|Home | About | Journals | Submit | Contact Us | Français|
The 22q11.2 deletion syndrome (velocardiofacial/DiGeorge syndrome) is a neurogenetic condition associated with visuospatial deficits, as well as elevated rates of attentional disturbance, mood disorder, and psychosis. Previously, we detected pronounced cortical thinning in superior parietal and right parieto-occipital cortices in patients with this syndrome, regions critical for visuospatial processing. Here we applied cortical pattern-matching algorithms to structural magnetic resonance images obtained from 21 children with confirmed 22q11.2 deletions (ages 8–17) and 13 demographically matched comparison subjects, in order to map cortical thickness across the medial hemispheric surfaces. In addition, cortical models were remeshed in frequency space to compute their surface complexity. Cortical maps revealed a pattern of localized thinning in the ventromedial occipital–temporal cortex, critical for visuospatial representation, and the anterior cingulate, a key area for attentional control. However, children with 22q11.2DS showed significantly increased gyral complexity bilaterally in occipital cortex. Regional gray matter volumes, particularly in medial frontal cortex, were strongly correlated with both verbal and nonverbal cognitive functions. These findings suggest that aberrant parieto-occipital brain development, as evidenced by both increased complexity and cortical thinning in these regions, may be a neural substrate for the deficits in visuospatial and numerical understanding characteristic of this syndrome.
The 22q11.2 deletion syndrome (velocardiofacial/DiGeorge syndrome; 22q11.2DS) is a neurogenetic disorder resulting from a microdeletion on the long arm of chromosome 22 (McDonald-McGinn et al. 1997). With an estimated prevalence of 1/4000 live births, it represents one of the most common genetic causes of cognitive disability (Scambler 2000). Its physical manifestations frequently include cleft palate, hypocalcemia, cardiac defects, and facial dysmorphology (McDonald-McGinn et al. 1997). There is now overwhelming evidence for a characteristic cognitive phenotype as well, involving attentional deficits and marked impairment in visuospatial cognition, arithmetic (Moss et al. 1999; Swillen, Devriendt, et al. 1999; Bearden et al. 2001) and spatial attentional orienting (Simon, Bish, et al. 2005).
The syndrome is also associated with strikingly increased rates of psychopathology, particularly psychosis (Bassett and Chow 1999; Murphy et al. 1999; Bassett et al. 2000; Murphy 2002). Thirty to 40 genes are encoded in the deleted region, several of which are highly expressed in the brain during development, and known to affect early neuronal migration (Maynard et al. 2003). This syndrome therefore provides a unique window into gene-brain-behavior relationships.
Given that cortical malformations due to abnormal neuronal migration, such as periventricular nodular heterotopia involve nodules of gray matter located along the lateral ventricles due to a failure of particular neurons to migrate to their proper locations (Guerrini and Marini 2006), this would suggest that midline structures may be more affected by defects of neuronal migration. Indeed, previous studies using traditional volumetric approaches and voxel-based morphometry show that developmental midline anomalies are common in this syndrome. In particular, dysmorphology of the corpus callosum (Shashi et al. 2004; Antshel et al. 2005; Machado et al. 2007), increased incidence of cavum septum pellucidum (Chow et al. 1999, 2002; van Amelsvoort et al. 2001), and reduced volume of the cerebellar vermis and pons (Eliez et al. 2001; Bish et al. 2006) have been documented. With recent advances in neuroimaging analysis, it is now possible to measure gray matter thickness across the cortex with submillimeter accuracy (Narr et al. 2005; Thompson et al. 2005). In a prior study of children with this syndrome, we mapped for the first time the profile of deficit across the lateral cortical surfaces; highly significant thinning was observed in superior parietal cortices and right parieto-occipital cortex, regions critical for visuospatial processing, and bilaterally in the pars orbitalis, a key area for language development (Bearden et al. 2007). Here, we used computational methods that were able to map the pattern of cortical thickness deficits on the medial hemispheric surface, in the same cohort. We examined key midline regions that may be affected by early abnormalities in processes of neuronal migration and cortical organization in this syndrome (i.e., the cingulate and medial frontal cortex; Benes et al. 1991). Cortical surface anatomy was carefully matched across individuals to provide spatially refined localizations of group differences relative to gyral landmarks over the medial cerebral surface.
In addition to mapping midline cortical thickness, we also aimed to investigate gyrification patterns in children with 22q11.2DS. Gyral formation begins at about 16 weeks in utero, but most cortical convolutions are formed during the late second and third trimester of pregnancy (Armstrong et al. 1995). Disorders of neural crest development such as a chromosomal deletion at 22q11.2 may disturb gyrification (Rakic 1988b). Polymicrogyria (PMG), a brain malformation involving abnormally small gyri that do not follow the normal gyral pattern, is a rare manifestation of chromosome 22q11.2DS (Sztriha et al. 2004; Robin et al. 2006). To quantify changes in cortical folding patterns associated with this syndrome, we applied an algorithm we recently developed to measure gyral complexity in 3D (Narr et al. 2004; Thompson et al. 2005), using high-resolution cortical surface models extracted from each individual magnetic resonance imaging (MRI) scan. Thus, based on prior imaging and neurocognitive studies in this syndrome, we predicted that parieto-occipital brain regions might show relatively greater cortical thinning than frontal regions, concomitant with altered gyral complexity in these regions.
