The Journal of Neuroscience, November 15, 2001, 21(22):8819–8829

Mapping Continued Brain Growth and Gray Matter Density

Reduction in Dorsal Frontal Cortex: Inverse Relationships during
Postadolescent Brain Maturation

Elizabeth R. Sowell, Paul M. Thompson, Kevin D. Tessner, and Arthur W. Toga

Laboratory of Neuro Imaging, Department of Neurology, University of California Los Angeles, Los Angeles, California
90095-1769

Recent in vivo structural imaging studies have shown spatial growth, showing an inverse relationship between cortical gray
and temporal patterns of brain maturation between childhood, matter density reduction and brain growth primarily in the su-
adolescence, and young adulthood that are generally consis- perior frontal regions that control executive cognitive function-
tent with postmortem studies of cellular maturational events ing. Inverse relationships are not as robust in the posterior
such as increased myelination and synaptic pruning. In this temporo-occipital junction where gray matter density reduction
study, we conducted detailed spatial and temporal analyses of is much less prominent despite late brain growth in these
growth and gray matter density at the cortical surface of the regions between adolescence and adulthood. Overall brain
brain in a group of 35 normally developing children, adoles- growth is not significant between childhood and adolescence,
cents, and young adults. To accomplish this, we used high- but close spatial relationships between gray matter density
resolution magnetic resonance imaging and novel computa- reduction and brain growth are observed in the dorsal parietal
tional image analysis techniques. For the first time, in this report and frontal cortex. These results suggest that progressive cel-
we have mapped the continued postadolescent brain growth lular maturational events, such as increased myelination, may
that occurs primarily in the dorsal aspects of the frontal lobe play as prominent a role during the postadolescent years as
bilaterally and in the posterior temporo-occipital junction bilat- regressive events, such as synaptic pruning, in determining the
erally. Notably, maps of the spatial distribution of postadoles- ultimate density of mature frontal lobe cortical gray matter.
cent cortical gray matter density reduction are highly consistent Key words: MRI; myelination; brain development; synaptic
with maps of the spatial distribution of postadolescent brain pruning; frontal lobe; adolescence

Magnetic resonance imaging (MRI) studies of human brain mat- internal capsule and arcuate fasciculus (Paus et al., 1999). Results
uration during the adolescent years have consistently shown sub- from an previous study by our group have shown dramatic accel-
tle increases in total brain volume along with regionally variable eration in frontal and striatal gray matter density loss to occur
patterns of reductions in gray matter volume and increases in during the postadolescent years along with a stabilization of
total white matter volume (Jernigan et al., 1991; Reiss et al., 1996; regressive gray matter density changes in the parietal lobes (Sow-
Giedd et al., 1999; Sowell et al., 2001b). The spatial and temporal ell et al., 1999a). Maturation, particularly in the frontal lobes, has
distribution of tissue density changes has also been mapped been shown to correlate with measures of cognitive functioning
(Sowell et al., 1999a,b), revealing a pattern of maturational (Reiss et al., 1996; Sowell et al., 2001a).
changes that is consistent with what would be expected given Although there is some evidence from in vivo volumetric stud-
findings from postmortem studies of myelination (Yakovlev and ies to suggest that continued brain growth during the adolescent
Lecours, 1967; Benes et al., 1994) and synaptic pruning (Hutten- years occurs in dorsal brain regions (Jernigan et al., 1991), a
locher, 1979; Huttenlocher and de Courten, 1987). Specifically, a detailed spatial mapping of brain growth has not yet been accom-
reduction in cortical gray matter density has been observed pri- plished in vivo. In the present report, we use novel surface-based
marily in the dorsal parietal and some frontal regions between image analysis methods that allow us to assess relationships
childhood and adolescence (Sowell et al., 1999b) along with an between gray matter density and late brain growth measured at
increase in white matter density in the posterior limb of the the cortical surface. In previous studies, brain anatomy had been
studied with volumetric methods in which inferences could typi-
Received May 11, 2001; revised Aug. 22, 2001; accepted Aug. 23, 2001. cally be made only at the gross lobar level and with voxel-based
This study was supported by National Institute of Mental Health Grant 5T32 methods (Wright et al., 1995) in which relatively crude anatomical
MH16381, by National Science Foundation Grant DBI 9601356, by National Center matching techniques had been used to create statistical maps of
for Research Resources Grant P41 RR13642, by National Institute of Neurological
Disorders and Stroke Grant NS38753, and by the pediatric supplement of the differences between age groups. In the present report, we care-
Human Brain Project, funded jointly by National Institute of Mental Health fully match brain surface anatomy across individuals by defining
(NIMH) and National Institute on Drug Abuse Grant P20 MH/DA52176 to A.W.T. cortical sulcal landmarks on brain surface renderings for each
and NIMH Grant K01 MH01733 to E.R.S. We acknowledge Terry Jernigan for
efforts in collecting the image data used here and for editorial comments on a subject, thereby ensuring accurate localization of group differ-
previous version of this report. ences relative to gyral landmarks. Understanding spatial and
Correspondence should be addressed to Dr. Elizabeth R. Sowell, Laboratory of
Neuro Imaging, University of California Los Angeles, 710 Westwood Plaza, Room
temporal relationships between brain growth on the one hand and
4 –238, Los Angeles, CA 90095-1769. E-mail: [email protected]. tissue density changes on the other hand could help shed light on
Copyright © 2001 Society for Neuroscience 0270-6474/01/218819-11$15.00/0 the biological processes contributing most to the brain maturation
8820 J. Neurosci., November 15, 2001, 21(22):8819–8829 Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter

