NeuroImage 20 (2003) 202–215

www.elsevier.com/locate/ynimg

Bright spots: correlations of gray matter volume with IQ in a normal pediatric population Marko Wilke,a,b,* Jin-Hun Sohn,a Anna W. Byars,c and Scott K. Hollanda a

Imaging Research Center, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH 45229-3039, USA Department of Psychiatry, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH 45229-3039, USA c Department of Neurology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH 45229-3039, USA

b

Received 9 September 2002; revised 14 March 2003; accepted 14 April 2003

Abstract The localization of brain areas related to cognitive functions has yet to be thoroughly explored in children. We therefore set out to apply volumetric, voxel-based, and structural connectivity analyses to magnetic resonance images from a large sample of healthy children. We could confirm a strong correlation of whole-brain gray matter volume and the individual intelligence quotient; however, this correlation only developed with age in our sample, in that it was not present in the younger children. With the application of an optimized protocol for voxel-based morphometry, the anterior cingulate was shown to be directly correlated with a measure of human intelligence. Furthermore, an analysis of structural connectivity identified gray matter volume in several distinct brain areas to be related to cognitive functions. The implications of our findings for normal development, pathological processes, and our understanding of cognition are discussed and related to previous findings. © 2003 Elsevier Science (USA). All rights reserved.

Introduction Intelligence is a construct to explain individual differences in cognitive abilities as assessed by a wide range of tests. There is some controversy about whether intelligence is better viewed as consisting of one single factor (termed g, for general intelligence) or is more accurately indexed by several factors specific for cognitive subfunctions (Deary and Caryl, 1997). However, the most common measure derived from standard tests is an intelligence quotient (IQ), designed to be a single integrative measure reflecting a subject’s general intellectual capacity. While this concept has drawn criticism, it is in widespread use in daily practice and is an acknowledged parameter predicting academic achievement (Lezak, 1995). With regard to the localization of “intelligence” in the brain, however, surprisingly little is known. While visuospatial abilities have been ascribed to the parietal cortex and

* Corresponding author. Department of Pediatric Neurology, Children’s Hospital, University of Tu¨bingen, Hoppe-Seyler Strasse 1, 72076 Tu¨bingen, Germany. Fax: ⫹49-7071-29-5473. E-mail address: [email protected] (M. Wilke).

a multitude of individual cognitive subfunctions have been attributed to frontal brain regions (for review, see Cabeza and Nyberg, 2000), no one central executive region has been identified in the brain. This might be due to both the dispute over whether such localization studies make sense at all (Peters, 1995) and the fact that large numbers of subjects are needed to investigate the small changes in tissue volume attributable to IQ differences (Andreasen et al., 1993). The individual variability in brain morphology is even more relevant in childhood, when developmental processes take place (Lange et al., 1997). However, several studies have now shown a robust correlation of whole-brain and whole gray matter volume with IQ, although this issue has been discussed controversially (for review, see Rushton and Ankney, 1995). Moreover, lesion studies in different patient populations have detected correlations between local gray matter alterations and cognitive deficits (Peterson et al., 2000; Pinter et al., 2001; Karussis et al., 2000). Despite these results, no study has as yet addressed the relationship between local and regional gray matter volume and intelligence in a large and truly normal pediatric population. We set out to address this issue in the present study, applying automated procedures to

1053-8119/03/$ – see front matter © 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S1053-8119(03)00199-X

M. Wilke et al. / NeuroImage 20 (2003) 202–215

analyze whole-brain magnetic resonance (MR) images from 146 healthy, normal children. We used both volumetric and voxel-based methods in order to investigate global and regional as well as local effects. Also, a connectivity analysis was employed in order to detect cognition-related neural networks.

203

Subjects and methods

of IQ and regional gray matter volume. Due to the substantial influence on brain morphology, age (in months at date of MR exam), gender, and handedness were considered covariates of no interest (nuisance variables) in order to allow factoring out their effects (Amunts et al., 1996; Courchesne et al., 2000; DeBellis et al., 2001; Good et al., 2001; Lange et al., 1997; Pfefferbaum et al., 1994). In the analysis of age effects, age was considered the parameter of interest, while IQ was considered a nuisance variable.

Subjects

MR image acquisition and processing

Children were recruited as part of an ongoing study of normal language development (Holland et al., 2001). The following exclusion criteria were applied: history of previous neurological illness, head trauma with loss of consciousness, current or past psychostimulant medication, learning disability, birth at 37 weeks or less of gestational age, pregnancy, abnormal findings on clinical neurological examination, and clinical or technical contraindications to an MRI examination (including orthodontic braces). Institutional review board approval and informed consent were obtained for all subjects. A board-certified pediatric neuroadiologist read all MRI scans for the presence of structural abnormalities. Of 200 children examined, 148 children met the quality criteria (Wilke et al., 2002). The subjects included in this study represent a subset of a group investigated earlier (Wilke et al., 2002, 2003). Two more subjects had to be excluded for the purpose of this study due to missing data points. Therefore, 146 healthy children could be included in this study, 78 girls (53.4%) and 68 boys (46.6%). Mean age was 136.3 ⫾ 41.9 months (11.7 ⫾ 3.5 years); range was 60 –226.5 months (5–18.87 years).

Images were acquired on a Bruker Biospec 30/60 3-T MRI scanner, equipped with a dedicated head gradient insert (Bruker SK330). A whole-brain, T1-weighted Modified Driven-Equilibrium Fourier Transform (Ugurbil et al., 1993) image was acquired (TR ⫽ 15 ms, TE ⫽ 4.3 ms, t-time ⫽ 550 ms, flip angle ⫽ 20°, matrix ⫽ 128 ⫻ 256 ⫻ 96, FOV ⫽ 19.2 ⫻ 25.6 ⫻ 14.4 cm, resolution ⫽ 1.5 ⫻ 1 ⫻ 1.5 mm). Images were preprocessed as described before, including the elimination of images with moderate or strong motion or blood flow artifacts (Wilke et al., 2002). The following procedures were completely automated and utilized functions available within the statistical parametrical mapping software package (SPM99, Wellcome Department of Cognitive Neurology, University College London, UK [Friston et al., 1995]) running in Matlab (MathWorks, Natick, MA) unless stated otherwise. Images were resliced in the axial plane to achieve a better overlay with the axially oriented templates and to reduce partial voluming during further steps. From this step on, all images were written out to 1 ⫻ 1 ⫻ 1 mm resolution. Normalization was achieved with a combined linear (12 parameter) and nonlinear transformation, using 7 ⫻ 8 ⫻ 7 discrete cosine transform basis functions, aiming at minimizing both the sum of squared differences between image and template and the energy cost function of this transformation (Ashburner and Friston, 1999, 2000). Tissue segmentation was achieved using the combined pixel-intensity/a priori knowledge approach implemented in SPM99 (Ashburner and Friston, 1997). Since we could recently show that spatial normalization and segmentation in the pediatric age group are profoundly influenced by the morphological differences between a pediatric and a standard adult reference population, all of our processing was based on custommade pediatric data. Details of this procedure have been described elsewhere (Wilke et al., 2002, 2003). We implemented the recently suggested optimized protocol for voxel-based morphometric studies (VBM; Good et al., 2001). This protocol aims at minimizing the contribution of non-brain and non-gray-matter tissue to spatial normalization and segmentation and also allows for the investigation of true gray matter volume instead of the more abstract concept of gray matter density. It consists of the