Finally, as an exploratory analysis, we sought to link these brain alterations with cognitive measures in these same subjects. To reduce the number of statistical comparisons, we focused on specific a priori comparisons, based on known functions of specific brain regions. In particular, as frontal and parietal regions are critically implicated in visuospatial, arithmetic and executive/attentional functions (Klingberg et al. 1997; Dehaene et al. 1999; Konrad et al. 2005), we predicted that cognitive abilities and behavior would be positively correlated with brain volumes in these regions.
Subject ascertainment procedures were identical to those reported in Bearden et al. (2004b). Briefly, 22q11.2DS participants were recruited through the Clinical Genetics Department at the Children's Hospital of Philadelphia (CHOP); genetic diagnosis of 22q11.2 microdeletion was confirmed by fluorescence in situ hybridization studies with the N25 (D2S75) molecular probe from Oncor (Gaithersburg, MD). The CHOP Institutional Review Board approved the study, and signed informed assents and consents were obtained from all subjects and their parents, respectively. A total of 22 22q11.2DS patients and 19 normal comparison subjects participated in the study; however, 7 subjects (1 22q11.2DS patient and 6 controls) had to be excluded from analysis due to motion artifact (N = 3) or use of different scanning acquisition parameters (N = 4). Thus, data were available from 21 subjects with 22q11.2DS and 13 control subjects (see Table 1). Normal comparison subjects were recruited through advertisement in the hospital community, and were group-matched with probands for race, handedness, parental social class, sex and age. A minimum IQ of 85 (1 SD below the population mean) and absence of previous neurologic or psychiatric disorder, and/or medical condition that might affect brain function, as determined by structured diagnostic interview with the subject and their parent, were used as inclusion criteria for controls.
At the time of the MRI scan, children with 22q11.2DS received neuropsychological testing. The 3-h neuropsychological test battery included measures of general intellectual function (Wechsler Scale for Children-III; WISC-III; Wechsler 1991), academic achievement (Wechsler Individual Achievement Tests; WIAT; Wechsler 1992), and memory and learning (subtests of the Wide Range Assessment of Memory and Learning; Sheslow and Adams 1990). Typically developing controls received IQ testing only (Wechsler 1991). A more complete description of the battery and test results is published in detail elsewhere (Bearden et al. 2001; Woodin et al. 2001).
Magnetic resonance images of each subject's brain were acquired using a 1.5 Tesla Siemens scanner in the Department of Radiology at CHOP. An average of 160 1-mm-thick sagittal slices were acquired using a 3D magnetization prepared rapid gradient echo (MP-RAGE) sequence, with a repetition time of 9.7 ms, an echo time of 4.0 ms, and inversion time of 300, a flip angle of 8°, and no interslice gap. The matrix size was 256 × 256 pixels, with a field of view of 25.6 × 25.6 cm and an in-plane resolution of 1.0 × 1.0 mm. Scans were analyzed at the UCLA Laboratory of Neuro Imaging (LONI) by image analysts blinded to all subject information, including age, sex, and diagnosis. All MR images were processed with a series of manual and automated procedures detailed in other reports (Thompson et al. 2003, 2004), and summarized below.
Image volumes passed through a number of preprocessing steps implemented in the LONI Pipeline Processing Environment (Thompson et al. 2004), including removal of all extra-cerebral tissue, creation of an intracranial mask of the brain using a brain surface extraction algorithm tool (Shattuck and Leahy 2002), and correction for head alignment and individual differences in brain size by using an automatic 9-parameter registration (Collins et al. 1994) to transform each brain volume into alignment with the target space and dimensions of the ICBM-305 reference brain created by the International Consortium for Brain Mapping (Mazziotta et al. 1995). After applying radiofrequency bias field corrections to eliminate intensity drifts due to magnetic field inhomogeneities, each image volume was segmented into different tissue types by classifying voxels based on their signal intensity values (Shattuck and Leahy 2002). Tissue classified brain volumes were resampled to 0.33 mm cubic voxels to improve the spatial resolution and precision of subsequent thickness measurements.
Cortical pattern-matching methods were used to localize disease effects on cortical anatomy, and increase the power to detect group differences (Thompson et al. 2003, 2004). The approach models, and controls for, gyral pattern variations across subjects. Briefly, we created 3D cortical surface models based on automatically generated spherical mesh surfaces that were continuously deformed to fit a threshold intensity value that best differentiates extra-cortical cerebrospinal fluid from underlying cortical gray matter (MacDonald et al. 1994). The generated 3D cortical surface models correspond globally in size and orientation, due to the linear transformation procedure. Nevertheless, the same parameter space coordinates, within each cortical surface model, do not yet index the same anatomy across all subjects and across hemispheres. Therefore, the cortical surface models from each individual were used to identify and manually outline 13 sulci in each hemisphere, where image analysts were blind to subject demographics and diagnosis. Spatially registered gray-scale image volumes in coronal, axial, and sagittal planes were available simultaneously to help disambiguate brain anatomy. Landmarks were defined according to a detailed anatomical protocol (Sowell et al. 2001) based on the Ono sulcal atlas (Ono et al. 1990). Interrater reliability estimates demonstrated < 2 mm root mean square difference in the matched 3D locations of sulcal landmarks traced on 6 test brains, compared with a gold standard arrived at by a consensus of expert raters. The written anatomical protocol can be obtained via the Internet:http://www.loni.ucla.edu/~khayashi/Public/medial_surface/.