observed in previous volumetric structural MRI studies. Addi- Image analysts (K .D.T. and E.R.S.) who were blind to subject gender
tionally, understanding the relationships between regional and and age drew each of 17 sulci [sylvian fissure, central sulcus, precentral
sulcus, postcentral sulcus, and superior temporal sulcus (STS) main body,
temporal patterns of brain growth and cortical tissue density STS ascending branch, STS posterior branch, primary intermediate sul-
changes could provide additional insight into patterns of cognitive cus, secondary intermediate sulcus, and inferior temporal, superior fron-
and affective development occurring during adolescence. tal, inferior frontal, intraparietal, transverse occipital, olfactory, occipi-
totemporal, and collateral sulci] in each hemisphere on the surface
MATERIALS AND METHODS rendering of each subject’s brain. These sulci were chosen because they
were the most readily identifiable in all subjects and because they covered
Subjects. Fourteen children (7–11 years; mean age, 9.3 ⫾ 1.3 years; 7 boys most of the brain surface superiorly, inferiorly, and laterally, facilitating
and 7 girls), 11 adolescents (12–16 years; mean age, 13.8 ⫾ 1.6 years; 6 the whole-brain surface analyses. In addition to contouring the major
boys and 5 girls), and 10 young adults (23–30 years; mean age, 25.6 ⫾ 2.0 sulci, a set of six midline landmark curves bordering the longitudinal
years; 5 men and 5 women) were studied with MRI. All child and
fissure were outlined in each hemisphere to establish hemispheric gyral
adolescent subjects were recruited as normal controls for a large, multi-
limits. Spatially registered gray-scale image volumes in coronal, axial,
disciplinary neurodevelopmental research center. Age ranges for the
and sagittal planes were available simultaneously to help disambiguate
child and adolescent groups were chosen because they correspond ap-
brain anatomy. We have developed detailed criteria for delineating the
proximately to prepubertal and pubertal status, although no direct mea-
cortical lines and for starting and stopping points for each sulcus using
sures of hormonal states were collected. All children and adolescents
brain surface atlases as references (Ono et al., 1990; Duvernoy et al.,
were right-handed, and each was screened for neurological impairments
1991). These criteria have been described previously (Sowell et al.,
and for any history of learning disability or developmental delay. In-
2001c) and complete details of the written anatomical protocol can be
formed consent was obtained from all children and adolescents and their
obtained from the authors.
parents. The 10 young adult subjects were recruited as normal controls
Points on the cortical surfaces surrounding and between the sulcal
for neuropsychiatric studies of adult patient populations. These subjects
contours drawn on each individual’s brain surface were calculated using
were all right-handed and were thoroughly screened for medical, neuro-
the averaged sulcal contours as anchors to drive three-dimensional cor-
logical, and psychiatric disorders. Although we did not specifically screen
patients for perinatal complications or prematurity, which can result in tical surface mesh models from each subject into correspondence
brain morphologic abnormalities (Peterson et al., 2000), it is unlikely that (Thompson et al., 2000, 2001). This allows the creation of average
significant perinatal complications were present in our participants, given surface models for the various age groups and the creation of group
that they denied any significant history of neurological or cognitive average maps of various features of the brain surface such as DFC and
abnormalities during screening interviews. Informed consent was ob- gray matter density. The DFC measure was developed primarily to
tained from each subject. All of these subjects have been studied in measure group differences in local growth. It is a measure of radial
previous reports (Sowell et al., 1999a,b). expansion measured from the center of the brain approximately at the
Imag ing protocol. The MRI protocol collected for each subject was a midline decussation of the anterior commissure (i.e., x ⫽ 0; y ⫽ 0; and
whole-brain, gradient-echo (spoiled gradient recalled acquisition in a z ⫽ 0) to each of the 65,536 brain surface points. Note that the measure
steady state) T1-weighted series collected in the sagittal plane with at each point for each individual is reflective of anatomical location in
repetition time of 24 msec, echo time of 5 msec, two excitations, flip angle addition to radial expansion; thus only relative differences in DFC are
of 45°, field of view of 24 cm, 124 slices with section thickness of 1.2 mm, meaningf ul in terms of growth. Given that the deformation maps (ac-
no gaps, and an imaging time of 19 min. quired during cortical surface matching) associate the same cortical
Image anal ysis. MR images from each individual were processed with anatomy in each subject, a local measurement of gray matter density (at