Intelligence and language testing Children underwent an assessment of language (Oral and Written Language Scores, Carrow-Woolfolk, 1996) and intelligence, using the age-appropriate version of the Wechsler scale (Wechsler Preschool and Primary Scale of Intelligence, revised, Wechsler Intelligence Scale for Children, third edition, Wechsler Adult Intelligence Scale, third edition). These scales are the most sound and widely used measures of general intellectual capacity, with excellent reliability and validity (Wechsler, 1989, 1991, 1997). Abnormal language functions or a full-scale IQ score of less than 80 were considered exclusionary criteria. The tests were administered in close temporal proximity to the MRI scanning. Covariates The Full Scale IQ score (hereafter referred to as IQ) derived from the Wechsler scales was considered the parameter of interest in the analyses addressing the correlation

204

M. Wilke et al. / NeuroImage 20 (2003) 202–215

Table 1 Summary of the steps in image preprocessing and all analysis approaches Image processing (based on custom-made pediatric reference data) Segmentation (yields tissue probability map [native space]) Brain extraction (yields individual, binary brain mask) Cleaning (removes non-brain tissue) Determine normalization parameters (yields normalization parameters optimal for gray matter) Apply parameters to original image (yields optimally normalized whole brain image) Segmentation (yields tissue probability map [optimally normalized]) Brain extraction (see step 2) Cleaning (see step 3) Modulate with Jacobian determinant (reintegrates volume changes from spatial normalization) Image analysis Volumetric study Apply lobar mask (yields lobar tissue volume) Relate to mask size (yields relative tissue volume)

Voxel-based study Smooth [FWHM ⫽ 12 mm] (yields local tissue volume) Statistical analysis (see text for details)

following step: images are segmented in native space and “cleaned” by modulating them with an individually derived brain-tissue mask. This cleaned gray matter probability map is then normalized to the average gray matter probability map. The parameters of this normalization are applied to the original whole-brain image, which is segmented and finally cleaned again. To reintegrate the volume changes occurring during the combined linear and nonlinear spatial normalization, images were modulated with the Jacobian determinant of the transformation matrix, finally representing gray matter volume (Good et al., 2001). These normalized and modulated gray matter images were analyzed in different ways as described below. Table 1 shows an overview of our processing and analysis procedures. Volumetric study Data processing In order to investigate regional volumetric differences, we aimed at analyzing the brains on the level of brain lobes while avoiding the error-prone and time-consuming steps of manual delineation in all individual brains. To achieve this, a model recently put forward (Tzourio-Mazoyer et al., 2002) containing anatomical delineations of a representative brain into 90 regions was used to allow for the automatic delineation of brain structures. The suggested sublobar breakdown was considered too fragmented for the purpose of this study, so the following regions of interest were defined by condensing the data: frontal, occipital, temporal, and parietal lobe; deep gray matter structures (thalamus and basal ganglia), and the cingulate gyrus. As every mask was constructed independently for the left and the right side of the brain, this resulted in 12 individual data sets, plus an integrative measure representing whole-brain gray matter. In order to assure an adequate overlap of the resulting masks with our images, the original partitioned image was matched to an average of our final gray matter partitions;

Connectivity study Analysis of connectivity (see text for details)

these normalization parameters were then applied to the individual lobar masks. To further account for interindividual anatomical variability, the masks (structure of interest ⫽ 1, rest ⫽ 0) were smoothed with a Gaussian filter (full width at half maximum [FWHM] ⫽ 6 mm). The final volumetric measure was derived by determining the overall pixel intensity of an image obtained by multiplying the individual gray matter partitions with the mask. Fig. 1 shows an overview over the procedure. In order to allow for a comparison of the data regardless of mask size, the volumetric measure was related to the individual mask size, therefore representing average gray matter volume per voxel of the corresponding mask. Data analysis The data obtained from the volumetric study was considered to represent influences of the effects of the covariate of interest (IQ) and the nuisance variables (age, gender, and handedness), on top of the sizable individual variability in brain structure (Lange et al., 1997). To allow for the investigation of the covariate of interest, a multiple regression analysis was performed on the results from the volumetric study to factor out the effects attributable to the nuisance variables. The results from these analyses were then used to assess a possible correlation of IQ and gray matter volume in all data sets. In order to further assess developmental effects, results from the volumetric study were divided into three equally sized subgroups (termed “young,” “medium,” and “old”). These data were independently analyzed as described in the previous paragraph to assess age-dependent effects. To investigate and further characterize the effect of age, a multiple regression analysis was done on the whole group, using age as the variable of interest and thus factoring out gender, handedness, and IQ. These results should then be related to prior studies on normal brain development to validate our data and approach.

M. Wilke et al. / NeuroImage 20 (2003) 202–215

205

Fig. 1. Overview of the procedure of the volumetric study: the parcellated standard brain (left) was condensed into six brain regions, which were then smoothed and applied to the individual gray matter partitions, yielding masked brain structures; their volume was subsequently analyzed. Fig. 2. Gray matter volume (corrected for gender, handedness, and IQ) versus age, separately for all regions and global gray matter.

In this analysis, age was found to best correlate nonlinearly with global gray matter volume (see Results). Therefore, a further exploratory analysis was done to allow for a closer assessment of the exact nature of the relationship between IQ and gray matter volume. Here, the effects of gender and handedness were accounted for as above by doing a linear multiple regression, but age was then factored out doing a nonlinear regression, using a third order polynomial function. The results

from this analysis were then entered into a curve-fitting algorithm in order to find the best fit to explain the relation between IQ and gray matter volume in the cingulate and global gray matter. This best fit was determined using an algorithm minimizing the “merit function” (describing the disagreement between the data and the model). In this approach, the model parameters are adjusted until the value of the merit function becomes as small as possible.

206

M. Wilke et al. / NeuroImage 20 (2003) 202–215

Fig. 3. Gray matter volume (corrected for age, gender, and handedness) versus IQ, separately for the cingulate (blue dots, red line) and global gray matter (gray diamonds, gray line).