As these cortical pattern-matching algorithms spatially relate homologous regions between individuals, anatomically comparable measures of cortical thickness may be derived for each subject. Cortical thickness—defined as the 3D distance, in millimetres, between inner gray matter (GM) / white matter (WM) border and the closest point on the outer surface (cerebrospinal fluid/GM border)—was calculated using the Eikonal fire equation (Thompson et al. 2005), applied to voxels classified as GM. More specifically, we identified the GM–WM interface as the set of voxels classified as GM that have at least one neighboring WM voxel, setting the distance values for this layer of voxels to zero. To quantify cortical thickness, successive layers of voxels were coded by assigning them a value equal to the closest 3D distance to the GM–WM interface. As such, the coding of distance values progresses from the inner surface of the cortex outwards, so the thickness field increases in a direction that is approximately perpendicular to the GM–WM interface (Luders et al. 2006).
We also applied an algorithm we recently developed to measure the fractal dimension (complexity) of the human cerebral neocortex in 3D (Thompson et al. 2005; Luders et al. 2006), in which the cortex is first divided into 4 separate surface meshes (frontal, temporal, parietal, and occipital regions) in each hemisphere using manually delineated anatomical constraints. Previous approaches for measuring gyrification have typically compared the length of an inner and outer cortical contour in 2D slices of the brain (see Fig. S1, adapted from Zilles et al. 1988), in order to compute a gyrification index (GI; Bartley et al. 1997; Vogeley et al. 2000). However, this value is biased by the orientation of slicing the brain (Thompson et al. 2005). Thus, here gyral complexity was defined as the rate at which the surface area of the cortex increases relative to increases in the spatial frequency of the surface model used to represent it (Lin et al. 2007). This methodology overcomes 2 shortcomings of gyrification indices. First, the surface-based fractal measure is independent of the direction in which the brain is sliced. Secondly, surface meshes can be adapted precisely to individual lobar anatomy, by using cortical pattern-matching methods to compare homologous brain regions across subjects and groups to the highest possible degree.
As seen in Supplementary Figure S1b, fractal dimension of the cortex was computed in 3D by using the surface mesh that represents the cortex, taking into account the 3D surface geometry (Gu et al. 2003). Frontal regions included cortex anterior to the central sulcus. Parietal regions were defined to include cortex posterior to the central sulci and anterior to the parietal–occipital fissure with temporal region boundaries used as the inferior limits. Temporal regions were defined as the cortex inferior to the Sylvian fissure and defined posteriorly by a line from the posterior limit of the Sylvian fissure (horizontal ramus) to the posterior extreme of the temporal sulci and collateral sulci on the inferior surface of the brain. Occipital regions included cortex bordered by parietal and temporal regions anteriorly. Cortical pattern matching was used to anchor sulcal landmarks to the reparameterized cortex so that corresponding sulci and cortical regions occurred in the same regions of the parameter space across subjects. Values above 2 indicate increasing surface detail and greater gyral complexity. This method is described in further detail in (Thompson et al. 2005; Lin et al. 2007).
We first calculated the mean thickness at each of 65536 cortical surface points across subjects, and then generated statistical maps indicating the degree to which local cortical thickness differences were statistically linked with diagnosis and cognitive/demographic variables (i.e., IQ, age). To do this, at each cortical point a regression was run to assess whether the thickness of the cortical gray matter at that point depended on the covariate of interest. The P value describing the significance of this linkage was plotted at each cortical point, using a color code to produce a statistical map (e.g., Fig. 1). The statistical maps (uncorrected) are crucial for allowing us to visualize the spatial patterns of gray matter deficits, but permutation methods were used to assess the significance of the statistical maps and to correct for multiple comparisons (Thompson et al. 2003). In each case, the covariate (group membership) was permuted 100000 times on an SGI Reality Monster supercomputer, and the number of significant results from these randomizations was then compared with the number of significant results in the true assignment to produce a corrected overall significance value for the uncorrected statistical maps. These permutations measure the distribution of features in the statistical maps that would be observed by accident if group assignment were random, and provide a P value for the observed effects that is corrected for multiple comparisons. The number of permutations N was chosen to control the standard error SEp of omnibus probability p, which follows a binomial distribution B(N, p) with (Edgington 1995).
For regions in which there were significant between-group differences in these neuroanatomic measures, exploratory correlational analyses were conducted within the 22q11.2DS group to examine any association between neuroanatomic and behavioral measures. It was hypothesized that, within brain regions identified in the main effects, there would be significant positive associations between gray matter thickness and general cognitive abilities, as well as arithmetic and attentional functions. Given the number of comparisons, we used a more conservative threshold of α ≤ 0.01, 2-tailed, for determining significance of correlational analyses. To maximize the reliability of these analyses, we examined composite factor scores from the WISC and WIAT batteries, rather than individual subtest scores (Nunnally 1978).