a series of manual and automated procedures that included the following each point over the surface of the brain) could be made for each subject
steps: (1) automated linear transformation (Woods et al., 1993) of the in addition to the DFC measures and averaged across corresponding
images into a standard orientation with scaling to remove global differ- regions of cortex (Sowell et al., 2001c; Thompson et al., 2001). Briefly, a
ences in head size allowing assessment of local changes in brain size or sphere with a radius of 15 mm centered at each cortical surface point was
tissue density; (2) classification of brain images into gray matter, white made and referenced to the same anatomical location in the gray matter
matter, and C SF (Kollokian, 1996; Sowell et al., 1999b); (3) removal of maps for each subject derived previously in the tissue classification. The
nonbrain tissue (i.e., scalp and orbits) and cerebellum from the trans- proportion of segmented gray matter pixels relative to the total number
formed images; (4) automated extraction of the cortical surface for each of pixels in this sphere was computed (at each point) and stored as a map
individual (MacDonald et al., 1994); (5) tracing of 23 sulcal and gyral of gray matter proportion (with values of 0.0 –1.0) for each subject. The
landmarks in each hemisphere on the cortical surface rendering of each proportion of gray matter or gray matter density in each sphere in each
individual (Sowell et al., 2001c); (6) estimating gray matter density or individual is reflective, in part, of local cortical thickness that varies over
local gray matter proportion over the entire cortical surface of each different regions of the brain.
individual’s brain (Thompson et al., 2001); and (7) estimating relative Finally, for some post hoc analyses, brain image volumes and surface
local brain growth measured at each cortical surface point [i.e., the radial renderings were transformed into standard space without scaling using
expansion or distance from the center (DFC) of the brain near the manually selected anatomical landmarks. This was accomplished by
anterior commissure to each cortical surface point (Thompson et al., delineating 80 standardized anatomical landmarks (40 in each hemi-
2001)]. sphere, the first and last points on each of 20 sulcal lines drawn in each
First, brain image volumes were transformed into standard Interna- hemisphere described above) in every individual and using a least
tional Consortium for Brain Mapping (ICBM)-305 space using a 12 squares rigid-body transformation to match each individual to the aver-
parameter linear (with scaling), completely automated image registration age of all individuals in the data set. In this way, every individual’s brain
algorithm (Woods et al., 1993). Semiautomated tissue segmentation was was matched in space, but global differences in DFC remained intact.
conducted for each volume data set to classif y voxels based on signal This step in the analysis was essential to ensure that our observations in
value as most representative of gray matter, white matter, or C SF. A the scaled image space were not attributable to confounds associated
simple minimum distance classifier was used, because it had previously with normalizing brain size differences. (i.e., brain structural differences
been shown to provide the best results (for this T1-weighted imaging attributable to transformation rather than actual brain shape differences
protocol) in a qualitative comparison of different tissue segmentation between groups). Because the transformation into ICBM-305 space was
algorithms. A detailed discussion of the reliability and validity of the linear (for both scaled and nonscaled image transformations), brain
tissue segmentation protocol has been published previously (Sowell et shape was generally preserved, and all images and anatomical delinea-
al., 1999b). tions were easily, automatically transformed between the nonscaled and
Each individual’s cortical surface was extracted using automated soft- scaled image data sets, thus eliminating any need for redundancy in
ware (MacDonald et al., 1994) that creates a spherical mesh surface that detailed anatomical delineations.
is continuously deformed to fit a cortical surface tissue threshold inten- Statistical anal yses. After the basic preprocessing steps were conducted
sity value (signal value that best differentiates cortical C SF on the outer for each individual, statistical maps of differences between age groups
surface of the brain from the underlying cortical gray matter) from the (i.e., child vs adolescent and adolescent vs adult) were created for gray
brain volume aligned in standard ICBM-305 space (Mazziotta et al., matter density and DFC in the scaled image data sets. In these analyses,
1995). The resulting cortical surfaces are represented as a high-resolution the correlation (Pearson’s r) between group membership and gray matter
mesh of 131,072 triangulated elements spanning 65,536 surface points. density or DFC at each brain surface point was calculated for the child
Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter J. Neurosci., November 15, 2001, 21(22):8819–8829 8821

