Cortical thickness and neurocognitive performance in former high‐level female soccer and non‐contact sport athletes

Long‐term effects of playing soccer (football) on the brain structure and function of the brain are vividly debated. While some studies showed differences in neurocognitive performance and structural brain changes in retired male players, data on female players are scarce. The present study compares cortical thickness and neurocognitive performance in former high‐level female soccer (SOC) and non‐contact sport athletes (CON).


| INTRODUCTION
Soccer is one of the most popular sports worldwide and is associated with various health benefits. 1,2 Besides these positive impacts, soccer players have also been reported to be at higher risk for neurodegenerative diseases, which might be affected by playing position and career length. 3 Although those studies might be interpreted with caution, because methodological aspects (e.g., disregard of modifying factors, selection biases, cross-sectional or retrospective study designs) impair generalization of study results, those results most certainly stimulate further scientific discussion. Possible negative (long-term) effects of soccer and other contact sports regarding neurodegenerative processes and their underlying mechanisms are still not fully understood. Factors that are discussed to influence long-term effects are sustained concussions and repetitive head impacts (RHIs), which are prominent in collision and contact sports, such as American football, rugby, ice hockey, or soccer. 4 Concussions seem to functionally and structurally alter the brain. 5 The latter might classically be revealed by diffusion tensor imaging analyzing the white matter (WM) integrity, which was found to be altered in concussed athletes. 6 In American football players, a history of concussion might also be associated with gray matter (GM) abnormalities like cortical thinning. 7 Moreover, volume loss in the amygdala and the right hippocampus was demonstrated in retired boxers compared to active boxers and a control group 8 and professional rugby players with a history of concussion and RHIs exhibited smaller whole brain, smaller bilateral hippocampal, and smaller left amygdala volumes. 9 A recent review summarized the effects of a concussion history in retired athletes from different sports on neurocognitive performance. Memory, executive functioning, and psychomotor function seem to be negatively associated with a concussion history. 10 Furthermore, (mild) neurocognitive deficits were found in former elite rugby players, 11 one of the sports with a high risk of concussion 4 and a high exposure to RHIs. Overall, comparison and generalization of research on concussion as well as on RHIs is limited due to heterogeneous samples with variable injury severity or definition and various RHI characteristics. 4,14 Soccer is the only sport, in which the head is purposefully and repetitively used to play the ball. Thus, RHIs are an area of concern in soccer in addition to concussion. 12 But in contrast to concussion, results on (long-term) effects of RHIs are still heterogeneous. 13,14 Reports on potential micro-and macrostructural changes in relationship to headers have sparked a vivid discussion in the scientific community and in (professional) sports. 12 Female soccer players demonstrated greater WM alterations than male soccer players, 15 but the underlying stimuli and clinical consequences still remain to be elucidated. Retired professional male players showed a significantly greater decrease of cortical thickness (the distance between GM/ WM-and GM/cerebrospinal fluid (CSF) boundary) with age than a matched control group of swimmers in brain regions including the temporal, parietal, and occipital lobes bilaterally. 16 In contrast, Oliveira et al., 17 did not find any associations of cortical thickness or less GM volume with heading in amateur soccer players. However, the variability of study designs, utilized methods, and examined cohorts often make it difficult to pool results and eliminate nuisance factors, when causality of cerebral changes and adaptations to playing soccer or headers should be established.
Similar to structural changes, the long-term effect of RHIs (i.e., heading) on neurocognitive performance is not clear. Koerte et al. 16 reported decreased memory performance in former professional, male soccer players when compared to non-contact sport athletes. By contrast, Guskiewicz et al. 18 found no impaired neurocognitive performance in collegiate soccer athletes. Prien et al. 19 reported that retired elite female soccer players performed similar to age-matched elite female non-contact sport athletes on various neurocognitive tests, except for significantly lower scores on verbal memory and verbal fluency tests. To evaluate neurocognitive performance in relation to concussion and to RHIs a large variety of tests was used. Computerized test batteries or structured individual testing, however, may cover all domains that are possibly affected by concussion. Similar to studies investigating effects of concussion, studies on heading and neurocognitive performance are characterized by heterogeneous designs (especially concerning determination of external loads/headers). When assessing the association between heading frequency and neurocognitive performance, results are somewhat inconclusive showing either no effects of heading 20,21 or better performances in verbal and visual memory, motor processing and speed in female soccer players compared with male soccer players. 22 Functional deficits within the spectrum of traumatic brain injury may be associated with underlying alterations of architecture of brain network. WM integrity in the corpus callosum, fornix, internal capsule, arcuate, and uncinate fasciculi, for example, has been linked to memory and/or attention deficits following traumatic brain injury, although establishing the link between structural network integrity and cognitive deficits still needs to be replicated. 23 How those changes might relate to RHIs (vs. traumatic brain injuries) or affect cortical architecture within the affected functional networks (vs. diffusion tensor imaging-based WM integrity) is largely unknown, although those relationships may not appear unlikely, when plastic changes in aging and pathological changes in neurodevelopmental disorders are considered and regarded as models. 24 In general, many more studies on heading and RHIs investigated male soccer players, although female soccer players had a higher incidence of concussions than male players, 4 which might be due to different neck strength or females being more open on reporting such incidences. 25 Furthermore, concussed females report a higher symptom burden and longer recovery periods. 25 Therefore, females might also be more vulnerable to long-term effects of RHIs than male players 26 despite the tendency toward less heading and subjective overestimation of heading burden in females. 27 Most studies investigating potential long-term effects of soccer, however, only rarely include and focus on female soccer players. 14 Although some studies differentiate between the sexes 15,22 others do not. 17,18 Furthermore, studies focusing on female soccer cohorts are limited by only investigating a specific age range of players (e.g., highschool or college 20,21 ).
These aspects and various methodological differences and inconclusive results necessitate further investigations regarding potential (long-term) effects of soccer on structural brain changes and the neurocognitive performance, specifically in high-level retired female soccer players. Therefore, the aim of this study was to compare cortical thickness and neurocognitive performance of former high-level female soccer players to a cohort of former high-level female non-contact sport athletes.