Three-dimensional average maps of cortical thickness across the medial hemispheric surface are shown for the 22q11.2DS and control groups in Figure 1a,b. As evident in (a) and (b), regional thickness patterns are similar in the 2 groups, with the thinnest cortex in striatal regions, and relatively thicker cortex in medial frontal regions. However, healthy subjects showed more pronounced thickening in anterior cingulate cortex. Patients with 22q11.2DS showed 3 major anatomical regions of highly significant cortical thinning: 1) in the anterior cingulate cortex bilaterally, and portions of the medial frontal gyrus, 2) in the subgenual prefrontal cortex, and 3) in inferior–posterior brain regions encompassing the posterior cingulate gyrus, the lingual gyrus, and the cuneus. Effects of greatest magnitude were found in the right ventromedial occipital–temporal cortex (red colors; Fig. 1c, right panel), where the average gray matter thickness in the 22q11.2DS group was more than 14% below the control average. This region of significantly thinner cortex includes the posterior cingulate (retrosplenial) cortex and lingual gyrus, regions critical for spatial navigation and orientation (Committeri et al. 2004). Similar to the pattern of cognitive deficits seen in 22q11.2DS, patients with lesions to this region reportedly have marked difficulty with map drawing tasks and show “topographical disorientation” (Aguirre and D'Esposito 1999). No significant excesses in thickness compared with controls were observed in any cortical location. We also compared the mean difference in thickness with a pointwise estimate of the standard error (SE) in cortical thickness to measure the significance of the thickness decreases (Fig. 1d). The resulting significance map, corrected for multiple comparisons, showed that cortical thickness reduction in the 22q11.2DS group was highly significant, for both the left and right hemisphere (P = 0.019 and P = 0.012, corrected, respectively).
These maps were corroborated by quantitative analyses of lobar thickness values, which indicated that cortical thickness was significantly reduced in 22q11.2DS in frontal (P = 0.005), parietal (P = 0.007), and occipital cortices (P = 0.017), but did not significantly differ from controls in the temporal lobes (P = 0.18).
Statistical comparisons of gyral complexity values revealed a significant increase in gyral complexity in children with 22q11.2DS that was specific to the occipital lobe (F1,32 = 6.52, P < 0.01; Fig. 2). Although gyral complexity was not significantly different in 22q11.2DS and control subjects in frontal, temporal, and parietal regions, this difference remained significant in occipital cortex even after Bonferroni correction for the multiple regions assessed (0.05/4 = 0.0125).
To better understand the implications of this finding of increased complexity, we examined the individual brains of participants with 22q11.2DS and control participants. As seen in Figure 3, increased fissuration was observed in the posterior regions of many 22q11.2 brains. This may have contributed to the anomalous increases in gyral complexity in occipital cortex identified in the 22q11.2DS group.
To determine whether decreased cortical thickness was associated with reductions in cortical complexity, we performed correlation analyses of these measurements. At each cortical point, thickness was regressed against the corresponding lobar gyral complexity value. After controlling for multiple comparisons, no significant correlations were identified in either the 22q11.2DS or control group. Therefore, consistent with prior studies in other populations (Thompson et al. 2005; Lin et al. 2007), there does not appear to be a straightforward relationship between these 2 measures of cerebral anatomy.
Using cognitive evaluations conducted on the same day the scans were acquired, we examined brain-behavior relationships within the 22q11.2DS group by conducting regressions of gray matter thickness/volume and gyral complexity against hypothesized cognitive measures. Because Full Scale IQ (FSIQ) was not significantly correlated with right or left hemisphere cortical thickness (r = −0.08, P = 0.73; r = −0.05, P = 0.83, respectively), nor gyral complexity (right: r = −0.19, P = 0.41; left: r = −0.28, P = 0.22), we did not follow up correlations with FSIQ with post hoc tests of correlations with other cognitive measures. However, total gray matter volume was highly significantly correlated with FSIQ within the 22q11.2 group (r = 0.65; P = 0.002). As such, we conducted post hoc tests of correlations of regional gray matter volumes with IQ Factor Scores (Verbal Comprehension, Perceptual Organization, Freedom from Distractibility [FD], and Processing Speed), and other cognitive measures thought to involve frontoparietal circuitry (e.g., reading and mathematical abilities; Dehaene et al. 1999; Temple et al. 2003). Due to the number of comparisons, we reduced the risk of Type I error by using a more conservative significance threshold of P ≤ 0.01. As seen in Table 2, cognitive measures showed the strongest association with frontal gray matter volumes. Verbal Comprehension (comprised of Vocabulary, Similarities, Information, and Comprehension subtests of the WISC) was highly correlated with frontal (r = 0.70, P = 0.001) and parietal GM volume (r = 0.56, P = 0.01), but was not significantly associated with occipital GM volume (r = 0.44, P = 0.053). Both Perceptual Organization (PO; Block Design, Picture Completion, Picture Arrangement, and Object Assembly subtests) and FD (comprised of Digit Span and Arithmetic subtests) were significantly correlated with frontal and parietal GM volume, although only PO was significantly associated with occipital GM volume at P ≤ 0.01. Notably, Processing Speed (Digit Symbol Coding and Symbol Search subtests) was not significantly correlated with gray matter volume (P > 0.50 for all 3 comparisons). Regarding composite scores for the WIAT Achievement tests, only the Reading Composite score (Reading Comprehension and Basic Reading) was significantly associated with frontal gray matter volume (r = 0.56, P ≤ 0.01). Mathematics Composite scores (Numerical Operations and Math Reasoning) were associated with frontal and parietal gray matter at a nonsignificant trend level (P = 0.06). All significant correlations remained significant at P ≤ 0.01 after controlling for total intracranial volume (Table 2).
