Abstract

Introduction

Most maps of human cerebral cortex have partitioned it according to histological features, such as the distribution of cell bodies or myelin, and, more recently, molecular markers (von Economo and Koskinas, 1925; Ongur et al., 2003; Zilles et al., 2004). Comparisons of similar classifications among several species provide an evolutionary perspective, and such analyses have identified two broad cortical types. One type, allocortex, has a primitive three-layered form that strongly resembles its homologues in reptiles. Another type, isocortex, lacks such homologues and has a more derived structure characterized by more than three layers (typically six) and a more complex pattern of afferent and efferent projections (Kaas, 1987; Puelles, 2001; Allman et al., 2002; Striedter, 2005). Between allocortex and isocortex, areas sometimes called “transition cortex” have intermediate properties. Using structural neuroimaging of the in vivo developing brain, we explored the possibility that these different kinds of cortex show differing levels of complexity in the trajectories of their growth in childhood and adolescence.

Using computational neuroanatomy, we defined cortical thickness at over 40,000 points throughout the cerebrum in a cohort of 375 healthy children and adolescents. Cortical thickness was chosen as a metric which both captures the columnar architecture of the cortex and is sensitive to developmental change in typically developing and clinical populations (Lerch et al., 2005; O'Donnell et al., 2005; Makris et al., 2006; Shaw et al., 2006a,b; Lu et al., 2007; Sowell et al., 2007).

Most of the children in our cohort had repeated neuroanatomic imaging, and such longitudinal data can be combined with cross-sectional data to model developmental change, with the longitudinal data being particularly informative. For cortical thickness, the simplest trajectory that can be fitted to describe its change over time is a straight line. More complex growth models include distinct phases of increase and decrease in cortical thickness: a quadratic model has two such phases (typically an initial increase which reaches a peak before declining) and a cubic model has three. The complexity of growth may vary across the cerebral cortex, and we sought to explore whether this variation aligned with cytoarchitectural properties.

Derived properties of developmental curves, such as the age of reaching various points of inflection, are frequently used as developmental indices (Tanner et al., 1976; Jolicoeur et al., 1988). For cortical thickness, the age at which peak cortical thickness is reached (the point where increase gives way to decrease in cortical thickness) can be determined for cortical points with either a cubic or quadratic (but not linear) trajectory and, thus, emerges as a potentially useful index of cortical development. We therefore additionally examined the pattern of attaining peak cortical thickness across the cerebrum, to confirm and expand previous observations of a heterochronous sequence, with primary sensory areas attaining peak cortical thickness before polymodal, high-order association areas (Gogtay et al., 2004).

Materials and Methods

Participants.

Three hundred and seventy-five children and adolescents, healthy children with no personal or family history of psychiatric or neurological disorders, had a total of 764 magnetic resonance images. Each subject completed the Childhood Behavior Checklist as a screening tool and then underwent a structured diagnostic interview by a child psychiatrist to rule out any psychiatric or neurological diagnoses (Giedd et al., 1996). Handedness was determined using the PANESS (Physical and Neurological Examination for Soft Signs), and 336 (90%) were predominately right-handed, 20 (5%) predominately left-handed, and 19 (5%) ambidextrous. The mean intelligence quotient (IQ) was 115 (SD, 13) as determined from age appropriate versions of the Wechsler Intelligence Scales (Shaw et al., 2006b). Socioeconomic status (SES) was determined from the Hollingshead Scales and the mean score was 40 (SD, 19) (Hollingshead, 1975). The age range spanned from 3.5 to 33 years, and the age distribution of scans is illustrated in Figure 1. The subjects came from 292 different families; 196 (52%) were male. The age range spanned from 3.5 to 33 years. All subjects had at least one scan (mean age at initial scan, 12.3 years; SD, 5.3); 203 (54.1%) had at least two scans (mean age, 13.8; SD, 4.6); 106 (28.3%) had at least three scans (mean, 15.3; SD, 4.2); and 57 (15.2%) had four or more scans (mean 18, SD 4.5).

Open in a separate window

Figure 1.

The age distribution of the data. The age at each scan is indicated by a blue diamond. For each subject, the first scan is always the leftmost; subjects with repeated scans have a horizontal line drawn connecting the age at first scan with the age at later scans.

Neuroimaging.