| Design
This case-control study was approved by the ethical review boards of the Westfalian Medical Board, Münster, Germany (2016-449-f-S) and the VU University Medical Center Amsterdam, the Netherlands (2017.360).

| Study procedure
A subsample of a cohort presented in detail elsewhere (see Prien et al. 19 ) was analyzed in the presented study. Female soccer players (SOC) who retired at least 2 years before the study and who played in the first league and/ or the national team and non-contact sports athletes (CON) who participated at national championships and/ or for the national team were included. Exclusion criteria for participation were: 1. severe non-sport related head injury (diagnosed by a physician) or 2. disease affecting the central nervous system, or for CON: 3. more than one concussion or any severe head injury. SOC and CON were matched for age and level of competition on a national level.
After informed consent was obtained, subjects underwent an online survey, a semi-structured interview, and neurocognitive testing. The subsample presented in this study additionally had an MRI scan of the brain. Demographic and sport-related characteristics were obtained using an online survey. The semi-structured interview was used to gather information about the history of concussion, which was defined as proposed by the Concussion in Sports Group, 28 other injuries to the head, neck, or face, and heading frequency. SOC had to characterize themselves as a rare, moderate, or a frequent header.

| MR imaging data acquisition and image processing
Whole-brain MR imaging was performed on a 3 T Philips Ingenia_CX with a 32-channel receive-only head coil, located at the Spinoza Centre for Neuroimaging, Amsterdam, the Netherlands. For the determination of cortical thickness, T1-weighted magnetization-prepared rapid acquisition gradient-echo images were obtained (MPRAGE, repetition time 8.3 msec; echo time 3.8 msec; inversion time 929 msec; flip angle 8°; field of view 240 mm × 188 mm (AP × RL); slice thickness 1 mm). The quality of the dataset was checked visually and automatically using MRIqc (v. 0.15.2rc1). 29 No data had to be rejected because of quality issues.
Preprocessing of the T1-weighted (T1w) images was executed by use of the automated pipeline of fMRIprep (v. 20.0.0), a Nipype based tool 30 combining established software packages for neuroimaging data, including FSL, ANTs, FreeSurfer, and AFNI. 31 In summary, the following steps were carried out: First, all T1w images were corrected for intensity non-uniformity (INU) by using N4BiasFieldCorrection (v. 2.1.0) 32 and skull-stripped using antsBrainExtraction. sh (v. 2.1.0) with the OASIS template. Recon-all from FreeSurfer (v. 6.0.1) was used for reconstruction of brain surfaces and the brain mask estimated previously was refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentations of cortical gray matter of Mindboogle. 33 Spatial normalization to the ICBM 152 Nonlinear Asymmetrical template version 2009c 34 was performed through nonlinear registration with the antsRegistration tool of ANTs v2.1.0, 35 using brain-extracted versions of both T1w volume and template. Using the "fast"-stream from FSL (FSL v. 5.0.9) on the brain-extracted T1w brain tissue segmentation into CSF, WM, and GM was performed.
Data analysis was based on FreeSurfer's "recon-all" processing step in the fMRIprep pipeline. Surfaces were first checked manually for quality and corrected, when necessary, using FreeSurfer's freeview (v. 6.0.0). To include the corrected surfaces, "recon-all" was carried out again with repetition of necessary steps on the corrected data. For the parcellation of the cerebral cortex the Desikan-Killiany Atlas 36 was used.
