003), leading to a myriad of chemical abnormalities including increased mI (Joyce et al., 2022).However, subconcussive impacts are inherently milder in nature (Bailes et al., 2013), and may result in less diffuse abnormalities.More longitudinal studies are needed to confirm the robustness of this effect and better understand regional variations.

| 1 H MRS Neurometabolites in Case-Control studies
Research comparing neurometabolites in contact/collision sport athletes, without a history of concussion, to matched controls is scarce.Our review identified only three studies measuring 1 H MRS neurometabolites; two involving young males and females during the off-season/pre-season in a variety of contact/collision sports (Churchill et al., 2017;Lefebvre et al., 2018), and the other, involving retired male soccer athletes (Koerte et al., 2015).Despite the limited available data and heterogenous populations, the two studies that measure mI found significantly higher levels in contact/collision sport athletes than controls (i.e. both non-contact sport athlete and non-athlete controls).In contrast, an earlier meta-analysis found no difference in the neurometabolites (including mI levels) of contact/ collision sport athletes and controls (Joyce et al., 2022).The reason for the disparity is unclear.Age is one factor that can influence mI levels (Lind et al., 2020), however the age of participants in the current review (Churchill et al., 2017;Koerte et al., 2015;Lefebvre et al., 2018) and the former review (Gardner et al., 2017;Koerte et al., 2015;Poole et al., 2014) did not differ distinctly.The only major difference between the reviews was that the earlier meta-analysis included (rather than excluded) participants with a confirmed or possible history of concussion/mild TBI (Gardner et al., 2017;Poole et al., 2014).However, this could be expected to exacerbate (not attenuate) the increase in mI (particularly considering that the authors found that mI was increased >3 months post-TBI (Joyce et al., 2022)).
Nonetheless, the observed increase in mI levels could indicate glial proliferation and/or a neuroinflammatory response (Schuhmann et al., 2003;Simon et al., 2017), especially considering one of the studies found a synchronous increase in tCho (Koerte et al., 2015; both metabolites have been suggested as markers of glial reactivity after TBI (Schuhmann et al., 2003)).Finally, NAA and tCr were not altered in any case-control study in this review.This is in agreement with the earlier meta-analysis (Joyce et al., 2022)

| Quality assessment
We used standardised assessment tools to assess the quality of the MRS data acquisition (i.e.MRS-Q) and RoB (i.e.NOS) in included studies.The majority (89%) of studies satisfied the MRS-Q criteria and were therefore considered to be of high MRS acquisition quality.This was predominantly related to strong acquisition parameters and/or justification of these parameters.However, no studies achieved a rating of 'good' on the RoB assessment.Instead, most were rated 'fair', and two were rated 'poor'.The main reason for this was insufficient reporting of the study methods.For example, few studies reported comprehensive participant demographics (such as playing positions) or specified the length of time participants had refrained from playing contact/collision sports prior to their baseline assessment.Thus, certain playing positions may be overrepresented in our analyses.In addition, participants' historical exposure to head impacts (i.e. over many years prior to baseline assessment) could diminish the magnitude of effects seen at the time of assessment.

| Limitations
There are important methodological limitations to consider in our review.Firstly, it is worth noting that studies that provided more (i.e.multiple) effect estimates could have had a greater influence on the overall result (Van den Noortgate et al., 2013).Secondly, some of the component meta-analyses were conducted using only a small number of studies (e.g.only two for tCr).Third, some studies had missing data, which may compromise the ability to detect significant differences or introduce selection bias.Finally, only observational studies were available for inclusion in this review.
Observational research can provide important preliminary data regarding the association between two outcomes (i.e.subconcussive impacts and neurometabolites alterations; Boyko, 2013).
However, such designs cannot identify causation, and can be confounded by a wide variety of internal and external variables (Grimes & Schulz, 2002).

| Future directions
This review did not identify any interventional studies investigating the effects of subconcussive impacts on neurometabolite levels.Such protocols do exist, where an acute subconcussive soccer heading task Two investigators (N.D. & B.D.) independently screened all titles and abstracts against the following inclusion criteria: (1) English language; (2) full-length article; (3) original research; (4) performed MRS; and (5) examined current or former contact/collision sport athletes.The remaining articles were then screened for eligibility by full text.The following exclusion criteria were applied: (1) no comparator condition; and (2) one (or more) of the contact/collision sport athletes studied; (a) experienced a concussion during the period of observation (longitudinal study designs, only) or (b) had a history of concussion (case-control study designs, only; and were not removed from the final sample).The final decision to include or discard potential research studies was made by two investigators (N.D. & B.D.).Any uncertainties were resolved in consultation with a third investigator (D.M.).The reference lists of all included studies and those included in a recent meta-analysis on a similar topic (Joyce et al., 2022) were hand-searched for missing publications.No additional restrictions (e.g.language or publication period) were imposed on the search.Details of the selection process are presented in Figure 1 with the reasons for exclusion detailed in File S1.

F I G U R E 4
The effect of contact/collision sport participation on tCho concentration in longitudinal studies (baseline vs. the ToA).CI, confidence interval; ES, effect size; M, mean; RE, random effects; SD, standard deviation; ToA, Time of Assessment. of contact/collision sport participation on mI(Chamard et al., 2012;Poole et al., 2014;Schranz et al., 2020;Vike et al., 2022).The overall weighted mean effect indicated no significant change in mI concentration across the sporting season (i.e.baseline vs. the ToA; g = −0.357,95% CI: −0.755 to 0.041, p = 0.076, I 2 = 66.0%,
and suggests alterations observed across the sporting season may dissipate during the off-season or upon retirement.The other markers assessed (i.e.Glx, GABA and GSH) either demonstrated inconsistent effects or were only measured in one study.As such, the field would benefit from further case-control studies that examine neurometabolic alterations resulting from contact/collision sport involving cumulative subconcussive episodes.

3 | Risk of bias assessment
a MeSH term.2.
These control group were not included in the review because of the athletic population (Participants) having possible or not reported concussion histories.

Imaging Timing of assessment Strength of field (tesla) Pulse sequence TR/TE (ms) Voxel size (mm) Region of interest Metabolites measured
. As such, only the 'second half of the season' measurements were included in the moderator analyses (i.e. in the 'During Late Season' category) under the reasonable assumption the season was >12 weeks in length.All regression analyses were examined for influential cases and outliers (i.e.studentised residuals, Cook's distance and centred leverage values).Statistical significance was accepted as p < 0.05.