Participants in this study were children followed either at the children psychiatry department of Lyon's Neurological Hospital or the PSYRENE Center, both specialized in HIQ evaluation and psychological follow-up. Children were also recruited by placing ads in public schools (mostly for the recruitment of control subjects). Prior to enrollment, all children received a medical examination and were fully informed along with their parents of the study details. Children with any neurological, psychiatric disorders, learning disabilities such as dyslexia or dyspraxia, or MRI contra-indications were excluded from the study. None of the children received any medical treatments. Ethical committee approval (“CPP Sud-Est IV”) and written informed consent from the children and their parents were obtained. Three other subjects were excluded from this study: one for his inability to perform the MRI exam and two for motion artifacts.

All children underwent the WISC-IV with its four subscales [Verbal comprehension index (VCI), Perceptual reasoning index (PRI), Working memory index (WMI), and Processing speed index (PSI)] to calculate the FSIQ. Children scoring 130 or greater in VCI or PRI were included in HIQ groups. Children with a VCI significantly higher than the PRI were included in the Het-HIQ group, while those with no significant difference between VCI and PRI were included in the Hom-HIQ group. The significance of VCI-PRI differences was defined by the WISC-IV score with a range of 15–46 in the Het-HIQ group. Since the number of Het-HIQ children with significantly higher PRI than VCI was under represented (3 subjects out of 27), these subjects were excluded for group homogeneity purposes.

A total of 44 right-handed (according to the Edinburgh handedness scale Oldfield, 1971) HIQ children were included in this study. The study consisted of 8 girls and 36 boys aged between 8 and 12 years old [mean (± SD) age = 10.4 ± 1.3 y]. The Hom-HIQ group was composed of 20 children [5 girls and 15 boys; mean (±SD) age = 10.2 ± 1.2 y] with a FSIQ of 139.9 ± 11.1. The Het-HIQ group was composed of 24 children [3 girls and 21 boys, mean (±SD) age = 10.5 ± 1.4 y] with a FSIQ of 129.4 ± 10.6. Thirteen age-matched subjects [6 girls and 7 boys, mean (±SD) age = 10.5 ± 1.2 y] with a FSIQ of 105.2 ± 8.8 served as control subjects. Except for the two subjects excluded for motion artifacts, the quality of MRI data was optimal.

All children were assessed using the adapted French version of the Child Behavior Checklist (CBCL) for children from ages 4 to 16 (Achenbach, 1991). The CBCL is a 113-item parental report inventory that uses a Likert response format (not true, somewhat true, very true) for a variety of behaviors. The total Competence score was obtained from three subscales: activities, social, and school. The total Problem score was obtained from eight subscales: (I) withdrawn, (II) somatic complaints, (III) anxious, (IV) social, (V) thought, and (VI) attention problems, (VII) delinquent and (VIII) aggressive behavior. Two other subtotal scores, internal and external, were obtained from subscales (I+II+III) and (VII+VIII), respectively. Total scores were calculated by summing items, which was transformed by square root to approximate a normal distribution. Individual raw scores were converted into T-scores.

The Conners' Parent Rating Scale (CPRS) (Conners et al., 1998) was used to assess symptoms of Attention Deficit Hyperactivity Disorder (ADHD), and included six subscales: hyperactivity/impulsivity, psychosomatic, learning problems, anxiety, and conduct disorder. Individual raw scores were converted into T-scores.

MRI examinations were performed without any sedation at the MRI department of CERMEP-Imagerie du Vivant, on a 1.5T Siemens Sonata system (Erlangen, Germany) with an 8-channel head-coil and 40 mT/m gradients. The MRI protocol included a 3D T1-weighted magnetization prepared rapid gradient echo (MPRAGE) sequence (time of repetition/time of echo/time for inversion [TR/TE/TI] = 1,970/3.93/1,100 ms; flip angle = 15°; matrix size = 256 × 256; field of view = 256 × 256 mm; slice thickness = 1 mm; voxel size = 1 × 1 × 1 mm³; acquisition time = 8 min).

