Study design and participants
This cross-sectional study is part of the ActiveBrains project (https://profith.ugr.es/activebrains?lang=en), a randomized controlled trial, with the primary aim of examining the effects of exercise on brain, cognition and academic performance in children with overweight or obesity according to sex and age specific World Obesity Federation cut-off points16,17. The complete methodology of the project has been described elsewhere18. In total, 110 children with overweight or obesity, ages 8-to-11 years, were recruited from Granada (southern Spain). Of these, 104 (10.04 ± 1.15 years old; 43 girls) were included in the present analyses. Data were collected from November 2014 to February 2016. Parents or legal guardians were informed of the goal of the study and written informed parental and child consents were obtained. This study was conducted according to the Declaration of Helsinki, approved by the Human Research Ethics Committee of the University of Granada, and registered in ClinicalTrials.gov (identifier: NCT02295072).
Physical fitness components and magnetic resonance imaging (MRI) procedure
Physical fitness components
Physical fitness components (i.e., cardiorespiratory fitness, muscular fitness, and motor fitness) were assessed using the extended version of the ALPHA (Assessing Levels of Physical fitness and Health in Adolescents) health-related physical fitness test battery19. This battery has been shown to be valid, reliable, feasible, and safe for the assessment of the physical fitness components in children and adolescents19.
Cardiorespiratory fitness was estimated by the 20-m shuttle-run test20. This test was always performed at the end of the fitness battery testing session. The total number of completed laps were registered. Upper- and lower-body muscular fitness were assessed using the handgrip strength test and the standing long jump test, respectively. A digital hand dynamometer with an adjustable grip (TKK 5101 Grip D, Takei, Tokyo, Japan) was used to assess upper-body muscular fitness. Each child performed the test twice, and the maximum scores of left and right hands were averaged and used as a measurement of absolute upper-body muscular fitness in kilograms (kg). The standing long jump test was performed three times and the longest jump was recorded in centimeters (cm) as a measurement of relative lower-body muscular fitness. In addition, we computed a relative-to-body weight measurement from upper body muscular fitness (kg/body weight) and an absolute measurement from lower body muscular fitness (cm * kg), according to previous research in children with obesity21. Motor fitness was assessed using the 4 × 10-m shuttle-run test. Participants were required to run back and forth twice between two lines 10-m apart. Children were instructed to run as fast as possible and every time they crossed any of the lines, they were instructed to pick up (the first time) or exchange (second and third time) a sponge that had earlier been placed behind the lines. The test was performed twice and the fastest time was recorded in seconds. Since a longer completion time indicates a lower fitness level, for analysis purposes we inverted this variable by multiplying test completion time (s) by − 1. Thus, higher scores indicated higher motor fitness levels.
MRI data were collected with a 3.0 T Siemens Magnetom Tim Trio scanner (Siemens Medical Solutions, Erlangen, Germany). Diffusion tensor imaging (DTI) data were acquired using an echo planar imaging (EPI) sequence with the following parameters: repetition time (TR) = 3,300 ms, echo time (TE) = 90 ms, flip angle = 90, matrix = 128 × 128, field of view (FOV) = 230 mm × 230 mm, slice thickness = 4 mm, number of slices = 25 and voxel resolution = 1.8 × 1.8 × 4 mm3. One volume without diffusion weighting (b = 0 s/mm2) and 30 volumes with diffusion weighting (b = 1000 s/mm2) were collected.
DTI is able to sample features of the microstructural architecture of white matter22. To quantify total DTI metrics, we use fractional anisotropy (FA) and mean diffusivity (MD), as two of the most common derived scalar metrics from DTI23. FA expresses the degree to which water diffuses preferentially along one axis, and has shown to increase with age23 during development and to be lower in the context of various neurological and psychiatric diseases24. MD is a scalar describing the average diffusion in all directions, with higher levels indicating relatively unimpeded diffusion (i.e., negatively correlated with FA)25.
Functional MRI of the Brain Software Library (FSL) (https://fsl.fmrib.ox.ac.uk) was used to processed MRI data26,27. First, images were adjusted for minor head motion28, which included a Gaussian process for outlier replacement29. Then, the resulting transformation matrices were used to rotate the diffusion gradient direction table30,31. Non-brain tissue was removed using the FSL Brain Extraction Tool32. Lastly, the diffusion tensor was fit, and common scalar maps (i.e., FA and MD) were subsequently computed.
Probabilistic fiber tractography
Fully automated probabilistic fiber tractography was performed using the FSL plugin, “AutoPtx” (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/AutoPtx). Diffusion data were processed using the Bayesian Estimation of Diffusion Parameters Obtained using Sampling Techniques (BEDPOSTx), accounting for two fiber orientations at each voxel33,34. Then, for each subject, the FA map was aligned to the FMRIB-58 FA template image with the FSL nonlinear registration tool (FNIRT). Next, the inverse of this nonlinear warp field was computed, and applied to a series of predefined seed, target, exclusion, and termination masks provided by the AutoPtx plugin35. Probabilistic fiber tracking was then execute with the FSL Probtrackx module using these supplied tract-specific masks (i.e., seed, target, etc.) that were warped to the native diffusion image space of each subject33. Lastly, the resulting path distributions were normalized to a scale from 0 to 1 using the total number of successful seed-to-target attempts and were subsequently thresholded to remove low-probability voxels likely related to noise.