Voxel-based morphometry For the voxel-based analysis, the image volumes were smoothed with a Gaussian filter (FWHM ⫽ 12 mm). This step was necessary in order to condition the images to conform to the normality assumption underlying the tests applied during the analysis; it also determines the spatial scale at which structural changes are most sensitively detected (Ashburner and Friston, 2000; White et al., 2001) since it converts pixel-wise into local gray matter volume. Employing the framework of the general linear model, the data were analyzed within SPM99 using age, gender, and handedness as covariates of no interest. To account for global effects, the individual global gray matter volume was determined using a stand-alone Matlab script and was entered as a further covariate of no interest. Since this measure was highly correlated with IQ (see Results), we reparameterized IQ to form three groups: low, middle, and high IQ values in our sample. This new scale (designated compacted intelligence quotient [cIQ]) did not significantly correlate

with global gray matter volume, thus allowing for the assessment of regional (instead of global) effects. Two contrasts were calculated, testing for a positive or negative correlation of gray matter volume with the parameter of interest (cIQ). Significance was set at a P value of P ⫽ 0.0001, not corrected for multiple comparisons, and an additional extent threshold of 25 voxels as suggested before (Sowell et al., 1999; Wilke et al., 2001). Significance was not based on spatial extent since this has been shown to be too liberal in structural studies (Ashburner and Friston, 2000). Connectivity analysis In order to assess whether and which other brain regions follow similar patterns as the one detected to be correlated with IQ (see Results), we employed the framework of functional connectivity to our data. Briefly, this approach tries to detect regions in the search volume that, over the course of the experiment, show a pattern significantly correlated with the raw data pattern extracted from a selected region of

M. Wilke et al. / NeuroImage 20 (2003) 202–215 Table 2 Age and gender distribution of our sample Age Gender

5–6

7–8

9–10

11–12

13–14

15–16

17–18

Female Male

5 8

15 15

18 13

11 14

13 9

8 3

8 6

interest (Bu¨ chel and Friston, 2000). These raw data, in SPM99, correspond to the first eigenvariate (or the first/ principal component, derived using a principal components analysis) and explains most of the variability in the data. In our case, we aimed at detecting regions in the brain that show the same gray matter volume pattern as clusters significantly correlated with IQ (as detected in the prior analysis; see Results for details). The use of this individual regional regressor allows us to uncover effects not detectable by the more simple model used in the first analysis. As no a priori hypotheses could be established from our data regarding the location of these areas, significance was only assumed at a level of P ⫽ 0.01, corrected for multiple comparisons (Genovese et al., 2002), again excluding small clusters below 25 voxels.

Results Subjects Intelligence testing revealed an average IQ of 113.8 ⫾ 13.8, range 83–147. Language functions were normal in all children, and all but 15 subjects were right-handed. None of the epidemiological measures (IQ, gender, age, or handedness) correlated significantly with any of the other measures (or with image quality; data not shown). For complete data, see Tables 2 and 3. Global gray matter correlated significantly with gender, with boys having significantly more global gray matter (r ⫽ 0.18, P ⫽ 0.029). Age and IQ also correlated positively with global gray matter (see below); IQ was therefore parameterized to yield three groups (cIQ): low cIQ (58 subjects), middle cIQ (54 subjects), and high cIQ (34 subjects). These three groups were not significantly different with regard to age, gender, or handedness, but were significantly different with respect to IQ (low: 100.7 ⫾ 7.5, middle: 115.9 ⫾ 4.3, high: 132.7 ⫾ 6; P ⬍ 0.000001 for all comparisons). This simplified classification did not correlate with global gray matter. Since both age and global gray matter were considered nuisance variables, no attempt was made to decorrelate these variables. For the assessment of age effects in the volumetric study, the sample was subdivided into three age groups: young (48 subjects, 91.2 ⫾ 14.6 months). medium (49 subjects, 131.6 ⫾ 10.7 months), and old (49 subjects, 185.2 ⫾ 22.3 months). Again, these three groups were not significantly

207

different with regard to IQ, gender, or handedness, but were significantly different with respect to age (P ⬍ 0.000001 for all comparisons). Volumetric study Effects of age Age showed a strong negative correlation with global (r ⫽⫺0.67) and all regional measures of gray matter volume (range, ⫺0.52 [right cingulate] to ⫺0.69 [left occipital lobe]; P ⬍ 0.000001 for all correlations). The relation between age and gray matter volume was best expressed by a third order polynomial function, with little variation among the different regions (shown for the combined left ⫹ right results in Fig. 2). The pattern most different from the global relation was seen for the cingulate, which showed the smallest gray matter decline with age (largest, parietal lobe). Effects of IQ Our volumetric analyses showed a significant correlation between IQ and global gray matter volume (r ⫽ 0.28, P ⫽ 0.0006). When the three age groups were analyzed separately, this effect turned out to be dominated by the oldest age group: in the young and medium age group, no single volumetric measure (including global gray matter) reached significance when correlated with IQ. However, it is interesting to note that the strongest correlation with IQ in the young and medium age group could be found for the deep gray matter structures, while in the oldest and the whole group, the correlation was most significant in the cingulate. For the whole group, every single measure was significantly correlated with IQ (see Table 4). Remarkably, only the cingulate showed a correlation with IQ (r ⫽ 0.282/0.3 [L/R]) that was higher than the correlation between global gray matter and IQ (r ⫽ 0.28). Relationship between IQ and gray matter The relation between IQ and gray matter volume in the cingulate was best described by a third order polynomial function (r ⫽ 0.3026, P ⫽ 0.0002). Global gray matter followed the same trend, but was slightly less correlated with IQ (r ⫽ 0.279, P ⫽ 0.0006; Fig. 3). In both cases, the polynomial fit explained more of the variance in the data than a simple linear fit (cingulate: 9.1% versus 6.2%; global gray matter: 7.7% versus 5.2%).

Table 3 IQ distribution of our sample IQ 80–90 91–100 101–110 111–120 121–130 131–140 140–150 Gender Female 3 Male 3

13 7

21 19

15 21

15 9

7 9

4 0

208

M. Wilke et al. / NeuroImage 20 (2003) 202–215

Table 4 Correlation of gray matter volume measures with IQ in the whole sample and the three age subgroups, corrected for gender, handedness, and age (see text for details)

Young (n ⫽ 48) Medium (n ⫽ 49) Old (n ⫽ 49) Whole group (n ⫽ 146)

Cingulate

Frontal lobe

Deep GM

Occipital lobe

Parietal lobe

Temporal lobe

Global GM

L: R: L: R: L: R: L: R:

L: R: L: R: L: R: L: R:

L: R: L: R: L: R: L: R:

L: 0.174 R: 0.111 L: 0.077 R: ⫺0.004 L: 0.360 R: 0.412 L: 0.260 R: 0.231

L: R: L: R: L: R: L: R:

L: 0.135 R: 0.150 L: 0.013 R: ⫺0.066 L: 0.297 R: 0.316 L: 0.208 R: 0.178

0.165

0.206 0.196 0.003 0.055 0.518 0.533* 0.282 0.300*

0.107 0.127 0.016 0.027 0.483 0.487 0.234 0.244

0.219 0.237* 0.212* 0.092 0.328 0.268 0.270 0.244

0.058 0.058 0.056 0.063 0.398 0.427 0.226 0.238

0.058 0.501 0.280

Note. Bold, P ⬍ 0.05. * Strongest correlation for every group; L, R: left, right side. Note that the significance of r values depends on group sizes and that therefore only the results from the subgroups are directly comparable with each other.