T1-weighted images with contiguous 1.5 mm axial slices and 2.0 mm coronal slices were obtained using three-dimensional spoiled gradient recalled echo in the steady state on a 1.5-T General Electric (Milwaukee, WI) Signa scanner. Imaging parameters were as follows: echo time, 5 ms; repetition time, 24 ms; flip angle, 45°; acquisition matrix, 256 × 192; number of excitations, 1; and field of view, 24 cm. Head placement was standardized as described previously (Giedd et al., 1999). The same scanner was used throughout the study. The native MRI scans were registered into standardized stereotaxic space using a linear transformation and corrected for nonuniformity artifacts (Sled et al., 1998). The registered and corrected volumes were segmented into white matter, gray matter, CSF, and background using an advanced neural net classifier (Zijdenbos et al., 2002). A surface deformation algorithm was applied which first fits the white matter surface, then expands outward to find the gray matter-CSF intersection defining a known relationship between each vertex of the white matter surface and its gray matter surface counterpart; cortical thickness is defined as the distance between these linked vertices (and measured at 40,962 such vertices) (MacDonald et al., 2000). A 30-mm-bandwidth blurring kernel was applied; this size was chosen on the basis of population simulations which that this bandwidth maximized statistical power while minimizing false positives (Lerch and Evans, 2005). This kernel allows anatomical localization, as 30 mm blurring along the surface using a diffusion smoothing operator preserves cortical topologic features and represents considerably less cortex than the equivalent volumetric Gaussian blurring kernel.

The validity of this automated measure against expert manual neuroanatomical estimation of cortical thickness has been demonstrated previously for selected cortical regions in an adult population (Kabani et al., 2001) We repeated this validation study in our pediatric population in the cortical regions included in the original study (the pre and post central gyri, the superior frontal gyrus, the superior temporal gyrus, the cuneus, the superior parietal lobule and supramarginal gyrus) (Kabani et al., 2001). We also examined regions of particular interest for this study. These were the insula, the orbitofrontal cortex (measured bilaterally in its anterior, posterior, medial, and lateral divisions), and medial cortical regions (the anterior and posterior cingulate, the medial dorsal prefrontal cortex, and the parahippocampal gyrus). Twenty scans were chosen at random from the cohort (from ages 6 through 15). For each brain region the neuroanatomist (N.K.) used image analyses software (MacDonald, 1996) to mark one point or tag on the CSF and gray matter border which represents the outer surface of the cortex, and another point of the gray and white matter border which represents the inner surface of the cortex. The distance between the two tags was calculated, mimicking the algorithm used by the automated tool. For a given tag placed by the neuroanatomist on the outer cortical surface, the closest vertex on the automatically extracted cortical surface was identified and its associated cortical thickness was noted. The output of the manual and automatic methods were compared using a repeated measures ANOVA followed by paired t tests to identify regional differences. There was a significant difference for type of measurement with the automated estimates being larger (mean, 4.62; SE, 0.06) than the manual (mean, 4.41; SE, 0.04; F_(1,684) = 8.8, p = 0.02). There was a significant interaction of the type of measurement and region (F_(35,684) = 2.59, p < 0.001) which was further explored. Overall there was no significant difference between the manual and automated measures in 30 of the 36 regions, with poorer performance noted bilaterally in the precentral gyrus, and in the left postcentral gyrus, and in the middle frontal gyrus, gyrus rectus and the cuneus in the left hemisphere. Notably, only one of these regions lay in an area of particular interest for this study (the left gyrus rectus). There was no correlation between age and the difference between the automated and manual estimates (r = 0.02, p = 0.53). Thus, there was no evidence that the differences between the two metrics had any significant age related bias.

To determine developmental trajectories at each cortical point, mixed model regression analysis was chosen as it permits the inclusion of multiple measurements per person, missing data, and irregular intervals between measurements, thereby increasing statistical power (Pinheiro and Bates, 2000). Our classification of developmental trajectories was based on a step-down model selection procedure: at each cortical point we modeled cortical thickness using a mixed-effects polynomial regression model, testing for cubic, quadratic and linear age effects. If the cubic age effect was not significant at p < 0.05, it was removed and we stepped down to the quadratic model and so on. In this way, we were able to classify the development of each cortical point as being best explained by a cubic, quadratic, or linear function of age. We consider cubic models to be more complex than quadratic, which in turn are held to be more complex than linear models. A random effect for each individual was nested within a random effect for each family, thus accounting for both within-person and within-family dependence. Thus, for cortical points with a cubic model, the kth cortical thickness of the ith individual in the jth family was modeled as thickness_ijk = intercept + d_ij + β₁(age) + β2*(age)**2 + β3*(age)**3 + e_ijk, where d_ij are nested random effects modeling within-person and within family dependence, the intercept and β terms are fixed effects, and e_ijk represents the residual error. Quadratic models lacked the cubic age term, and linear models the cubic and quadratic age terms. The analyses were repeated entering SES and IQ as covariates.

The age at which peak cortical thickness was attained was calculated for cubic and quadratic models from the first order derivatives of the fitted curves.