Correlational analyses of age with cortical thickness, analyzed separately for the 2 groups, revealed nonsignificant negative correlations between age and overall cortical thickness in the control group, following multiple comparison correction (corrected P value, left hemisphere: P = 0.14, right: P = 0.13). However, this inverse relationship was highly significant in the 22q11.2DS group (right: P = 0.01, left: P = 0.008), suggesting a more pronounced effect of age on cortical thickness. These effects are visualized in the maps depicted in Figure 4.
To further examine the effects of age on group differences in cortical thickness, we reanalyzed our group comparisons using analysis of covariance controlling for age. After covarying for age, results of diagnosis remained significant (F3,29 = 3.24, P < 0.04). Age was also a significant predictor in the model (F3,29 = 10.49, P < 0.001), indicating that age and diagnosis both exerted significant independent effects on cortical thickness.
In this study, we used computational methods to examine key midline regions that may be affected by early malformations of cortical development resulting from a chromosome 22q11.2 deletion. Cortical thickness maps and gyral complexity analyses provided a detailed characterization of areas of anomalous cortical development in this syndrome, revealing 3 main findings. First, we detected discrete regions of reduced cortical thickness in bilateral ventromedial occipital–temporal cortex, as well as the anterior cingulate, medial frontal, and subgenual prefrontal cortex. Second, fractal dimension (complexity) measures of the human cerebral neocortex in 3D revealed that children with 22q11.2DS had significantly increased gyral complexity specifically in the occipital cortex. Third, variation in gray matter volume, particularly in the frontal cortex, was strongly correlated with measures of cognitive ability, suggesting that, within 22q11.2DS, regionally specific neuroanatomic alterations may underlie cognitive and behavioral differences.
Here we visualized for the first time the profile of cortical thickness deficits across the medial hemispheric surface associated with the 22q11.2 chromosomal deletion. Decreases in cortical thickness were most pronounced (by 10–15%) in the ventromedial occipital–temporal cortex. Functional neuroimaging studies have shown that this region is critical for visuospatial navigation and directing spatial attention (Galati et al. 2001), cognitive areas of disproportionate deficit in patients with 22q11.2DS (Simon et al. 2002; Simon, Bearden, et al. 2005). Cortical thinning in this region of the brain may underlie the prominent mathematical difficulties characteristic of this syndrome (Barnea-Goraly et al. 2005; De Smedt et al. 2007), as numerical–spatial interactions are believed to arise from common parietal circuits for attention to external space and internal representations of numbers (Hubbard et al. 2005). Using a large battery of mathematical tests (De Smedt et al. 2007) found that children with 22q11.2DS had specific difficulty with the semantic manipulation of quantities, a cognitive function which depends on a bilateral inferior parietal network (Dehaene et al. 1999).
Although prior volumetric studies have primarily detected deficits in posterior brain regions, with relative preservation of frontal lobe regions, we also identified localized decreases in cortical thickness in frontal regions, particularly within the subgenual prefrontal cortex and anterior cingulate. Both functional neuroimaging and lesion studies in human and nonhuman primates have demonstrated that these brain regions are critical for emotional modulation, motivation and attention (Goldman et al. 1974; Pardo et al. 1990; Devinsky et al. 1995; Drevets et al. 1998), functions which are consistently reported to be significantly affected in both children and adults with 22q11.2DS (Swillen, Vandeputte, et al. 1999; Arnold et al. 2001). Structural and functional abnormalities in these brain regions are also frequently reported in children with attention deficit hyperactivity disorder (see Durston 2003 for a review), a diagnosis received by 30–50% of children and adolescents with 22q11.2 deletions (Papolos et al. 1996; Feinstein et al. 2002; Antshel et al. 2005). Additionally, a recent functional MRI study identified hypoactivation in the dorsolateral prefrontal cortex and anterior cingulate in children with 22q11.2 deletions during performance on a verbal working memory task, which was not attributable to performance decrements, suggesting that the frontal component of the distributed network subserving executive function and working memory may be disrupted in youth with this syndrome (Kates et al. 2007).
Despite the pronounced cortical thinning observed in posterior brain regions in children with 22q11.2DS, we found concomitant increases in gyral complexity in these regions. These findings may be related, as it may be that in order to fit more GM into the same surface, the cortex must become more gyrified in those regions (Toro and Burnod 2005). According to the radial unit model of cortical evolution (Rakic 1988b), an increase in cerebral convolutions results in a net increase in cortical surface due to the addition of radial units, or minicolumns (Rakic 2004). Thus, in disorders involving a larger than normal number of convolutions (i.e., PMG), the affected cortex is thinner, but is nevertheless associated with an increase in total cortical surface, suggesting that—in the extreme—increased gyral complexity is associated with cortical thinning. However, the lack of direct correlation between gyral complexity and thickness suggests that there is not a straightforward, linear relationship between surface geometry (e.g., gyrification measures) and regional tissue concentration (e.g., cortical thickness and GM volume). Only one other study, to our knowledge, has examined gyral complexity in 22q11.2DS (Schaer et al. 2006). Using a semiautomated method for calculating a GI, this study observed decreased gyrification in the frontal and parietal lobes. Methodological differences are the most likely explanation for these contrasting findings. In particular, because gyrification indices are known to be affected by the orientation of slicing the brain (Thompson et al. 2005), the derivation of the GI from coronal sections may have obscured differences that would be easier to detect in other planes.