Utilizing the intensity and continuity information from the entire three-dimensional MR volume in segmentation and deformation procedures cortical thickness was calculated as the closest distance between GM/ WM boundary and GM/CSF boundary at each vertex of the surface. In a next step, the data were assembled. Subjects were already resampled onto FreeSurfer's average subject ("fsaverage") and smoothed at various full width/half maximum (FWHM) values in the "reconall" step. With FreeSurfer's function "mris_preproc" all smoothed thickness data with a 10-mm FWHM from each subject was stacked into one file and used for further group analysis.

| Neurocognitive performance
For the evaluation of the neurocognitive performance the computerized test battery CNS Vital Signs (Morrisville, NC, USA) consisting of seven tasks generating 11 domain scores was used. As in the study of Prien et al., 19 seven domain scores were included in the analysis (see Table 1). Excluded domain scores were considered redundant. In accordance with Prien et al. 19 and for ease of interpretation, raw scores were automatically standardized to a normative dataset of subjects with an age between eight and 90 years and to a mean of 100 and a standard deviation (SD) of 15. Standardized scores were compared between the groups based on sport, concussion history, and heading frequency. For the cortical thickness analysis between the groups, FreeSurfer's general linear modeling (GLM) function ("mri_glmfit") was used with a "different onset, different slope" scenario. GLM allows a whole brain vertex-wise surface analysis 37 so that no region-of-interest boundaries affect the results. In the GLM, the particular group (SOC or CON; concussion history; heading frequency) was the discrete variable, and age was applied as nuisance regressor in the model. Multiple comparison correction was performed with a Monte Carlo Null Z Simulation. 10 000 iterations and a clusterwise p-value, as the probability of seeing a maximum cluster of that size or larger during the simulation, of 0.05 were applied using the clusterwise procedure implemented in FreeSurfer ("mri_glmfitsim"). Results were visualized by overlaying significant clusters on the inflated cortical surfaces of the "fsaverage brain", an average and standard space into which the individual subject data was registered to. Image processing and MRI data analyses were run on a high-performance computing cluster. Neurocognitive performance was compared between the groups (SOC minus CON; concussion history; heading frequency) using a multivariate linear model with age as control variable. The comparison between the main effects was adjusted by Bonferroni to test for multiple comparisons. Correlations between cortical thickness and neurocognitive performance were calculated with Pearson correlation. Alpha was set at p < 0.05 for all analyses.

| Sample
The present study analyzed a subsample of a larger cohort study on former high-level female athletes 19 who participated in the MRI. The 15 SOC (mean age 38.3 ± 5.1 years) and 16 CON (mean age 36.6 ± 5.8 years) were similar regarding demographical information and sports career characteristics (see Table 2) and did not differ from the larger cohort. CON included different non-contact sports (volleyball (n = 7), korfball (n = 3), hockey (n = 2), (wheelchair) tennis (n = 2), ice skating (n = 1), and athletics (n = 1)). Abbreviations: CI, confidence interval; CON, non-contact sport athletes; SD, standard deviation; SOC, soccer players.