versus adolescent comparison and the adolescent versus adult compari- RESULTS
son. In all statistical maps, a surface point significance threshold of p ⫽
0.05 was used to illustrate local changes in gray matter density or DFC. Regional and temporal patterns of gray matter
To correct for multiple comparisons (i.e., statistical tests at each of density reduction
65,536 surface points), subjects were randomly assigned to groups for Statistical maps for gray matter density differences in the scaled
10,000 new correlational analyses (at each surface point), and the num- image data sets (Fig. 1) between children and adolescents and
ber of significant results (i.e., gray matter density or DFC at any surface between adolescents and adults reveal distinct patterns as ex-
point that significantly differed between groups at the threshold of p ⫽
pected given previous results from our laboratory (Sowell et al.,
0.05) that occurred in the real group difference test was compared with
the null distribution of significant results that occurred by chance in the 1999a,b). Between childhood and adolescence, local gray matter
permutation analyses. In other words, the threshold for assessing signif- density loss is distributed primarily over the dorsal frontal and
icance of statistical maps based on the permutation tests [whole hemi- parietal lobes. Between adolescence and adulthood, a dramatic
spheres or within smaller regions of interest (ROIs)] was determined increase in local gray matter density loss is observed in the frontal
objectively by calculating the surface area (number of surface points) of lobes; parietal gray matter loss is reduced relative to the earlier
significant effects in the real group difference test. That surface area
years; and a relatively small, circumscribed region of local gray
within any tested ROI was used as the threshold for comparison with the
random tests for that ROI, and if ⬍5% (i.e., p ⬍ 0.05) of the results from matter density increase is observed in the left perisylvian region.
random tests reached or exceeded the surface area of the real test, the Again, these results are similar to those observed in the same
statistical map (within ROIs or whole hemispheres) was considered subjects in previous studies in which different methods were used
significant. For the DFC analyses, total left and right hemispheres were to assess gray matter density changes (Sowell et al., 1999a,b).
assessed for overall growth or shrinkage with the permutation tests. Permutation tests confirmed that the overall amount of gray
Additionally, for the DFC group difference maps, an ROI was used to
matter density reduction that occurs in the child to adolescent age
reduce the search area for effects in the permutation analyses. This was
because we had a priori predictions that changes in DFC would be more range does not occur by chance (left hemisphere, p ⫽ 0.007; right
likely to occur in regions of greatest gray matter changes. Because the hemisphere, p ⫽ 0.014). Gray matter density loss over the entire
greatest changes in gray matter density occur in the dorsal frontal cortex brain was highly significant between adolescence and adulthood
between adolescence and adulthood (Sowell et al., 1999a), this was the as well according to the permutation tests (left hemisphere, p ⫽
region that we targeted in the permutation analyses (all gray matter brain 0.0001; right hemisphere, p ⫽ 0.002). Overall gray matter density
tissue anterior to the central sulcus and superior to the axial plane half gain was not significant in either age range according to the
the inferior– superior distance between the intersection of the precentral
sulcus and the posterior extent of the inferior frontal sulcus and the permutation tests. Unlike in our previous reports, here we
intersection of the superior frontal sulcus and the precentral sulcus in mapped differences between the Pearson’s correlation coefficients
each hemisphere). For the gray matter permutation analyses, total left for the two gray matter age–effect comparisons, finally confirm-
and right hemispheres were assessed separately for the overall signifi- ing that there are regions of accelerated gray matter loss in the
cance of gray matter density gain or loss. postadolescent age range, primarily in the dorsal frontal cortices.
To test the statistical significance of the difference between the maps Statistically significant deceleration in gray matter loss is also