Notably, a similar pattern of increased gyral complexity in posterior cortex has been observed in Williams Syndrome (WS), a genetic deletion syndrome with a characteristic cognitive profile—involving relative strengths in verbal memory, in contrast to marked deficits in visuospatial memory—that bears striking similarity to that seen in 22q11.2DS (Bearden et al. 2002). A recent study using similar methods to analyze gyrification with excellent spatial resolution revealed increased gyrification bilaterally in occipital regions and over the cuneus in WS subjects (Gaser et al. 2006). Thus, it is tempting to speculate that a similar neuroanatomic substrate may underlie the visuospatial impairments observed in patients with Williams Syndrome and 22q11.2DS, by affecting the flow of information through distributed neural systems; this hypothesis awaits further testing using multimodal imaging approaches.
Perhaps surprisingly, there was not a significant correlation of IQ with overall cortical thickness in children with 22q11.2DS. The thickness of the cortex, ranging between 1.5 and 4.5 mm across cortical regions, reflects cytoarchitectural characteristics of the neuropil including the density and arrangement of neurons, neuroglia, and nerve fibers (Armstrong et al. 1995). As such, we might expect measures of cortical thickness to more closely link with cognitive abilities than volumetric measures. However, only a few studies have examined the relationship between intelligence and cortical thickness, with mixed results. Shaw et al. (2006) examined the trajectory of change in cortical thickness from childhood to young adulthood, and found that relationships between IQ and cortical thickness were not significant in early childhood, but that these relationships shifted toward positive correlations (predominantly in frontal cortical regions) in older subjects, suggesting that the correlation is highly age dependent and may be zero or even negative in younger individuals. Using stepwise multiple regression analyses, a prior study of healthy adults found gray matter to be the best predictor of variation in IQ, when including overall intracranial, gray, and white matter volumes in the model (Narr et al. 2007). Although this study also found significant associations between FSIQ and cortical thickness in prefrontal (BA 10/11 and 47) and temporal (BA 37 and 36) cortical regions, the sample consisted of a large group of healthy young adults (N = 65), so differences across studies may be either a function of sample size, and/or developmental differences.
Nevertheless, we did find that gray matter volume deficits in frontal and parietal regions were significantly linked to IQ scores in 22q11.2DS, with the strongest associations found between frontal gray matter and verbal and visuo-constructive abilities. The same GM deficits were also less strongly linked with academic abilities in reading and arithmetic. This pattern is highly consistent with 2 studies examining the relationship between gray matter density and IQ in normal adults (Haier et al. 2004) and normal twin pairs (Thompson, Cannon, et al. 2001), both of which found the strongest linkage between gray matter in frontal regions and measures of intelligence. Additionally, functional brain imaging studies support the notion that activation within the frontal lobes is the primary source of differences in performance on tasks of “general intelligence” (g) (Duncan et al. 2000). Interestingly, the correlations identified here in patients with 22q11.2DS are markedly higher than those generally reported in healthy populations, as a recent meta-analysis of the relationship between in vivo brain volume and intelligence in normal individuals, involving 37 studies and a total of 1530 people, estimated the population correlation to be 0.33 (McDaniel 2005). This difference may be partially attributable to the greater variability in cognitive function in 22q11.2DS, although this possibility requires confirmation in a larger sample of patients. In any case, these findings clearly demonstrate that the observed neuroanatomic alterations have functional significance in patients with this syndrome.
Although this study was not designed to examine the developmental trajectory of cortical thickness alterations in 22q11.2DS, our data provide preliminary evidence for differences in the pattern of age-associated cortical thinning in this syndrome. In healthy control subjects, localized cortical thinning, primarily in superior parietal and cingulate cortices, was associated with increasing age. This pattern is fairly consistent with the trajectory of cortical thickness seen in healthy comparison subjects in a recent longitudinal study of child onset schizophrenia (Vidal et al. 2006). However, children with 22q11.2DS showed a more posterior and more widespread pattern of age-related cortical thinning, particularly in ventromedial occipital–temporal regions. The time course of the observed differences will be assessed in the future in a prospective longitudinal design, in order to establish when in the course of development these differences emerge, and how they change over time within individual patients with 22q11.2DS.
Cortical thinning over the course of normal development likely reflects synaptic pruning, a process which is generally associated with increased cognitive efficiency (Johnson and Munakata 2005). However, excessive pruning may indicate a pathological process, as suggested by the pattern of gray matter loss seen in adolescents with childhood-onset schizophrenia, in which early gray matter deficits in the parietal cortex progressed anteriorly into the temporal lobes and engulfed sensorimotor and dorsolateral prefrontal cortices, resulting in dramatic gray matter loss over a 5-year period (Thompson, Vidal, et al. 2001; Vidal et al. 2006).