DTI protocol was based on a 2D multi-slice spin-echo echo-planar imaging (EPI) sequence (TR/TE = 6,900/86 ms, matrix size = 96 × 96, FOV = 240 × 240 mm, acquisition time = 7 min). Fifty-one contiguous axial slices, of 2.5 mm thickness were acquired in the anterior commissure—posterior commissure (AC-PC) plane. Twenty-four diffusion gradient directions (b = 1,000 s/mm²) were applied with a nominal isotropic resolution of 2.5 mm³. The b0 image was acquired four times to increase signal to noise ratio while the other directions were acquired twice. This protocol follows international recommendations on DTI acquisition (Jones, 2004).

Both T1-weighted, and DTI data were visually inspected (by S.H, D.S-M, and O.R) in order to detect artifacts arising from subject motion or scanner malfunction, and confirm the lack of visually detectable abnormalities.

Analysis of diffusion tensor data was performed using the Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library (FSL) (Smith et al., 2004). Eddy current correction using FMRIB's Diffusion Toolbox (FDT) was first applied followed by a non-brain voxels extraction using the Brain Extraction Tool (BET) with a factor of 0.35. FA and MD as well as AD and RD maps were then generated. Individual DTI maps were visually inspected for the presence of significant residual motion or other artifacts.

TBSS analysis of FA maps first consisted in aligning all subjects' data into a common space by means of a nonlinear registration. Since children brain anatomy is slightly different from adult's, the common template used in FSL was thereby not adapted for our subjects. Therefore, all subjects' FA images were aligned to every other one in order to find the most “typical” subject as a target image to align all other. A mean FA image was then created and used to obtain a mean skeleton of the major WM tracts on which all aligned subjects FA data were then projected. The resulting data were next fed into a voxel-wise statistical analysis, performed to identify FA differences in areas between HIQ and Control subjects, and between Hom-HIQ and Het-HIQ groups, including age and sex as covariates. For a better assessment, TBSS was also applied to diffusion-derived data other than FA. The nonlinear warps and skeleton projection were therefore applied to AD, RD, and MD maps. The resulting warped maps were then merged and projected onto the original mean FA skeleton. Voxel-wise analysis was finally performed exactly as for the FA maps.

The employed voxel-wise analysis was based on a non-parametric approach using the permutation test theory with a standard generalized linear model design matrix. By allowing inference on the statistical maps when the null distribution is not known, this approach provides an easily implementable solution to the multiple testing problems. The permutation testing was performed using the Randomise module of FSL. Group comparisons were performed between each group of HIQ children and control subjects with age and gender as co-factors. The threshold-free cluster enhancement (TFCE) option of Randomise was applied on the resulting statistical maps. This method enhances cluster-like structures in an image without having to define an initial cluster-forming threshold or to carry out a large amount of data smoothing. The resulting statistical parameter maps were corrected for sex and age and for multiple comparisons by the family-wise error rate (FEW-corrected p < 0.05). The anatomical location of significant clusters was identified based on WM atlases (JHU ICBM-DTI81 White Mater Labels and JHU White-Matter Tractography Atlas) in FSL. In order to confirm the voxel-wise analysis results, quantitative values of diffusion metrics were extracted from 48 regions of interest (ROI) and 21 fiber-bundles of the JHU atlases, by multiplying atlas labels with the TBSS skeleton obtained from all 57 subjects.

Statistical analysis was performed using the R library (https://www.r-project.org). ANOVA followed by Tukey's post-hoc tests were performed on demographic data, IQ, CBCL, and CPRS scores, and DTI metrics (ROI and fiber bundles) to detect group-wise differences between the four children groups (Total-HIQ, Hom-HIQ, Het-HIQ, and Control) with age, gender and CBCL as co-factors.

Figures 1, 2 illustrate the significant AD and FA differences observed when comparing HIQ and Control groups. No significant differences were reported for MD and RD diffusion metrics in any comparison.

