Since all of the subjects were specifically recruited for the investigation of normal brain development and their neurological, psychological, and structural measures were all within the normal range, we consider our data representative of a healthy population and are able to avoid any potential problem of transferability of results (Rivkin 2000
), within the limitations of our sample being skewed toward the higher end of the IQ range and containing significantly more girls than boys.
It has been well documented (Clark and Le Bihan 2000
; Mulkern, et al. 1999
) that the diffusion properties of brain tissue follow a multiexponential, rather than a monoexponential, model. The range of diffusion coefficients in vivo
may approach a continuum and not reflect only two components (Pfeuffer, et al. 1999
). The exact origins of the diffusion components continue to remain a topic of debate. For example, results from a recent spectroscopic study (Inglis, et al. 2001
) performed on an excised rat brain are consistent with the hypothesis that the faster-diffusing component corresponds predominantly to the extracellular water. However, studies performed on rat brains in vivo
(Duong, et al. 1998
; Duong, et al. 2001
) have challenged the assumption that intracellular and extracellular water have different intrinsic diffusion coefficients.
Our observed changes in FA and MD are due almost completely to a faster-diffusing component, as the relatively low b-value used in this experiment of 710 s/mm2
will minimize any significant contribution from a slower-diffusing component (Clark and Le Bihan 2000
). The fraction of the fast diffusion component has been shown to be larger in newborns than in adults (Mulkern, et al. 2001
), a result attributed to cellular development and myelination. However, since the lower age limit of our study is five years old, at which time gross myelination is virtually complete (Nakagawa, et al. 1998
), we do not expect there to be significant differences in the fast diffusion fraction. Still, further research will be necessary in order to confirm this hypothesis, especially with regard to evidence of ongoing white matter tract development in older children (Schmithorst, Radiology 2002
With regard to development, it has previously been proposed (Schmithorst, et al. 2002
) that there are at least two processes contributing to the maturation of white matter in the normal pediatric population: 1) increasingly dense and ordered packing of the fiber tracts, resulting in more directionally restricted extracellular space; and 2) changes in the intracellular compartment including a greater concentration of membranes and a greater membrane surface-to-cell volume ratio. Process 1) should lead to a marked increase in diffusion anisotropy as well as a possible decrease in MD, due to the increasingly restricted diffusion perpendicular to the axon direction; process 2) should lead only to a decrease in MD, as suggested by Baratti et al.
(Baratti, et al. 1999
) in a detailed longitudinal study investigating maturation of the cat brain.
Based on the diffusion properties of white matter laid out above, the areas exhibiting increases in FA with increasing cognitive abilities reflect an overall correlation of fiber organization and/or density with IQ. The areas which display negative correlations of MD with IQ should either display a greater concentration of macromolecules and a greater membrane surface-to-cell ratio due to a decrease in processes and organelles (Caley 1971
), but might also be compatible with a more dense extracellular space (corresponding to process 1 above). To examine this question further we investigated whether there was a significant correlation between the average MD values and the average FA values in the regions displaying correlations with MD (). As all regions display a significant negative correlation of the MD values with the FA values, we think it quite likely that the changes in MD do not represent a separate developmental process than the changes in FA, but are representative of the same mechanism (greater fiber organization), with the failure of some of these regions to reach significance on the FA correlational analyses due to insufficient sensitivity and/or sample size. Thus our results are in agreement with an earlier study (Schmithorst, et al. 2002
) investigating normal development of white matter in the pediatric population displaying regionally-specific increases in FA, with more widespread decreases in MD, and indicate that efficient organization of white matter association fibers is essential for optimal cognitive performance.
Correlations between the average mean diffusivity and average FA for regions listed in .
Our findings also show a striking spatial overlap with earlier findings of Klingberg et al
(Klingberg, et al. 2000
) on a small group of reading impaired and normal adults. They found a positive correlation of FA in a left temporo-parietal area with higher reading skills, and the spatial location of that region is extremely close to the region exhibiting changes in FA with general cognitive ability (, middle slice, middle row). Reading is a task that heavily relies on highly specialized brain areas in frontal, temporo-parietal, and occipito-temporal regions and their fast and efficient connection (Habib 2000
; Pugh, et al. 2000
). In fact, it has been hypothesized that in reading-impaired individuals, “temporo-parietal difficulties disrupt this developmental trajectory” (Pugh, et al. 2001
), which is consistent with neuroimaging studies demonstrating pathological patterns of posterior activation in reading-impaired subjects, with a compensatory shift involving stronger activation in frontal areas (Pugh, et al. 2001
). The DTI results in the reading-impaired (adult) individuals thus reflect the disruption of a normal (anatomical and functional) connection between cooperating brain regions. Our results, however, reflect normal variations in healthy children that correlate with their individual, overall cognitive abilities as measured by a broad range of tasks in a test of “general” intelligence (which is normal in most dyslexic individuals (Habib 2000
)). To further investigate we looked at the average FA from the ROI of correlated voxels in the slice at Z = +25 in left temporo-parietal white matter (detailed in ) and computed the partial correlation coefficients for the average FA versus Full-scale IQ (R = 0.58), Verbal IQ (R = 0.6), and Performance IQ (R=0.43), with subject age as the covariate, indicating that the majority of the FA differences in this area may be traced to verbal proficiency. Incorporating Full-scale IQ as well as age as a covariate, while neither correlation with Verbal IQ (R = 0.21) or Performance IQ (R = −0.19) reached significance the signs of the correlation coefficients provide a further indication of the relation of FA in this area to verbal proficiency.
Our results also match well with a recent voxel-based morphometry study investigating correlations of brain structure with intelligence in the normal pediatric population (Wilke, et al. 2003
). Gray matter volume positively correlated with cognitive function in anterior brain regions and in posterior temporal/inferior parietal regions. Our findings of increased FA in adjoining white matter areas are compatible with a shift of cognitive functions to frontal and temporal regions during normal development (Schlaggar, et al. 2002
; Wilke, et al. 2003
). Therefore, our findings could be related and secondary to gray matter developmental processes. However, more research addressing this issue will be necessary, especially with regard to questions relating to causality.
In order to investigate changes in white matter structure, we chose to use a well-known statistical software package (SPM99) (Friston, et al. 1995
) to remove gray matter and CSF voxels from further analysis. Since we desired to investigate changes in white matter microstructure rather than gross morphological differences related to intelligence, we chose to use a rather strict threshold in order to safeguard against partial-volume effects, despite the fact that that might result in the discarding of some voxels actually in white matter and a subsequent loss of sensitivity. This was necessary due to the rather low acquired spatial resolution (2 × 3 × 5 mm), which limits detectability of smaller fiber tracts, and due to the affine spatial normalization, which, while shown to be fairly reliable for subjects in our age range (Muzik, et al. 2000
; Wilke, et al. 2002
), is nevertheless not as robust as it is for adult subjects. In addition, since our analysis was, of its nature, exploratory, we chose to use a voxelwise approach than a region-of-interest approach for the analysis. Hence it is likely that some white matter areas actually related to intelligence and cognitive function were not found in our analysis.