T A B L E 2
Comparison of demographic and sport-related data of soccer and non-contact sport athletes.

| Cortical thickness
No significant differences were found between SOC and CON nor between subgroups with different concussion history after correction for multiple comparisons. There were no differences between never headers and all other heading subgroups. Frequent headers did not differ to moderate and rare headers. Greater cortical thickness was found for moderate headers compared with rare headers in one cluster (clusterwise p = 0.016; see Supplemental Material Appendix S1).

| Neurocognitive performance
SOC and CON performed similar in neurocognitive tests in five of seven domains (Table 3)

| Correlation cortical thickness and neurocognitive performance
The correlation of cortical thickness values of the inferior parietal region and the seven domains of neurocognitive performance did not reveal significant results, except for a negative correlation with visual performance in the CON group which is identical to the never header subgroup (Pearson correlation coefficient = −0.523, p = 0.038). The same correlation, with a different direction, was found in the frequent header subgroup (Pearson correlation coefficient = 0.721, p = 0.044), but not in the moderate or rare header subgroups (Figure 1). Further statistical investigations of this correlation in other subgroups (e.g., concussion) were not feasible due to the small sample size.

| DISCUSSION
This study compared cortical thickness of the brain and neurocognitive performance of 15 retired high-level female soccer players with a matched control group of 16 retired high-level female non-contact sport athletes. No cortical thinning in SOC compared with CON could be demonstrated. Self-reported heading frequency was not T A B L E 3 Mean differences between soccer and non-contact sport athletes in the CNS Vital Signs domains.

F I G U R E 1
Correlation between cortical thickness and visual memory for frequent (n = 8) and never (n = 16) headers. associated with differences in cortical thickness, except for a greater thickness in the right inferior parietal region when moderate headers (n = 4) were compared to rare headers (n = 3). SOC and CON performed similar in five neurocognitive domains, but SOC performed worse on psychomotor speed and verbal memory than CON.
No relationship of neurocognitive performance and selfreported heading frequency was found except for a worse verbal memory of soccer players, who classified themselves as moderate headers compared to rare or frequent headers. Significant correlations were found between visual memory performance and cortical thickness in the frequent header group and CON, although directions were opposed with frequent headers showing a positive and CON showing a negative correlation.

| Cortical thickness
Results on cortical thickness are in contrast to previous observations in retired male soccer players that revealed cortical thinning in right inferolateral parietal, temporal, and occipital regions when compared to swimmers, which increased with age. 16 Besides the differences in sex, the male cohort was older and the technique used to calculate cortical thickness (voxel based) was different from our study, which might contribute to the different results. Recently, Oliveira et al. 17 did not find concussion or heading to be associated with reduction in gray matter volume either, but rather described an association between heading and greater volume of the left inferior parietal cortex in a large cohort of amateur male and female soccer players. Furthermore, greater cortical thickness was recently seen in American football athletes compared with volleyball players in a longitudinal approach. 38 In our study, after correction for multiple comparisons, SOC who headed the ball moderately often demonstrated a bigger cortical thickness in the right inferior parietal region than players who rarely headed the ball (see Supplemental Material Appendix S1). The absolute numbers furthermore show that frequent headers have the highest cortical thickness compared to moderate, rare and never headers (see Supplemental Material Appendix S1) but this result is not significant, and the differences were small. The small sample size and unknown clinical modifiers of this possibly neuroplastic effect (e.g., lifestyle and diet) might forbid to draw more detailed conclusions. A more in-depth investigation of sport-specific aspects (e.g., techniques or player positions) might be needed to gain a more detailed insight in the results. While for example less GM volume decrease has been shown in American football players with greater position-based head impact risk 38 , correlations of structural changes with specific positions in other types of sports are still sparse.
Since the location and lateralization of greater cortical thickness are still not uniform, the interpretation of these findings still seems speculative. Training per se might be associated with gray matter changes, as described by a recent study revealing that aerobic exercising was associated with greater cortical thickness values in middle-aged adults (>45 years) in motor and somatosensory brain regions. 39 But underlying mechanisms are still discussed and sample sizes as well as cross-sectional study designs still forbid any further conclusions. Soccer training (including heading) resembles a complex motor skill as players need to coordinate their whole body very precisely and especially the right inferior parietal region might have a key role for higher-order motion regulation and eye-body coordination. 40 Since there was no difference in the current level of physical activity, training sessions per week and competitions per year during the respective careers of the SOC and CON in our study, the intriguing yet at this point speculative possibility of training-associated structural differences (of parietal cortical regions) should be the target of larger longitudinal studies to possibly identify features of "healthy" headers and heading training.