Voxel-based study Two clusters in the anterior cingulate showed a significant positive correlation with the compacted IQ (Fig. 4a). No other clusters were detected anywhere in the brain (also indicating that the variance in the data was adequately explained by the covariates). One small cluster in a high-posterior parietal region correlated negatively with cIQ. After correction for multiple comparisons (based on the whole brain volume), neither of these results reached significance. This, however, is not a conditio sine qua non in the case of expected results, and our volumetric data had already pointed to the cingulate as the brain region correlated most with IQ, in line with a large number of previous studies (see Discussion of results below). To ascertain that these results were not detected simply by chance, a small volume correction was done, assessing the significance of these results if expected within the search volume (the complete cingulate). This analysis showed that the results were significant after correcting for multiple comparisons (P ⫽ 0.04). Additionally, the extracted raw data for all detected clusters were analyzed in order to verify that they correlate with IQ and are not a result of the parameterization of the IQ data. This could be confirmed in the case of the clusters in the anterior cingulate, the extracted raw data of which correlated highly with the original IQ values (r ⫽ 0.35, P ⫽ 0.00001). The raw data from the high-parietal cluster, which was found to have a negative correlation with the compacted IQ in the voxel-based study, showed no correlation with IQ. We therefore do not consider this finding significant in this analysis. Connectivity analysis This analysis showed that large parts of the surrounding anterior cingulate were correlated with the initial clusters. Clusters were also detected bilaterally in the thalamus, the medial temporal lobe, and the posterior temporofrontoparietal junction area (Figs. 4b and 5). Their extracted raw data

again correlated highly with individual IQ (r ⫽ 0.26, P ⫽ 0.001). In the inverse correlation, several clusters in the parietal lobe were found to be negatively correlated with the clusters from the anterior cingulate, indicating an inverse volumetric relationship between these areas (Figs. 4b and 5). Again, the correlation between the raw data and IQ was significant (r ⫽ 0.20, P ⫽ 0.01). Discussion Subjects and methods Our sample consisted of a large group of carefully selected healthy and normal children with a wide range of cognitive abilities. Regarding age and IQ, we find an overall nicely normally distributed pattern while still covering sufficiently the extreme ends of the spectrum. We therefore think that our data make up a representative sample of normal children, large enough to account for the high variability in brain structure measures (Lange et al., 1997) and allowing for the generalization of our findings to the population as a whole (Rivkin, 2000). At this point, the necessity of studying adequate group sizes should be stressed: the analysis of a smaller group may not have had enough power to detect the patterns we have been able to show in this sample. Regarding our methodology, we were able to use reference data for spatial normalization and segmentation based on our own pediatric data as described before (Wilke et al., 2002, 2003). As we were thus able to avoid the problem of processing our data based on inappropriate reference data, an increased sensitivity and specificity should result. Also, the application of an optimized processing protocol (Good et al., 2001) should improve not only sensitivity by ensuring optimal segmentation, but also specificity by using cleaned gray matter partitions, which minimizes the contribution of non-brain voxels. Finally, applying a correction for the regional effects of nonlinear normalization allowed for the

M. Wilke et al. / NeuroImage 20 (2003) 202–215

209

Fig. 4. a. Voxel-based morphometry: detection of gray matter volume significantly correlated with cIQ; P ⫽ 0.0001, uncorrected, extent threshold ⫽ 25 voxels, overlaid on the averaged gray matter partition. b. Connectivity analysis: detection of gray matter volume significantly correlated with the region detected in a; P ⫽ 0.01, corrected, extent threshold ⫽ 25 voxels; red, positive correlations; blue, negative correlations; note different slice thickness (a, 3 mm; b, 10 mm) and scaling (according to color bars) in a and b. Fig. 5. Connectivity analysis: rendering of results on the corresponding averaged brain surface; P ⫽ 0.01, corrected, extent threshold ⫽ 25 voxels; red, positive correlations; green, negative correlations. See also Fig. 4b.

examination of real gray matter volume changes (as opposed to “gray matter density”) while at the same time ensuring adequate overlap of corresponding structures.

We decided to use the age-corrected, full-scale IQ data in our analyses, which is an excellent predictor of academic achievement (Lezak, 1995) and allows for a direct compar-

210

M. Wilke et al. / NeuroImage 20 (2003) 202–215

ison of the overall cognitive abilities of a 5-year-old and an 18-year-old, and thus, gray matter volumes putatively correlated with this measure. Although the Wechsler test also provides verbal and performance IQ scores, these fail to reflect distinct neuropsychological features and show a high degree of overlap (Lezak, 1995). Additionally, they are highly correlated with each other (r ⫽ 0.58) and with the full-scale IQ (performance and full-scale: r ⫽ 0.87; verbal and full-scale: r ⫽ 0.90, P ⬍ .000001 for all correlations in our sample), making an assessment of possibly different effects on brain gray matter virtually impossible. Volumetric study Effects of age The results from the volumetric study, examining the relation between gray matter volume and age, are very well in line with previous cross-sectional studies on normal brain development (Courchesne et al., 2000; Pfefferbaum et al., 1994) in that a strong decrease of gray matter volume occurs in childhood. The larger number of subjects studied here provided more statistical power to assess the exact nature of this relationship, both globally and regionally. The developmental gray matter volume decline has been described to be regionally specific (Giedd et al., 1999), and regional variability can also be seen in our sample (Fig. 2). The fact that all curves seem to follow the same trend would be compatible with a global process that is regionally modified in that it is either enhanced or attenuated, regulated, and determined by a hitherto unknown factor (or a combination of factors). Possible implications of these regional variations are discussed below. Effects of IQ The correlation between IQ and global gray matter volume is very strong and consistent with earlier findings (Andreasen et al., 1993; Plomin and Kosslyn, 2001; Reiss et al., 1996; Rushton and Ankney, 1995). The individual correlations of regional gray matter volume and IQ are all very similar to each other and, consequently, to the correlation of global gray matter and IQ. This pattern of global rather than local correlations has been observed before (Andreasen et al., 1993; Reiss et al., 1996; Pennington et al., 2000; for a discussion of the volumetric variability and its interrelation with age, see below). For example, Reiss et al. found that only their measure of prefrontal gray matter correlated (slightly) more strongly with IQ than global gray matter, consistent with our finding of the strongest correlation between IQ and gray matter volume in the cingulate. Our results also confirm other findings from the Reiss et al. study: in their population, IQ explained about 15% of the variance in gray matter, similar to the 9% we found in this study. The correlation between gray matter volume and IQ was best described by a third-order polynomial function (Fig. 3), while a second-order polynomial function was used by Reiss et al. (1996). The difference is that a second-order

function allows the curve to descend toward higher values: in our data, the correlation accelerates toward higher IQ values. This might be explained by the fact that we have 20 subjects in the very high IQ range of ⬎130 (as opposed to 7 in the Reiss et al. study), allowing for a more accurate estimation. Although such a nonlinear acceleration has been suggested before (Gur et al., 1999), dedicated samples of very high IQ individuals might be necessary to ascertain that the relationship becomes more steep in the high IQ range. It should be noted that the necessity of smoothing the data prior to doing VBM studies (see Subjects and Methods) precludes a one-to-one correspondence with the volumetric study (done on not-smoothed data). However, the differences should be small. Voxel-based study As stated before, “simple correlations would probably show that all brain regions correlate with g [general intelligence], not just the frontal region” (Plomin and Kosslyn, 2001). The accuracy of this prediction is underscored by the profound and regionally almost indistinguishable correlation of IQ and global gray matter found in our volumetric study, underlining the necessity of accurately controlling for global effects. It should be noted that global gray matter (unlike in the volumetric study) was used as a covariate of no interest. Additionally, we used a parameterized, compacted IQ score that was not correlated with global gray matter volume anymore and thus allowed for the independent assessment of both global effects (as an effect of no interest) and local effects. Therefore, global tissue effects do not explain the results from this approach. In this voxel-based study, assessing the positive correlation of local gray matter volume and our compacted IQ scale, two small clusters in the anterior cingulate were found to be significant (Fig. 4a). The anterior cingulate has been implicated over and over in cognitive processes, including divided attention, novelty detection, working memory, memory retrieval, Stroop tasks, evaluative judgment, motivation, and performance monitoring (Cabeza and Nyberg, 2000; Casey et al., 1997; Dehaene et al., 1998; Gehring and Fencsik, 2001; Swick and Jovanovic, 2002; Zysset et al., 2002). It is considered an initiating and/or an inhibiting region, in that it initiates appropriate or suppresses inappropriate behavior (Cabeza and Nyberg, 2000; Paus et al., 1993; Swick and Jovanovic, 2002). These kind of functions would make it assume an active role in approaching and solving many of the tasks in the standard tests of human intelligence as used in this study. In fact, a unique position in the interplay among emotion, cognition, and motor control has been attributed to the anterior cingulate (Allman et al., 2001; Paus, 2001), supported by the presence of specialized projection neurons (so-called spindle cells) only in this structure and only in the brains of humans and certain great apes (Nimchinsky et al., 1999). It therefore seems likely that we are able to detect this structure as the “tip of