Based on the JHU atlases, the diffusion metrics were measured in different ROIs (ICBM DTI-81) and fiber-bundles (White-Matter Tractography) in HIQ and Control groups to confirm our voxel-wise findings. When comparing Hom-HIQ and Het-HIQ groups, FA values were significantly increased in several ROIs, the body of corpus callosum, the left inferior cerebellar peduncle (ICP), the left posterior corona radiata (PCR), and the right fornix stria terminalis (FX/ST), and in one fiber bundle, the cingulum cingulate gyrus (CCG). When comparing with the Control group, AD values were significantly increased in 17 ROIs and six fiber bundles of the Tot-HIQ group, in four ROIs and three fiber bundles of the Hom-HIQ group, and four ROIs and two fiber bundles of the Het-HIQ group, as reported in Table 4.

Effects of full-scale IQ, and CBCL score (Achenbach) on AD values were estimated using General linear models for WM region reported in Table 4. The relative importance of these predictors was estimated for each WM region. A mean value of 77.88% (78.21% for JHU and 77.74% for ROI) and 22.12% (21.79% for JHU and 22.26% for ROI) was obtained for the full-scale IQ and CBCL score respectively. These results showed that differences observed in WM between HIQ and Control groups are mainly driven by IQ.

When compared to the Control group, both HIQ groups presented greater AD and FA in widespread WM regions of frontal, central and associative pathways. Central regions included mainly the body and the genu of the corpus callosum, the fornix and the cingulum as well as brainstem and cerebellar regions. These findings are in agreement with previous DTI studies on intelligence showing a bilateral fronto-parietal network as a neural substrate of enhanced information processing and intelligence (Dunst et al., 2014; Navas-Sánchez et al., 2014). Similarly, a previous fMRI study demonstrated the use of more extensive and bilateral cortical connections in gifted adolescents (Desco et al., 2011). Such improvement of bilateral and frontal WM organization contributes to improved cognitive performance in working memory, fluency, and executive functioning such as inhibition and task-switching (Seghete et al., 2013). Other studies suggested that corpus callosum related changes are crucial for higher cognitive capabilities (Singh and O'Boyle, 2004; Hutchinson et al., 2009; Prescott et al., 2010). Also, it has been demonstrated that the genu (Sidtis et al., 1981) and the splenium (Funnell et al., 2000) are involved in high order transfer of semantic information. In addition, the higher FA in the fornix may reflect a greater episodic memory (Gaffan and Wilson, 2008). Projection fibers of the capsule and the corona radiata also showed higher FA. These fibers regroup ascending fibers from the thalamus and descending fibers from the fronto-parietal cortex to subcortical nuclei. They constitute the backbone of perceptual and motor functions and other cognitive functions (Schmahmann and Pandya, 2008). Additionally, our study showed greater FA and AD in other regions of associative WM fasciculi, namely the uncinate, the superior longitudinal (SLF) and the superior fronto-occipital (SFOF). The SLF supports bidirectional networks involved in high order cognitive executive functions such as attention, inhibition as well as memory and language (Mesulam, 1998) while the uncinate (UNC) fasciculus may play a prominent role in semantic tasks (Catani and Mesulam, 2008; Catani et al., 2013). Thus, these findings show that both inter- and intra-hemispheric connections are enhanced in HIQ children suggesting that they may benefit from a better WM integrity, providing higher performances in most abilities.

When comparing Hom-HIQ and Het-HIQ groups to the Control group, we found AD increases in numerous WM regions. However, their spatial distribution was clearly different between the two HIQ groups. The Het-HIQ group showed more AD changes in the left hemisphere in contrast to the Hom-HIQ group which showed more AD differences in the right hemisphere.