for the child versus adolescent comparison and the adolescent versus
adult comparison, maps of the difference between the correlation coef-
observed in various brain regions.
ficients at analogous surface points for the two comparisons for gray The results are also shown in tabular format to help summarize
matter density and the two comparisons for DFC were created using a the locations of significant gray matter density change. We list the
Fisher’s Z transformation (Cohen and Cohen, 1983). number of clusters and the surface area in square millimeters of
The statistical map of the child versus adolescent differences in gray all significant surface points combined separated by hemisphere
matter and the statistical map of the child versus adolescent differences and by lobe (frontal, parietal, and temporo-occipital). Negative
in DFC in the scaled images were overlaid to qualitatively assess the
spatial correspondence between changes with age in the two anatom-
(gray matter density reduction) and positive (gray matter density
ical features (gray matter density and DFC). A Fisher’s Z transforma- increase) effects are shown separately. Table 1 represents gray
tion was used to statistically test the difference between the age – effect matter density changes between childhood and adolescence and
correlation coefficients for scaled gray matter and scaled DFC at each between adolescence and adulthood and corresponds to the sta-
surface point. We predicted that the regions where an inverse rela- tistical maps shown in Figure 1, A and B, respectively. Clusters
tionship existed would appear the most different because correlation are tabulated by lobe and by hemisphere, and negative and
coefficients for gray matter would be most strongly negative (i.e., gray
positive effects are shown separately. As shown, in the left hemi-
matter density loss) in the exact locations where the DFC age – effect
correlation coefficients were most strongly positive (i.e., growth or sphere, gray matter loss is much greater in the frontal lobes
increase in DFC). A similar set of analyses was conducted for the gray between adolescence and adulthood than between childhood and
matter and DFC maps for the adolescent to adult contrasts. To assess adolescence, and left hemisphere parietal lobe gray matter loss is
relationships between changes in gray matter density and changes in much more prominent between childhood and adolescence than
DFC while correcting for multiple comparisons, we created one ROI between adolescence and adulthood. Differences between age
that included all surface points where significant gray matter loss was
groups are much less prominent in the right hemisphere, but as
observed (one for each hemisphere; Fig. 1, top, red reg ions) and one
ROI that included all surface points where positive gray matter cor- shown in the statistical maps in Figure 1 B, there are regions of
relation coefficients were observed (Fig. 1, top, blue to pink reg ions) for accelerated gray matter loss in the right hemisphere as well when
the child to adolescent group comparison. We created similar ROIs group differences are assessed on a point-by-point basis.
for the adolescent to adult contrast and used permutations to test for
positive and negative group differences in DFC within the gray matter Regional and temporal patterns of brain growth
ROIs. Finally, to assess general relationships between DFC and gray For the first time, in this report we show spatial and temporal
matter, a statistical map was created for all 35 subjects in whom gray maps of brain growth and surface contraction between childhood,
matter density at each cortical surface point was assessed for correla- adolescence, and young adulthood. It should be noted that be-
tions with DFC at each corresponding cortical surface point in the
cause the brain surfaces were scaled to remove global size differ-
scaled images.
Post hoc analyses for DFC were also conducted for nonscaled brain ences for these analyses, local brain growth and contraction
image data sets in which we looked at the difference between groups in observed in these maps must be considered relative to global
DFC in millimeters. differences in brain size between groups. Notably, the relative
8822 J. Neurosci., November 15, 2001, 21(22):8819–8829 Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter

Figure 1. Gray matter density age ef-

fect statistical maps (left, right, and top
views) showing gray matter density
changes between childhood and adoles-
cence ( A) and between adolescence and
adulthood ( B). Anatomically, the cen-
tral sulcus and sylvian fissure are shown
in black. Shades of green to yellow rep-
resent negative Pearson’s correlation
coefficients (gray matter loss with in-
creasing age), and shades of blue, purple,
and pink represent positive Pearson’s
correlation coefficients (gray matter gain
with age) according to the color bar on
the right (range of Pearson’s correlation
coefficients from ⫺1.0 to 1.0). Regions
shown in red correspond to correlation
coefficients that have significant nega-
tive age effects at a threshold of p ⫽ 0.05
(gray matter loss), and regions shown in
white correspond to significant positive
age effects at a threshold of p ⫽ 0.05
(gray matter density gain). C, Statistical
map of the Fisher’s Z transformation
of the difference between Pearson’s
correlation coefficients for the child to
adolescent and the adolescent to adult
contrasts (see color bar on right repre-
senting Z scores from ⫺5.0 to 5.0).
Shades of green to yellow represent
regions where the age effects are more
significant in the adolescent to adult
contrast ( B) than in the child to ado-
lescent contrast ( A). Highlighted in
red are the regions where the differ-
ence between Pearson’s correlation
coefficients is statistically significant
( p ⫽ 0.05). Shades of blue, purple, and
pink represent regions where the age
effects are more significant in the child
to adolescent contrast than the adolescent to adult contrast. Highlighted in white are regions where these effects are significant at a threshold of
p ⫽ 0.05.

Table 1. Gray matter density reduction based on statistical maps shown in Figure 1

Positive effects (increase) Negative effects (decrease)

Left Right Left Right
Clusters Area Clusters Area Clusters Area Clusters Area
Child to adolescent gray matter
Frontal 3 162 6 228 51 2481 32 3251
Parietal 2 14 0 0 13 666 18 461
Temporo-occipital 3 4 0 0 13 294 30 280
Adolescent to adult gray matter
Frontal 2 15 2 4 23 4463 23 2832
Parietal 13 205 2 22 12 359 18 425
Temporo-occipital 0 0 0 0 12 980 13 431

The number of clusters (sets of two or more significant surface points; p ⱕ 0.05) and area in square millimeters is presented by hemisphere and by lobe, and positive and
negative effects are shown separately.

maps reveal little local growth (increased DFC) occurring be- surfaces of the children’s brains. When comparing the adoles-
tween childhood and adolescence (Fig. 2) once overall brain size cents with the adults, the permutation test for positive age effects
differences are controlled. Permutation tests for age effects in in DFC (i.e., growth) over the entire brain surface was significant
DFC over the entire brain surface in the scaled image data sets (left hemisphere, p ⫽ 0.030; right hemisphere, p ⫽ 0.028), with
were not significant between childhood and adolescence. It ap- various local regions of significant brain growth contributing to
pears that some local surface contraction or shrinkage (decreased the overall significance of the DFC measure. Permutation tests
DFC) is occurring, however, in various regions over the dorsal for negative age effects (i.e., brain surface contraction) were not
cortex, most prominently in the parietal lobes where the surfaces significant. Notably, during this age range there is some regional
of the adolescents’ brains are closer to the center than the specificity with prominent local growth or increased DFC occur-
Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter J. Neurosci., November 15, 2001, 21(22):8819–8829 8823