M. Wilke et al. / NeuroImage 20 (2003) 202–215

the iceberg” due to its important role for many of the tasks contributing to the final assessment of human intelligence, as reflected in the full-scale IQ. To conclude, we can show a direct and positive correlation between gray matter volume in a distinct area of the brain (not explained by global effects of gray matter volume) that is known to be involved in a multitude of cognitive tasks and a measure of human intelligence. Connectivity analysis As laid out above, IQ had to be reparameterized in order to decorrelate it from global gray matter volume. To overcome the limitation of such a rough classification (our compacted IQ scale) and to allow for the detection of related areas, we employed a connectivity analysis. It must be borne in mind, though, that inferences about the relationship between the areas detected in this analysis and intelligence can only be indirect, as opposed to the direct implication of the anterior cingulate as detected in the first voxel-based analysis. Connectivity analyses have been applied to structural (Bullmore et al., 1998), functional (Bu¨ chel and Friston, 2000), and combined MRI data sets (Sporns et al., 2000b), including the anterior cingulate (Koski and Paus, 2000). In the context of our study, this analysis should detect brain areas following volume patterns very close to the one in the original, seeding-point region (the anterior cingulate). A negative relation should point to areas where a volume increase in the marker region is accompanied by decreases in this area (K.J. Friston, personal communication). Although such a (positive or negative) relation cannot be taken to automatically reflect a causal or immediate interdependency, the strong correlation detected here and the findings of synchronization playing an important role in neural networks (Arshavsky, 2001; Crick and Koch, 1998; Gonza´ lezHerna´ ndez et al., 2002) strongly support a close “working relationship.” Distributed networks likely play a substantial role in the sustentation of the neural substrates of cognitive functions (Dehaene et al., 1998; Koski and Paus, 2000; McIntosh, 2000; Sporns et al., 2000a), although the role of the network versus the role of single cell contributions remains a subject of debate (Arshavsky, 2001). Therefore, the application of such connectivity analyses may be especially helpful, if not the only way, to fully assess these foundations. This analysis, not surprisingly and verifying the validity of the approach, showed a large portion of the anterior cingulate and adjacent gray matter structures to be correlated with the originally detected, nearby clusters. Furthermore, bilateral clusters were detected in the thalamus, in the medial temporal lobes, and in the posterior frontotemporoparietal junction, the latter with a slight right predominance. Interestingly, several regions in the high parietal lobe showed an inverse correlation with the IQ-correlated anterior cingulate, indicating an antipodal growth pattern (Figs. 4b and 5).

211

The regions that were closely linked with the anterior cingulate have all been implicated in cognitive functions before: the medial temporal lobe, due to its role in topographical memory and memory retrieval (Maguire, 1997; Ryan et al., 2001); the mediodorsal aspects of the thalamus (the largest association nuclei of the thalamus) have been implicated in cognitive defects in schizophrenia (Byne et al., 2002) and in lesion studies (Karussis et al., 2000; Leathem and Martin, 2001); the role of the anterior cingulate has been discussed above. These findings may also have implications for diseases with an impairment of cognitive functions: for example, both the anterior cingulate (Sanders et al., 2002) and the thalamus (Byne et al., 2002; Kim et al., 2000) have been implicated in the pathophysiology of cognitive deficits in schizophrenia, with structural abnormalities found in both regions (Byne et al., 2002; Wilke et al., 2001). The fact that these regions were indirectly correlated with intelligence in the present study lends support to the hypothesis that such alterations may underlie the cognitive deficits in patients with schizophrenia. We also detected bilateral clusters in the posterior frontal/ superior temporal/anterior parietal regions as part of a network associated with cognitive abilities. Interestingly, very similar regions have been implicated in the cognitive processes involved in the Wisconsin Card Sorting test (Gonza´ lez-Herna´ ndez et al., 2002). In this detailed study of electrical brain activity, fast oscillations (related to information processing) have been detected in these regions, with hemispheric specificity for certain types of oscillations. These areas may also occupy an exceptional position in this cognition-related network in that they are among the very few regions that show a positive correlation of gray matter volume with age and continue to grow in early adulthood (Sowell et al., 2001). In fact, this pattern of volume increase was also evident in our sample (data not shown), supporting a role of this structure in aspects of cognition that continue to evolve in and beyond adolescence. The importance of this region is also underlined by its close spatial proximity to white matter regions showing IQcorrelated changes in MR diffusion images (Schmithorst et al., 2002), confirming earlier 1H-MR spectroscopy findings (Jung et al., 1999). Both studies are compatible with a progressive elimination of some and strengthening of remaining white matter fibers with higher individual IQ scores. This is especially intriguing due to the interposition of these white matter findings between regions positively and negatively correlated with cognitive performance (frontotemporoparietal interface and high parietal lobes, respectively). However, these suggested relations will need to be tested more directly in future studies. Implications of our findings IQ by age interactions It is interesting to notice that the two regions detected to have a strongly positive (cingulate) or negative (parietal lobe) correlation with IQ in both voxel-based analyses also