White matter tract segmentation was performed by thresholding the normalized tract density images based on previously established values by de Groot et al.35 (i.e., cingulate gyrus part of cingulum (CGC): 0.01, CST: 0.005, forceps major (FMA): 0.005, forceps minor (FMI): 0.01, inferior longitudinal fasciculus (ILF): 0.005, SLF: 0.001, uncinate fasciculus (UNC): 0.01). Average FA and MD values were then computed for each fiber bundle. Connectivity distributions were estimated for the 7 large fiber bundles previously named and selected based on previous reports36,37,38. Average of FA and MD in the left and right hemisphere was calculated in those tracts present in both hemispheres (i.e., CGC, CST, FMA, FMI, ILF, SLF, and UNC).
To assess whether physical fitness components (i.e., cardiorespiratory fitness, muscular fitness, and motor fitness) were related to global measures of white matter microstructure (i.e., global FA, MD), selected tracts were combined into a single factor (“global factor”). The global factor was computed by averaging all tracts and weighting this average by the size (volume) of the tracts.
Tract-based spatial statistics
Tract-based spatial statistics (TBSS) was used to perform voxel-wise statistical analyses of the DTI data (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/TBSS/UserGuide39. A mean FA image was calculated and thinned to create a mean FA skeleton, which represents the center of white matter tracts. A threshold of FA > 0.2 was selected to exclude voxels not belonging to white matter. FA maps of each participant were then projected onto the skeleton. The same procedure was applied to the MD maps.
Image quality assurance
Raw image quality was assessed via visual inspection. In addition, the sum-of-squares error (SSE) maps from the tensor estimation were calculated and visually inspected for structured noise12. Image quality was rated using a 4-point scale, with 1 = “excellent”, 2 = “minor”, 3 = “moderate”, and 4 = “severe”. Datasets determined to be of insufficient quality (i.e., moderate and severe) for statistical analyses were excluded (n = 2). Lastly, probabilistic tractography data were inspected visually. First, the native space FA map registration was inspected to ensure images were all properly aligned to the template (masks were properly mapped to native space). Second, all tracts were visualized to ensure accurate path reconstruction.
Body weight and height were performed with participants having bare feet and wearing underclothes; weight was measured with an electronic scale (SECA 861, Hamburg, Germany) and height (cm) with a stadiometer (SECA 225, Hamburg, Germany). Both measurements were performed twice, and averages were used. BMI was expressed in kg/m2. PHV is a common indicator of maturity in children and adolescents40. PHV was obtained from anthropometric variables (weight, height and/or seated height) using Moore’s equations41. The total composite IQ was assessed by the Spanish version of the Kaufman Brief Intelligence Test (K-BIT), a validated and reliable instrument42. This test consists of vocabulary and matrices subtests which provided indicators of crystallized intelligence and fluid intelligence, respectively. The typical punctuation of both, crystallized and fluid indicators of intelligence, were computed and a total intelligence score was obtained from the sum of them. Parental education was assessed by the educational level of mother and father reported (i.e., no elementary school, elementary school, middle school, high school and university completed). Parent answers were combined into a trichotomous variable (i.e., none of the parents had a university degree, one of the parents had a university degree and both parents had a university degree). Lastly, the Behavior Assessment System for Children (BASC), level-2 for children aged 6–12 years old, was used to assess behavioral and emotional functioning. A total behavioral symptoms index (including aggressively, hyperactivity, attention problems, atypical behaviors, anxiety and depression) was extracted from the questionnaire43.
All analyses, with the exception of TBSS analyses, were performed using the Statistical Package for Social Sciences (IBM SPSS Statistics for Windows, version 22.0, Armonk, NY, P set at < 0.05). The characteristics of the study sample are presented as means and standard deviations (SD) or percentages. In addition, we tested the correlation of BMI with global DTI metrics and physical fitness components. Interaction analyses of sex with physical fitness variables were also performed. No significant interactions with sex were found (P ≥ 0.10) and therefore analyses are presented for the whole sample. In addition, we explored the association of several confounders (i.e., sex, PHV, BMI, IQ, parental education, and emotional and behavioral problems) with tractography-derived white matter variables using a Pearson’s bivariate correlation analysis (data no shown). Among all of the potential confounders, parental education, socioeconomic status, and emotional and behavioral problems were not significantly related to white matter microstructure (all P values > 0.1) and were therefore excluded from the subsequent analyses.
Separate linear regression analyses adjusted for sex, PHV, BMI and IQ were performed to examine the association between physical fitness components and global-extracted DTI scalar metrics (i.e., global FA and MD). Each regression model examined separately the relationships between a single physical fitness component and a single DTI scalar metric.
Then, in order to determine whether the association of physical fitness with white matter microstructure was indeed only global or restricted to a particular set of white matter bundles, and to facilitate comparison with future studies, we applied two commonly used methodologies: (1) probabilistic tractography of large, commonly studied white matter tracts and (2) TBSS, which is a voxel-based approach. For probabilistic tractography analyses, false discovery rate (FDR. Benjamini–Hochberg method) was used to adjust for multiple comparisons44. Correction for multiple comparisons was based on 7 tracts, 2 DTI metrics and 6 physical fitness components for a total of 84 tests. For TBSS analyses, the association between physical fitness components and DTI scalar metrics were tested voxel-wise using general linear models, including sex, PHV, BMI and IQ as covariates. A permutation-based statistical approach (5,000 permutations) within FSL’s Randomise39 was performed including the threshold-free cluster enhancement (TFCE) multiple comparison correction method. Significance was set at P < 0.05, corrected for family-wise error.