The left hemisphere is well known to play a major role in lingual functions (Szaflarski et al., 2012) with a strong involvement of the SLF, the SFOF, and UNC fasciculi (Catani et al., 2002). In Het-HIQ group, AD is only significantly increased in the left UNC fasciculus while in Hom-HIQ group, it is also increased the right UNC, which is involved in emotion processing and memory, and plays an important role in lexical retrieval, semantic associations and naming aspects of the language (Grossman and Ash, 2004; Agosta et al., 2010; Galantucci et al., 2011). Thus, the left lateralization of increased diffusivity observed in Het-HIQ children can be associated to their high language performances (high VCI) in contrast to their relatively low visuo-spatial capacities (low PRI). Indeed, the right hemisphere is more specialized in visuo-spatial capacities such as facial recognition and attention monitoring (Gazzaniga, 2000; de Schotten et al., 2011; Caeyenberghs and Leemans, 2014).

Recently, Marinsek et al. (2014) proposed that both hemispheres diverge by their reasoning strategies. On one hand, the left hemisphere is considered as an interpreter, tending to reduce uncertainty by creating explanations, filling gaps of information, and making inference (Braun and Suffren, 2011). On the other hand, the right hemisphere is considered as a realist, striving to reduce inconsistencies between hypotheses and reality by detecting conflicts, updating information, supporting mental set shifting (Goel et al., 2013), and monitoring and inhibiting thoughts and behavior (Chatham et al., 2012). Moreover, Vendetti et al. (2015) found hemispheric differences in relational reasoning. Indeed, the left hemisphere has been shown to excel in relational encoding, particularly when the information is ordered linearly. It interacts more with itself, whereas the right hemisphere demonstrates connectivity patterns associated with both hemispheres (Gotts et al., 2013), and acts more as a monitor of evaluation and inhibition.

However, the complementarity or independence of hemispheric specialization remains an open question (Badzakova-Trajkov et al., 2016). The causal hypothesis suggests that strengths in language and weaknesses in visuo-spatial abilities of Het-HIQ may result from an earlier development of language, which thereby reinforces a hemispheric asymmetry specialization. Also, disabilities observed in Het-HIQ children may derive from an initial sub-development of the right hemisphere due to genetic and/or environmental causes.

FA is a useful measure of brain WM integrity that can be derived from DTI. It provides a simple and robust mean to assess the degree of anisotropic diffusion occurring within a region. FA is determined by both microscopic factors, such as myelination and axonal density, and macroscopic factors such as crossing fibers. Thus, the greater values of FA in HIQ children could be interpreted in terms of higher directionally coherent organization of fibers within voxels (Beaulieu, 2002), increased fiber tracts myelination (Jones et al., 2013), or higher axonal density and/or diameter (Mori and Zhang, 2006). However, increased FA could also be associated with less integrity, due to neurodegenerative processes in crossing fibers' regions (Douaud et al., 2011). Nevertheless, this effect is unlikely to occur in children. Furthermore, the increased FA observed in HIQ children is supported by the higher AD found in all HIQ groups compared to controls. AD quantifies the amount of water diffusion along the primary axis of the axon. It is believed to better reflect axonal density and/or caliber, while RD constitutes a more specific marker of myelin integrity (Song et al., 2005; Jones et al., 2013). Based on these assumptions, our findings suggest that HIQ children might present a greater axonal density or caliber than control subjects. The higher degree of axonal density or caliber could increase the signal transmission speed and thereby might account for more efficient brain functioning.

Among the diffusion metrics, significant differences were observed in FA and AD, but not in MD and RD. We can hypothesize that RD values might be decreased in HIQ children, leading to insignificant changes in MD but significant increase in FA. Such a hypothetical RD decrease would reinforce the argument of a higher degree of myelination in relation with intelligence (Dunst et al., 2014).

Nevertheless, this study may suffer from the relatively small number of children in each group. This is mostly related to the targeted population of children, which is difficult to recruit. Also, the ratio of girls and boys was different between HIQ and Control groups. This bias is probably related to girls' observed ability of “over-adaptation” which leads to less clinical consultations, thus reducing the number of girls diagnosed as HIQ, and particularly as Het-HIQ. In contrast, our control group was well balanced in that matter. This potential bias was taken into account by including age and sex as covariates in the voxel-wise statistical analysis.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.