212

M. Wilke et al. / NeuroImage 20 (2003) 202–215

show the most aberrant volume change with age. The cingulate shows a more shallow course of gray matter reduction, whereas the parietal lobe displays a stronger-thanaverage volume loss with age, a finding consistent with earlier studies on a smaller dataset (Sowell et al., 1999, 2002). We think this observation ties in with the fact that the gray matter volume/IQ correlations were dominated by the oldest group of children: regions strongly correlated with IQ show a lower-than-average volume loss (cingulate), while regions negatively correlated with IQ show a stronger-thanaverage decline in gray matter volume (parietal lobe; Table 4, Figs. 4 and 5). An explanation for this would be that the well-established process of gray matter pruning (Casey et al., 2000) selectively eliminates gray matter not effectively contributing to cognitive processes. This process would also explain the lack of correlation between IQ and gray matter volume in the younger age groups: the relation between contributing and noncontributing gray matter is still unfavorable, as also evident by a larger variance of the data in the younger age groups (data not shown). If, by reaching the age range covered by the older group, noncontributing gray matter has been eliminated more effectively, a correlation with a parameter of cognitive function is much more likely to be detected. This is consistent with what we find in our data. Correlation of gray matter volume and cognitive abilities A surprisingly large portion of the correlation between intelligence and global gray matter volume has been estimated to be of genetic origin (Posthuma et al., 2002; Thompson et al., 2001). It seems interesting to apply the results from these studies, done in adults, to our findings. Taken together, it would seem that not only the final volume, but also the process leading to this volume, i.e., the (slightly region-specific) systematic reduction of gray matter, is subject to a tight genetic control. This is especially interesting since, therefore, both a negative (in the case of the “pruned” gray matter) and a positive correlation of gray matter volume with IQ (in the case of the remaining gray matter) can be postulated to be regulated by genetic influences. “Frontalization” of cognitive functions The relation between the positively (frontal) and negatively (parietal) correlated areas is interesting insofar as that this means that, opposite to the overall trend, a higher level of cognitive abilities seems to be associated with lower gray matter volumes in posterior areas, in accordance with the negative correlation of IQ and gray matter postulated above. This could be indicative of a posterior-to-anterior shift as a function of intelligence, suggesting that higher IQ individuals rely more on frontal regions than their lower IQ counterparts. A number of recent findings in completely unrelated populations would support this notion: in individuals with

Down’s syndrome, only parietal gray matter was found not to be reduced in volume when compared to healthy controls (Pinter et al., 2001). This was taken to explain the relative sparing of visuospatial abilities in this population but could also be interpreted as a lack of shifting cognitive processes toward frontal regions as hypothesized here, thus not subjecting the parietal gray matter to a consecutive volume loss. Also, an fMRI study of cognitively impaired patients with multiple sclerosis (MS) indicates that, while control subjects activated the anterior cingulate, patients with MS showed activation in left-parietal regions instead (Staffen et al., 2002). This was interpreted as a compensatory mechanism and could be seen as “falling back to an alternative/ earlier pattern” in the light of our data. Finally, in a recent fMRI study, we found that the individual cognitive abilities in healthy controls determined activation in the anterior cingulate during a visuospatial rotation task (Sohn et al., 2002). It seems noteworthy that this process may be similar to the functional and anatomical frontal maturation seen in children as a function of age (Casey et al., 2000; Moses et al., 2002; Schlaggar et al., 2002). However, age was controlled for in our analysis. It should thus be considered a separate process which takes place in childhood but is influenced not necessarily by age itself but rather by the unfolding “inherent intellectual capacity” of a child’s brain. It could even be argued that there is only one process, the detection of which in childhood is attributable to the profound improvements of intellectual capacity in children. At present, not enough data is available to definitively resolve this possible interdependency regarding the presence of one or two processes (modulated by age and/or intelligence). We therefore put forward the testable hypothesis that a posterior-to-anterior shift of the neural substrates of cognitive abilities occurs not (only) as a function of normal development in children, but may rather be a more general process depending on the individual level of cognitive abilities. Possible limitations of this study As mentioned above and as is apparent from Table 2, our sample is slightly weighted toward the median age group. This leads to our three age groups covering different age ranges, with the medium group only covering about 39 months (as opposed to 54 and 73 months in the young and old group, respectively). However, since the effect of age was not the primary focus of this study, we do not see this as a major confound. Our approach to studying volumetric effects has the advantage of being fast, accurate, and reproducible, but (in a given single case) may yield less accurate results than a careful manual delineation of the structures of interest. However, our large sample size practically prohibits such labor-intensive procedures, effectively making an automated procedure of some sort the only alternative in order to

M. Wilke et al. / NeuroImage 20 (2003) 202–215

obtain reliable and robust data. Also, due to considerations of nonoptimal coverage in some subjects and possible effects of field inhomogeneities, volumetric measures of the cerebellum could not be included in the current analysis. In the voxel-based study, only relatively small clusters were detected, as opposed to the connectivity analysis. In our opinion, this relative inability of the initial study to detect these regions is due to several factors; for one, the strong correlation between IQ and global gray matter might play a role, because it influences the results in two ways: first, it necessitated the reparameterization of IQ into three classes to decorrelate these variables. We think it is mainly the fact that the raw data from the global voxel-based analysis represents a much more individual and thus much more sensitive volumetric measure (correlated with individual IQ) that allowed the uncovering of the large-scale correlations found in the connectivity analysis. Secondly, in order to make sure that global confounds did not influence our results, global gray matter was modeled as a covariate of no interest. Global effects were additionally modeled by including age as a further nuisance variable, possibly leading to an overcorrection of this effect in the first analysis (where an individual measure of gray matter, but only a group measure of intelligence was used). It seems that by using an individual measure in the connectivity analysis, this effect was counterbalanced.

Acknowledgments We thank Christian Bu¨ chel (Cognitive Neuroscience Laboratories, University of Hamburg), Karl Friston (Functional Imaging Laboratory, University College London), and Thomas Nichols (Department of Biostatistics, University of Michigan) for helpful advice and discussions. This work was funded in part by a grant from the National Institutes of Child Health and Human Development, RO1HD38578-01.

References Allman, J.M., Hakeem, A., Erwin, J.M., Nimchinsky, E., Hof, P., 2001. The anterior cingulate cortex. The evolution of an interface between emotion and cognition. Ann. N. Y. Acad. Sci. 935, 107–117. Amunts, K., Schlaug, G., Schleicher, A., Steinmetz, H., Dabringhaus, A., Roland, P.E., Zilles, K., 1996. Asymmetry in the human motor cortex and handedness. NeuroImage 4, 216 –222. Andreasen, N.C., Flaum, M., Swayze 2nd, V., O’Leary, D.S., Alliger, R., Cohen, G., Ehrhardt, J., Yuh, W.T., 1993. Intelligence and brain structure in normal individuals. Am. J. Psychiatry 150, 130 –134. Arshavsky, Y.I., 2001. Role of individual neurons and neural networks in cognitive functioning of the brain: a new insight. Brain Cogn. 46, 414 – 428. Ashburner, J., Friston, K.J., 1997. Multimodal image coregistration and partitioning—a unified framework. NeuroImage 6, 209 –217. Ashburner, J., Friston, K.J., 1999. Nonlinear spatial normalization using basis functions. Hum. Brain Mapp. 7, 254 –266.