| Neurocognitive performance
The significantly worse performance of the SOC in the psychomotor speed domain is in concordance with study results by Prien et al., 19 and Stewart et al. 41 In the latter study, an association between heading and psychomotor speed was also described, which was not seen in our study. Worse memory performances in verbal, but not in visual domains in soccer players, as found in our study, have also already been reported in different soccer cohorts. 16 Although a thinner cortex was associated with lower cognitive processing speed, Koerte et al. 16 did not find significant differences between soccer players and a control group, which can also be seen in our study. The similar performances of SOC and CON in the processing speed domain may be explained by the level of expertise of the study subjects since elite athletes have been found to perform better in processing speed than amateur athletes 42 or by a higher level of general motor coordination. 43 Investigating potential effects of a history of concussion or header frequencies based on subjective judgment yielded significant differences in the verbal memory, but in no other neurocognitive domain. CON performed significantly better than moderate headers or soccer players with a history of one concussion. The lack of similar differences in soccer players with a higher header frequency or more concussions leaves the interpretation of this finding debatable. Since subgroup analyses were performed with very small sample sizes, those results should be interpreted with great caution and should not be generalized. In male cohorts of retired soccer players, there was no uniform neuropsychological deficit that can be attributed to heading in general, but executive function, attention, visuospatial, or memory performance in general might be related to headers at least in some studies. 14 Overall and similar to studies on (retired) male professional soccer players, our study did not demonstrate a clear influence of heading on neurocognitive performance.

| Correlation cortical thickness and neurocognitive performance
Cortical thickness and neurocognitive performance were not correlated with only two exceptions in the subgroup analyses. Visual memory and right inferior parietal cortical thickness were positively correlated in frequent headers and negatively correlated in CON (identical to the never header subgroup), but not in the other header groups. Future studies should analyze the relation of structural and functional changes of the brain and their clinical significance.

| Limitations and future studies
The presented study has several limitations. The sample size was rather small, but the groups did not differ significantly in any demographical data and the athletic profiles were quite similar. Subgroup analysis, however, lead to even smaller samples so that player position analyses could not be conducted. Heading was subjectively assessed and not objectively counted. The necessity of direct player observation in youth soccer has already been described and pointed out. 44 This should be implemented in future studies as well, as self-report methods are not validated and it has to be assumed that not every athlete is able to accurately estimate the number of headers. 27 The presented case-control study design is useful to generate hypotheses but restricts deriving causality from the results. 45 Therefore, prospective as well as longitudinal studies are needed for the evaluation of potential long-term effects of heading. 14 All participants were middle-aged and results might be different in an older cohort. Psychomotor speed results between SOC and CON are substantially different to the larger cohort 19 which might be due to a selection bias.

| PERSPECTIVE
To the best of our knowledge, this is the first study that compared cortical thickness and neurocognitive performance of retired high-level female soccer players to a matched group of non-contact sport athletes. SOC and CON had a similar cortical thickness and performed similar in a well-established neurocognitive test battery, except worse results in psychomotor speed and verbal memory of SOC. No clear relationship to the subjectively perceived heading frequency or history of concussion could be established. These results need to be confirmed in future studies with larger sample sizes and ideally a prospective design and an objective documentation of concussion and heading frequency.

AUTHOR CONTRIBUTION
FKH contributed to data analysis, interpretation of the data, drafting and writing of the manuscript, revision of the manuscript, approval of the final version to be published. AP contributed to study design, data collection, data analysis, interpretation of the data, approval of the final version to be published. LD contributed to study design, data analysis, interpretation of the data, approval of the final version to be published. NF-D contributed to study design, interpretation of the data, approval of the final version to be published. AJ contributed to study design, interpretation of the data, revision of the manuscript, approval of the final version to be published. CR contributed to study design, interpretation of the data, drafting and writing of the manuscript, revision of the manuscript, approval of the final version to be published.

FUNDING INFORMATION Fédération Internationale de Football Association (FIFA).
The Funder had no influence on the study design; the collection, analysis, and interpretation of the data; the writing of the manuscript nor on the decision to submit the paper for publication.

CONFLICT OF INTEREST STATEMENT
CR receives grant support from the German Federal Institute of Sports Sciences and the Heinz Nixdorf Westfalian Foundation on projects not related to the presented topic. He serves as hygiene officer and as a member of the medical commission of the German Football Association (DFB) and is a medical advisor on concussion for the Union of European Football Associations (UEFA). All other authors declare no conflicts of interest.