Although little is known regarding longitudinal brain development in children with 22q11.2DS, the first longitudinal study to examine the developmental trajectory of brain volume in 22q11.2DS was recently completed (Gothelf et al. 2005, 2007). Children and adolescents with 22q11.2DS who were rescanned, on average, 5 years later, showed a greater longitudinal increase in white matter, and superior temporal gyrus and caudate nucleus volumes compared with typically developing controls, as well as a more robust decrease in amygdala volume. The trajectory of change in the candidate brain regions examined in this study did not distinguish between psychotic and nonpsychotic adolescents with 22q11.2DS, although this may have been the result of limited power to examine subgroup differences.
The human neocortex is organized into functionally unique subdivisions that are distinguished by differences in cytoarchitecture, input and output connections, and gene expression patterns (O'Leary and Nakagawa 2002; Rash and Grove 2006). In adults, borders between neocortical areas can be sharply defined by differences in cytoarchitecture and neuronal connections, as well as differences in gene expression. These properties determine the functional specializations of particular cortical subregions.
The term “cortical arealization” refers to the process of regional and areal differentiation of the developing mammalian neocortex (Mallamaci and Stoykova 2006). This process involves an interplay between intrinsic genetic mechanisms and extrinsic information relayed to the cortex by thalamocortical input. Several transcription factors, with varying expression across the embryonic cortical axes, are now known to determine the size and position of particular cortical areas (O'Leary et al. 2007). Members of the fibroblast growth factor (Fgf) family of genes, for example, are implicated in the control of neocortical regionalization. Using a set of gene expression markers to distinguish subdivisions of the newborn mouse frontal cortex (Cholfin and Rubenstein 2007) found that loss of a particular fibroblast growth factor gene, Fgf17, selectively reduces the size of the dorsal frontal cortex, whereas the ventral and orbital frontal cortical regions maintained normal appearance. These changes were complemented by a rostromedial shift of sensory cortical areas in Fgf17 mutant mice. Thus, in addition to an overall effect on patterning the neocortical map, Fgf17 showed an unexpectedly selective role in regulating dorsal frontal cortical development. Genetic manipulations during embryonic development that result in fairly subtle decreases or increases in the sizes of somatosensory and motor areas in adults can result in significant deficiencies at tactile and motor behaviors, suggesting that these areas have an optimal size for maximum behavioral performance (see O'Leary et al. 2007 for a review). Our findings of highly robust relationships between neuroanatomic variation in specific brain regions and phenotypic variation in patients with 22q11.2 deletions are in line with this view.
According to the radial unit hypothesis, the embryonic cerebral ventricle contains proliferative units that provide a “proto-map” of prospective cytoarchitectonic areas (Rakic 1988b). Neurons proliferate in the germinal zones, within the walls of the lateral ventricles, then migrate along various pathways to the appropriate layers of distinct cytoarchitectonic areas in the developing cortex, where they disengage from the guide cell and begin to extend neurites and establish synaptic connections (Guerrini et al. 2008). This model thus offers a framework for understanding genetic and epigenetic regulation of the division of cytoarchitectonic regions, as well as the pathogenesis of particular disorders of cortical development (Rakic 2004). As our data suggest that a specific genetic mutation may affect the generation and migration of cortical neurons that are destined for particular regions of the cortex, our findings provide additional, albeit indirect, support for the protomap hypothesis.
Over the past several years, genetic studies of cerebral cortical development in mice and in humans have identified several genes that, when mutated, can disrupt each of the main stages of cell proliferation and specification, neuronal migration and late cortical organization, resulting in disorders of neuronal migration and cerebral cortical development (Barkovich et al. 2005; Guerrini et al. 2008). It is likely that the cortical dysmorphology observed here is shaped primarily by genetically programmed anomalous neurodevelopment that disrupts midline cortical development. Moreover, findings of increased gyrification in posterior cortical regions suggest that one or more genes in the 22q11.2 region is involved in cortical development during the critical period when cortical folding occurs. Several genes that direct early cell differentiation and are highly expressed in the brain, including Tbx1, goosecoid-like, ZNF74 zinc finger gene, GNB1L, which is widely expressed in the mouse forebrain, midbrain and hindbrain, and proline dehydrogenase, which regulates glutamate and γ-aminobutyrate neurotransmission (Paterlini et al. 2005), are located in the deleted region (Maynard et al. 2003). The expression of several zinc finger proteins has been shown to be critical for normal development of the cerebral cortex (Aruga et al. 1998; Chen, Schaevitz, et al. 2005; Chen, Rasin, et al. 2005). Four zinc finger genes map to 22q11.2 within the DiGeorge syndrome region, suggesting that these may be potential candidate genes for the developmental malformations associated with this syndrome (Aubry et al. 1992). The Zinc finger gene 312 (Zfp312, also known as Fez1) has also been demonstrated to play a critical role in the patterning of cortical axonal projections and the dendritic development of pyramidal neurons, the largest cellular elements in cortex (Chen, Rasin, et al. 2005).