213

Ashburner, J., Friston, K.J., 2000. Voxel-based morphometry—the methods. NeuroImage 11, 805– 821. Bu¨ chel, C., Friston, K., 2000. Assessing interactions among neuronal systems using functional neuroimaging. Neural Netw. 13, 871– 882. Bullmore, E.T., Woodruff, P.W., Wright, I.C., Rabe-Hesketh, S., Howard, R.J., Shuriquie, N., Murray, R.M., 1998. Does dysplasia cause anatomical dysconnectivity in schizophrenia. Schizophr. Res. 30, 127–135. Byne, W., Buchsbaum, M.S., Mattiace, L.A., Hazlett, E.A., Kemether, E., Elhakem, S.L., Purohit, D.P., Haroutunian, V., Jones, L., 2002. Postmortem assessment of thalamic nuclear volumes in subjects with schizophrenia. Am. J. Psychiatry 159, 59 – 65. Cabeza, R., Nyberg, L., 2000. Imaging cognition II: an empirical review of 275 PET and fMRI studies. J. Cogn. Neurosci. 12, 1– 47. Carrow-Woolfolk, E., 1996. Oral and Written Language Scales: Written Expression Scale Manual. American Guidance Service, Circle Pines, MN. Casey, B.J., Giedd, J.N., Thomas, K.M., 2000. Structural and functional brain development and its relation to cognitive development. Biol. Psychol. 54, 241–257. Casey, B.J., Trainor, R., Giedd, J., Vauss, Y., Vaituzis, C.K., Hamburger, S., Kozuch, P., Rapoport, J.L., 1997. The role of the anterior cingulate in automatic and controlled processes: a developmental neuroanatomical study. Dev. Psychobiol. 30, 61– 69. Courchesne, E., Chisum, H.J., Townsend, J., Cowles, A., Covington, J., Egaas, B., Harwood, M., Hinds, S., Press, G.A., 2000. Normal brain development and aging: quantitative analysis at in vivo MR imaging in healthy volunteers. Radiology 216, 672– 682. Crick, F., Koch, C., 1998. Consciousness and neuroscience. Cereb. Cortex 8, 97–107. Deary, I.J., Caryl, P.G., 1997. Neuroscience and human intelligence differences. Trends Neurosci. 20, 365–371. DeBellis, M.D., Keshavan, M.S., Beers, S.R., Hall, J., Frustaci, K., Masalehdan, A., Noll, J., Boring, A.M., 2001. Sex differences in brain maturation during childhood and adolescence. Cereb. Cortex 11, 552– 557. Dehaene, S., Kerszberg, M., Changeux, J.P., 1998. A neuronal model of a global workspace in effortful cognitive tasks. Proc. Natl. Acad. Sci. USA 95, 14529 –14534. Friston, K.J., Holmes, A.P., Worsley, K.J., Poline, J.P., Frith, C.D., Frackowiak, R.S.J., 1995. Statistical parametric maps in functional imaging: a general linear approach. Hum. Brain Mapp. 2, 189 –210. Gehring, W.J., Fencsik, D.E., 2001. Functions of the medial frontal cortex in the processing of conflict and errors. J. Neurosci. 21, 9430 –9437. Genovese, C.R., Lazar, N.A., Nichols, T., 2002. Thresholding of statistical maps in functional neuroimaging using the false discovery rate. NeuroImage 15, 870 – 878. Giedd, J.N., Blumenthal, J., Jeffries, N.O., Castellanos, F.X., Liu, H., Zijdenbos, A., Paus, T., Evans, A.C., Rapoport, J.L., 1999. Brain development during childhood and adolescence: a longitudinal MRI study. Nat. Neurosci. 2, 861– 863. Gonza´ lez-Herna´ ndez, J.A., Alcorta, C.P., Ceden˜ o, I., Bosch-Bayard, J., Gala´ n-Garcia, L., Scherbaum, W.A., Figueredo-Rodriguez, P., 2002. Wisconsin card sorting test synchronizes the prefrontal, temporal and posterior association cortex in different frequency ranges and extensions. Hum. Brain. Mapp. 17, 37– 47. Good, C.D., Johnsrude, I., Ashburner, J., Henson, R.N., Friston, K.J., Frackowiak, R.S., 2001. Cerebral asymmetry and the effects of sex and handedness on brain structure: a voxel-based morphometric analysis of 465 normal adult human brains. NeuroImage 14, 685–700. Gur, R.C., Turetsky, B.I., Matsui, M., Yan, M., Bilker, W., Hughett, P., Gur, R.E., 1999. Sex differences in brain gray and white matter in healthy young adults: correlations with cognitive performance. J. Neurosci. 19, 4065– 4072. Holland, S.K., Plante, E., Weber Byars, A., Strawsburg, R.H., Schmithorst, V.J., Ball Jr., W.S., 2001. Normal fMRI brain activation patterns in children performing a verb generation task. NeuroImage 14, 837– 843.

214

M. Wilke et al. / NeuroImage 20 (2003) 202–215

Jung, R.E., Brooks, W.M., Yeo, R.A., Chiulli, S.J., Weers, D.C., Sibbitt Jr., W.L., 1999. Biochemical markers of intelligence: a proton MR spectroscopy study of normal human brain. Proc. R. Soc. Lond. B. Biol. Sci. 266, 1375–1379. Karussis, D., Leker, R.R., Abramsky, O., 2000. Cognitive dysfunction following thalamic stroke: a study of 16 cases and review of the literature. J. Neurol. Sci. 172, 25–29. Kim, J.J., Mohamed, S., Andreasen, N.C., O’Leary, D.S., Watkins, G.L., Boles Ponto, L.L., Hichwa, R.D., 2000. Regional neural dysfunctions in chronic schizophrenia studied with positron emission tomography. Am. J. Psychiatry 157, 542–548. Koski, L., Paus, T., 2000. Functional connectivity of the anterior cingulate cortex within the human frontal lobe: a brain-mapping meta-analysis. Exp. Brain Res. 133, 55– 65. Lange, N., Giedd, J.N., Castellanos, F.X., Vaituzis, A.C., Rapoport, J.L., 1997. Variability of human brain structure size: ages 4 –20 years. Psychiatry Res. 74, 1–12. Leathem, J.M., Martin, G.T., 2001. Cognitive deficits due to asymmetrical bilateral thalamic lesions. Brain Inj. 15, 455– 462. Lezak, M.D., 1995. Batteries for Assessment of Brain Damage, in: Lezak, M.O. (Ed.), Neuropsychological Assessment, Third ed., Oxford Univ. Press, New York, pp. 686 –735. Maguire, E.A., 1997. Hippocampal involvement in human topographical memory: evidence from functional imaging. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 352, 1475–1480. McIntosh, A.R., 2000. Towards a network theory of cognition. Neural Netw. 13, 861– 870. Moses, P., Roe, K., Buxton, R.B., Wong, E.C., Frank, L.R., Stiles, J., 2002. Functional MRI of global and local processing in children. NeuroImage 16, 415– 424. Nimchinsky, E.A., Gilissen, E., Allman, J.M., Perl, D.P., Erwin, J.M., Hof, P.R., 1999. A neuronal morphologic type unique to humans and great apes. Proc. Natl. Acad. Sci. USA 96, 5268 –5273. Paus, T., Petrides, M., Evans, A.C., Meyer, E., 1993. Role of the human anterior cingulate cortex in the control of oculomotor, manual, and speech responses: a positron emission tomography study. J. Neurophysiol. 70, 453– 469. Paus, T., 2001. Primate anterior cingulate cortex: where motor control, drive and cognition interface. Nat. Rev. Neurosci. 2, 417– 424. Pennington, B.F., Filipek, P.A., Lefly, D., Chhabildas, N., Kennedy, D.N., Simon, J.H., Filley, C.M., Galaburda, A., DeFries, J.C., 2000. A twin MRI study of size variations in human brain. J. Cogn. Neurosci. 12, 223–232. Peters, M., 1995. Does brain size matter? A reply to Rushton & Ankney. Can. J. Exp. Psychol. 49, 577–583. Peterson, B.S., Vohr, B., Staib, L.H., Cannistraci, C.J., Dolberg, A., Schneider, K.C., Katz, K.H., Westerveld, M., Sparrow, S., Anderson, A.W., Duncan, C.C., Makuch, R.W., Gore, J.C., Ment, L.R., 2000. Regional brain volume abnormalities and long-term cognitive outcome in preterm infants. J. Am. Med. Assoc. 284, 1939 –1947. Pfefferbaum, A., Mathalon, D.H., Sullivan, E.V., Rawles, J.M., Zipursky, R.B., Lim, K.O., 1994. A quantitative magnetic resonance imaging study of changes in brain morphology from infancy to late adulthood. Arch.Neurol. 51, 874 – 887. Pinter, J.D., Eliez, S., Schmitt, J.E., Capone, G.T., Reiss, A.L., 2001. Neuroanatomy of Down’s syndrome: a high-resolution MRI study. Am. J. Psychiatry 158, 1659 –1665. Plomin, R., Kosslyn, S.M., 2001. Genes, brain and cognition. Nat. Neurosci. 4, 1153–1154. Posthuma, D., De Geus, E.J., Baare, W.F., Hulshoff Pol, H.E., Kahn, R.S., Boomsma, D.I., 2002. The association between brain volume and intelligence is of genetic origin. Nat. Neurosci. 5, 83– 84. Reiss, A.L., Abrams, M.T., Singer, H.S., Ross, J.L., Denckla, M.B., 1996. Brain development, gender and IQ in children. A volumetric imaging study. Brain 119, 1763–1774.