Results

Throughout most of the lateral frontal, lateral temporal, parietal and occipital isocortex, developmental trajectories are cubic, with a period of initial childhood increase, followed by adolescent decline and then stabilization of cortical thickness in adulthood (Fig. 2). Growth characterized by increase and decrease, but lacking the phase of stabilization within the first three decades of life (a quadratic model) is present in much of the insula and anterior cingulate cortex. A linear trajectory is seen in the posterior orbitofrontal and frontal operculum, portions of the piriform cortex, the medial temporal cortex, subgenual cingulate areas, and medial occipitotemporal cortex. Graphs illustrating individual data points from representative regions with a cubic, quadratic or linear trajectory are shown in Figure 3.

Open in a separate window

Figure 2.

Complexity of developmental trajectories throughout the cerebral cortex. The brain maps show the vertices having a cubic (red), quadratic (green) or linear (blue) developmental trajectory. The graphs show the growth pattern for each of these divisions. In order there are dorsal, right lateral, left medial, left lateral, and right medial views. The corpus callosum is blacked out.

Open in a separate window

Figure 3.

Graphs showing raw cortical thickness data in blue, with the fitted trajectory superimposed in pink. a–c, The first three images show in order the mean cortical thickness and trajectory for representative regions: the superior frontal gyri, which have a cubic trajectory (a); the portion of the insula which has a quadratic trajectory, seen in green in Figure 5 (b); the portion of the orbitofrontal cortex which has a linear trajectory, seen in blue in Figure 4 (c).

We examined the complexity of developmental trajectories with respect to cortical regions of differing cytoarchitecturonic types, using histological atlases to assign cytoarchitectonic fields (Ongur et al., 2003). This analysis revealed a clear parallel between basic types of cortex and the pattern of cortical development. The orbitofrontal cortex exemplifies the correspondence between cortical types and developmental trajectories (Fig. 4). In the most anterior part of this region, a cubic trajectory characterizes the homotypical (six-layered) isocortex of the frontal pole and lateral orbitofrontal regions. In contrast, most of the cortex on the posterior orbital surface follows relatively simple quadratic and linear growth trajectories. This region has a lamination pattern typical of transitional cortex: compared with homotypical isocortex it has fewer, less well developed layers and lacks the clear concentration of nonpyramidal cells of layer 4, the internal granular layer (Brockhaus, 1940; Mesulam and Mufson, 1982; Ongur et al., 2003). In the most posterior part of this region, linear and quadratic growth characterizes the piriform cortex, a primitive allocortical area that subserves olfaction.

Open in a separate window

Figure 4.

A, Complexity of developmental trajectories throughout the orbitofrontal cortex, projected onto a standard brain template. The anterior and lateral orbitofrontal cortex has a cubic fit (red); medial and posterior orbitofrontal cortex has simpler quadratic (green) and linear (blue) trajectories. B, The trajectories are superimposed on a cytoarchitectonic map of the region by Öngür et al. (2003) to illustrate the overlap between the cytoarchitectonic fields and regional differences in trajectories. C, The trajectory of each of the divisions.

Although Figure 4 focuses on the orbitofrontal cortex, the same principles are observed generally, where a transition from isocortex to simpler forms occurs. The results for the medial frontal cortex are similar to those in the orbitofrontal cortex, with cubic growth anteriorly, especially in the homotypical cortex of the medial frontal pole- and linear or quadratic trajectories more posteriorly in regions of dysgranular or agranular architecture (Fig. 5, top). For the insula (Fig. 5, bottom), the pattern is much the same. The anterior insula, with its agranular and poorly laminated cortex, has a linear developmental trajectory. Moving posteriorly to the dysgranular and homotypical insula, there is at first a more complex quadratic fit; still more posteriorly, as the cortex becomes increasingly homotypical, the trajectory becomes cubic. Likewise in the temporal lobe, an allocortical component such as the piriform cortex shows a predominately a linear trajectory. In contrast, the lateral temporal isocortex has a cubic trajectory and transition areas such as the entorhinal and perirhinal regions have quadratic and linear trajectories (Fig. 2). These results are summarized in Table 1. There are some cortical regions where this link between cortical types and developmental trajectories does not hold, most notably in the medial occipitotemporal and the anterior superior temporal areas, both of which are isocortical regions that nonetheless have a linear and quadratic trajectory, respectively. The pattern of results held when SES and IQ were entered as covariates, either separately or together.

Open in a separate window

Figure 5.