Several studies in mice have found Tbx1 to be the critical cause of the cardiac defects characteristic of the 22q11.2 phenotype (e.g., Vitelli et al. 2002), and may also be involved in some of the behavioral and psychiatric symptoms associated with the syndrome (Paylor et al. 2006). Another T-box transcription factor, T-box-brain2, leads to microcephaly with PMG and callosal agenesis (Baala et al. 2007), all of which have been reported in rare cases in individuals with 22q11.2DS (Cramer et al. 1996; Kraynack et al. 1999; Robin et al. 2006). In this regard, it is also interesting that another PMG syndrome, frontoparietal PMG, has been mapped to chromosome 16q12.2–21 (Piao et al. 2002), and subsequently associated with mutations of a gene in the G-protein-coupled receptor family, the Gpr56 gene (Piao et al. 2004). Thus, the gene expression patterns seen in a mouse Gpr56 mutant, as well as the pattern of cortical abnormalities seen in patients with homozygous Gpr56 mutations, offer convincing evidence that this gene regulates cortical patterning (Piao et al. 2004). Although it is not yet clear precisely what gene(s) in the 22q11.2 region may be critically implicated in the identified cortical malformations, it is intriguing that—consistent with our findings of predominantly right-lateralized cortical thinning in ventromedial cortex—Robin et al. (2006) observed a striking tendency toward right hemisphere PMG in a series of 22q11.2DS patients, suggesting that the relevant genes may be asymmetrically expressed in the brain.
Finally, the catechol-O-methyl transferase (COMT) gene, located within this region, has a major role in dopamine metabolism, particularly in the prefrontal cortex (Tunbridge et al. 2006). Several lines of evidence point to a role of COMT in both executive cognition and susceptibility to psychosis (Egan et al. 2001), leading to speculation that COMT haploinsufficiency is responsible for the behavioral and cognitive abnormalities in 22q11.2DS. Although there is some evidence that genetic variation in COMT may influence working memory and executive cognition, both in individuals with 22q11.2DS and in the general population (Bearden et al. 2004a; Gothelf et al. 2005; Shashi et al. 2006), evidence for its involvement in psychosis susceptibility remains equivocal (Murphy and Owen 2001; Bassett et al. 2007).
In addition, other neurobiological mechanisms occurring downstream of the genetic lesion are likely to also be involved in the observed neuroanatomic alterations; for example, abnormal mechanical constraints may arise from inappropriately targeted or missing neural connections. Lesion studies in nonhuman primates have shown that disruption of afferent pathways, when occurring very early in development, can lead to the emergence of abnormal sulcal and gyral patterning (Goldman and Galkin 1978; Rakic 1988a, 1988b). Although cortical architecture in 22q11.2DS is highly likely to be affected by haploinsufficiency for particular genes in the deleted region, epigenetic and environmental influences are likely contributors as well. As such, direct causality cannot be determined from this study, in which we correlate a genetic mutation with neuroanatomic alteration. Longitudinal studies are clearly warranted in order to begin to disentangle the complex genetic and nongenetic influences that contribute to the neuroanatomic and cognitive abnormalities in this syndrome.
Because comparison subjects were not IQ-matched to the 22q11.2DS patients, IQ significantly differed between groups. Because IQ is inextricably correlated with diagnosis, we did not control for IQ in our analyses, as doing so would incorrectly eliminate some disease-specific effects. In developmentally delayed populations the issue of appropriately matched comparison subjects is complex, as individuals with comparable IQ to those with 22q11.2 deletions are likely to have intellectual disability of heterogeneous etiology, such as undetected chromosomal abnormalities or unknown environmental exposures (e.g., lead exposure, birth complications), which are likely to lead to a variety of cortical anomalies that are not well characterized. Moreover, including children with familial low IQ and/or environmental exposures would likely lead to systematic unmatching on other demographic variables, such as parental education. Thus, we adopted this more straightforward approach for this investigation, but clearly an optimal design for future studies would include both normal as well as IQ-matched comparison groups, possibly involving another discrete genetic etiology in order to minimize heterogeneity.
These findings offer novel information regarding patterns of neuroanatomic alteration and their relationship to cognitive abilities in children with 22q11.2 deletions. The fact that the neuroanatomic abnormalities in this syndrome are localized to particular brain regions adds to the growing body of evidence from both human and experimental animal studies that specific cytoarchitectonic areas can be a selective target for gene mutations. Aberrant parieto-occipital brain development, as evidenced by both increased complexity and cortical thinning in these regions, suggests a compelling neuroanatomic substrate for the deficits in visuospatial and numerical understanding in 22q11.2DS. In addition, significant cortical thinning in medial frontal and cingulate regions may contribute to the emotional and attentional difficulties characteristic of this syndrome. These observations suggest new hypotheses regarding the effects of haploinsufficiency for genes in the 22q11.2 region, the investigation of which will result in a more complete elucidation of the associations between genes, brain, cognition and behavior, both in 22q11.2DS and in the broader population.
National Institute of Mental Heath (K23MH074644-01) to C.E.B., and (PO1-DC02027) to B.S.E.; National Institute on Aging; the National Library of Medicine; the National Institute for Biomedical Imaging and Bioengineering; the National Center for Research Resources; and the National Institute for Child Health and Development (AG016570, LM05639, EB01651, RR019771, and HD050735) to P.M.T. supported the Algorithm development.
We wish to thank all children and their families who participated in this study. We also thank Lara Zimmermann and Helen Tran for assistance with processing of neuroimaging data. Conflict of Interest: None declared.