Rivkin, M.J., 2000. Developmental neuroimaging of children using magnetic resonance techniques. Ment. Retard. Dev. Disabil. Res. Rev. 6, 68 – 80. Rushton, J.P., Ankney, C.D., 1995. Brain size matters: a reply to Peters. Can. J. Exp. Psychol. 49, 562–569. Ryan, L., Nadel, L., Keil, K., Putnam, K., Schnyer, D., Trouard, T., Moscovitch, M., 2001. Hippocampal complex and retrieval of recent and very remote autobiographical memories: evidence from functional magnetic resonance imaging in neurologically intact people. Hippocampus 11, 707–714. Sanders, G.S., Gallup, G.G., Heinsen, H., Hof, P.R., Schmitz, C., 2002. Cognitive deficits, schizophrenia, and the anterior cingulate cortex. Trends Cogn. Sci. 6, 190 –192. Schlaggar, B.L., Brown, T.T., Lugar, H.M., Visscher, K.M., Miezin, F.M., Petersen, S.E., 2002. Functional neuroanatomical differences between adults and school-age children in the processing of single words. Science 296, 1476 –1479. Schmithorst, V.J., Wilke, M., Dardzinski, B.J., Holland, S.K., 2002. IQ Correlates with diffusion tensor imaging parameters in normal children. Proc. Intl. Soc. Mag. Reson. Med. 10, 427. Sohn, J.H., Kim, Y.H., Lee, K.H., Park, S.H., Lee, J.M., Chung, S.C., 2002. An fMRI study to identify the brain structure for predicting the level of individual’s intelligence. NeuroImage 16 (Pt. 2), 1024. Sowell, E.R., Thompson, P.M., Holmes, C.J., Batth, R., Jernigan, T.L., Toga, A.W., 1999. Localizing age-related changes in brain structure between childhood and adolescence using statistical parametric mapping. NeuroImage 9, 587–597. Sowell, E.R., Thompson, P.M., Tessner, K.D., Toga, A.W., 2001. Mapping continued brain growth and gray matter density reduction in dorsal frontal cortex: inverse relationships during postadolescent brain maturation. J. Neurosci. 21, 8819 – 8829. Sowell, E.R., Trauner, D.A., Gamst, A., Jernigan, T.L., 2002. Development of cortical and subcortical brain structures in childhood and adolescence: a structural MRI study. Dev. Med. Child. Neurol. 44, 4 –16. Sporns, O., Tononi, G., Edelman, G.M., 2000a. Connectivity and complexity: the relationship between neuroanatomy and brain dynamics. Neural Netw. 13, 909 –922. Sporns, O., Tononi, G., Edelman, G.M., 2000b. Theoretical neuroanatomy: relating anatomical and functional connectivity in graphs and cortical connection matrices. Cereb. Cortex 10, 127–141. Staffen, W., Mair, A., Zauner, H., Unterrainer, J., Niederhofer, H., Kutzelnigg, A., Ritter, S., Golaszewski, S., Iglseder, B., Ladurner, G., 2002. Cognitive function and fMRI in patients with multiple sclerosis: evidence for compensatory cortical activation during an attention task. Brain 125, 1275–1282. Swick, D., Jovanovic, J., 2002. Anterior cingulate cortex and the Stroop task: neuro-psychological evidence for topographic specificity. Neuropsychologia 40, 1240 –1253. Thompson, P.M., Cannon, T.D., Narr, K.L., van Erp, T., Poutanen, V.P., Huttunen, M., Lonnqvist, J., Standertskjold-Nordenstam, C.G., Kaprio, J., Khaledy, M., Dail, R., Zoumalan, C.I., Toga, A.W., 2001. Genetic influences on brain structure. Nat. Neurosci. 4, 1253–1258. Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N., Mazoyer, B., Joliot, M., 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage 15, 273–289. Ugurbil, K., Garwood, M., Ellermann, J., Hendrich, K., Hinke, R., Hu, X., Kim, S.G., Menon, R., Merkle, H., Ogawa, S., Salmi, R., 1993. Imaging at high magnetic fields: initial experiences at 4 T. Magn. Reson. Q. 9, 259 –277. Wechsler, D., 1989. Manual for the Wechsler Preschool and Primary Scale of Intelligence, rev.. The Psychological Corp., San Antonio, TX. Wechsler, D., 1991. Manual for the Wechsler Intelligence Scale for Children, 3rd ed. The Psychological Corp., San Antonio, TX.

M. Wilke et al. / NeuroImage 20 (2003) 202–215 Wechsler, D., 1997. Manual for the Wechsler Adult Intelligence Scale-III. The Psychological Corp., New York. White, T., O’Leary, D., Magnotta, V., Arndt, S., Flaum, M., Andreasen, N.C., 2001. Anatomic and functional variability: the effects of filter size in group fMRI data analysis. NeuroImage 13, 577–588. Wilke, M., Kaufmann, C., Grabner, A., Pu¨ tz, B., Wetter, T.C., Auer, D.P., 2001. Gray matter-changes and correlates of disease severity in schizophrenia: a statistical parametric mapping study. NeuroImage 13, 814 – 824.

215

Wilke, M., Schmithorst, V.J., Holland, S.K., 2002. Assessment of spatial normalization of whole-brain MR-images in children. Hum. Brain Mapp. 17, 48 – 60. Wilke M., Schmithorst V.J., Holland S.K., 2003. Normative pediatric brain data for spatial normalization and segmentation differs from standard adult data. Magn. Reson. Med., in press. Zysset, S., Huber, O., Ferstl, E., von Cramon, D.Y., 2002. The anterior frontomedian cortex and evaluative judgment: an fMRI study. NeuroImage 15, 983–991.