Top, Detailed views of trajectories in the right medial prefrontal cortex, where isocortical regions have a cubic trajectory, and transitional areas have either a quadratic trajectory (e.g., the agranular and poorly laminated cortex of area 24a in the cingulate gyrus) or a linear decline in thickness (e.g., the thin and largely agranular cortex of the gyrus rectus). Bottom, The right insula shows progressively more complex trajectories moving: the posterior portion has a cubic trajectory (red), the body of the insula has a quadratic fit (green) and the anterior insula has a linear fit (blue). A similar pattern holds for the left insula.

We next determined the age at which peak cortical thickness was attained for all points with either a cubic or quadratic trajectory, using the first order derivative of the fitted curve for each point. A peak age cannot be determined for the points with a linear trajectory. The results are presented as a time-lapse dynamic sequence (supplemental Movies 1, 2, available at www.jneurosci.org as supplemental material), “stills” taken from the movies (Fig. 6), and the estimated age of peak cortical thickness for 56 brain regions (as defined by the ANIMAL segmentation tool).

Open in a separate window

Figure 6.

Age of attaining peak cortical thickness across the cerebral cortex. Peak thickness can only be estimated for regions with a cubic or quadratic trajectory and not for regions with linear change (which are indicated with a darker red shade). The changes are illustrated dynamically in supplemental Movies 1 and 2, available at www.jneurosci.org as supplemental material.

To summarize the results, within isocortex, the primary sensory and motor areas generally attain their peak cortical thickness before adjacent secondary areas, and also before other polymodal association areas. In the posterior brain, the first area to reach its peak thickness is the somatic sensory cortex (~7 years), followed by the occipital poles, containing much of the striate primary visual area (~7 years on the left, and ~8 years on the right) and then the remaining parieto-occipital cortex, with polymodal regions (such as the posterior parietal cortex) reaching peak thickness later (~9–10 years). In the frontal cortex, the primary motor cortex attains peak cortical thickness relatively early (~9 years), followed by the supplementary motor areas (~10 years) and most of the frontal pole (~10 years). High-order cortical areas, such as the dorsolateral prefrontal cortex and cingulate cortex, reach peak thickness last (~10.5 years). In the medial views, the occipital and frontal poles attain peak thickness early, and then a centripetal wave sweeps from these areas, with the medial prefrontal and cingulate cortex attaining peak thickness last. There is also a marked dorsal to ventral progression of development. Detailed results for each brain regions are given in Table 2.

Table 2.

The estimated age of peak cortical thickness is given for 56 brain regions

	Hemisphere	Trajectory	Age of peak cortical thickness (years)
Frontal
Superior	R	Cubic	10.2
	L	Cubic	10.1
Middle	R	Cubic	10.4
	L	Cubic	10.3
Inferior	R	Cubic	9.7
	L	Cubic	9.7
Medial	R	Cubic	10.6
	L	Cubic	10.0
Precentral	R	Cubic	9.6
	L	Cubic	10.5
Cingulate	R	Cubic	13.8
	L	Cubic	11.2
Medial orbitofrontal	R	Linear
	L	Cubic	8.6
Lateral orbitofrontal	R	Cubic	9.4
	L	Cubic	9.4
Temporal
Superior	R	Cubic	14.9
	L	Cubic	14.9
Middle	R	Cubic	11.7
	L	Cubic	11.6
Inferior	R	Cubic	11.2
	L	Cubic	11.1
Insula	R	Quadratic	18.1
	L	Quadratic	18.0
Periamygdaloid	R	Linear
	L	Linear
Parahippocampal	R	Linear
	L	Linear
Uncal	R	Linear
	L	Linear
Parietal
Postcentral gyrus	R	Cubic	8.4
	L	Cubic	8.5
Superior lobule	R	Cubic	8.3
	L	Cubic	9.0
Supramarginal gryus	R	Cubic	9.2
	L	Cubic	9.2
Angular gryus	R	Cubic	8.5
	L	Cubic	9.0
Precuneus	R	Cubic	9.8
	L	Cubic	10.1
Lateral occipitotemporal	R	Cubic	11.2
	L	Cubic	11.1
Medial occipitotemporal	R	Linear
	L	Linear
Occipital
Pole	R	Cubic	7.9
	L	Cubic	6.8
Superior	R	Cubic	8.3
	L	Cubic	8.3
Middle	R	Cubic	9.5
	L	Cubic	9.2
Inferior	R	Cubic	7.3
	L	Cubic	7.9
Lingual	R	Cubic	8.6
	L	Cubic	9.2
Cuneus	R	Cubic	9.2
	L	Cubic	8.8

Open in a separate window

This can be estimated only for regions with either a cubic of quadratic trajectory (thus, regions with a linear trajectory are marked with a dash). For regions with a mix of trajectories (e.g., the insula), the dominant trajectory, which applies to the majority of points within the region, is given. L, Left; R, right.

Discussion

Footnotes

References