Figure 2. DFC age–effect statistical

maps (left, right, and top views) showing
changes in DFC between childhood
and adolescence ( A) and between ado-
lescence and adulthood ( B). Anatomi-
cally, the central sulcus and sylvian fis-
sure are shown in black. Shades of
g reen to yellow represent positi ve
Pearson’s correlation coefficients (in-
creased DFC or brain growth), and
shades of blue, purple, and pink repre-
sent negative Pearson’s correlation co-
efficients (decreased DFC or shrink-
age) according to the color bar on the
right (range of Pearson’s correlation
coefficients from ⫺1.0 to 1.0). Regions
shown in red correspond to correlation
coefficients that have significant posi-
tive age effects at a threshold of p ⫽
0.05 (brain grow th), and regions
shown in white correspond to signifi-
cant negative age effects at a threshold
of p ⫽ 0.05 (brain shrinkage). C, Sta-
tistical map of the Fisher’s Z transfor-
mation of the difference bet ween
Pearson’s correlation coefficients for
the child to adolescent and the adoles-
cent to adult contrasts (see color bar
on right representing Z scores from
⫺5.0 to 5.0). Shades of green to yellow
represent regions where the age effects
are more significant in the adolescent
to adult contrast ( B) than in the child
to adolescent contrast ( A). High-
lighted in red are the regions where
the difference between Pearson’s cor-
relation coefficients is statistically sig-
nificant ( p ⫽ 0.05). Shades of blue,
purple, and pink represent regions
where the age effects are more signif-
icant in the child to adolescent con-
trast than the adolescent to adult contrast. Highlighted in white are regions where these effects are significant at a threshold of p ⫽ 0.05. Note the
sign of the differences between contrasts is opposite to that in the difference map for the gray matter density contrasts because of the inverse
relationship between gray matter density (negative effects) and late brain growth (positive effects).

ring in the dorsal aspects of the frontal lobes bilaterally in the age or stabilization of these processes in the parietal lobes during
same general region where we observed accelerated gray matter the postadolescent years.
density reduction. When a region of interest was used to limit the The results are also shown in tabular format, where we list the
search area of the permutation test to include only dorsal frontal central coordinates in ICBM-305 space, similar to Talairach co-
cortex, the rather prominent continued local growth or increased ordinates (Talairach and Tournoux, 1988) and the area in square
DFC in this region was significant for the right hemisphere ( p ⫽ millimeters of each cluster of significant surface points observed
0.013) but not the left hemisphere ( p ⫽ 0.132). Relative local in the statistical maps. Table 2 illustrates DFC differences be-
growth has occurred in the dorsal frontal cortex such that once tween childhood and adolescence and between adolescence and
overall brain size differences are removed, the surface of the adulthood and corresponds to the statistical maps shown in Fig-
brain lies farther from the center in the adults than in the ure 2, A and B, respectively. Clusters are tabulated by lobe
adolescents. Lateral growth also appears in the inferior, lateral (frontal, parietal, and temporo-occipital) and by hemisphere, and
temporo-occipital junction bilaterally where the brain surface is negative and positive effects are shown separately.
also significantly farther from the center of the brain in the adults
than in the adolescents. Finally, some growth is also observed in Inverse relationship between growth and gray matter
the orbital frontal cortex, more prominent in the left hemisphere. density changes
Local relative shrinkage or decreased DFC is occurring in the In Figure 3, relationships between gray matter density and DFC
lateral aspects of the frontal lobes, perhaps more in the left than changes can be seen in the child versus adolescent and the
the right hemisphere. The difference between correlation coeffi- adolescent versus adult comparisons. Notably, when comparing
cients for the child to adolescent and adolescent to adult compar- the adolescents with the adults, significant gray matter density
isons shown in Figure 2 B confirms accelerated local growth in loss in the frontal lobes is seen almost exclusively in locations
dorsal frontal regions in the older age range and accelerated local where positive age effects for DFC are observed, with very little
growth in the posterior temporo-occipital junction as well. Fi- gray matter loss observed in frontal regions that are not growing
nally, there appears to be acceleration of the local shrinkage in in this age range. In this composite map, the regions of significant
the lateral aspects of the frontal lobes and deceleration of shrink- gray matter loss overlap nearly perfectly onto the regions of
8824 J. Neurosci., November 15, 2001, 21(22):8819–8829 Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter

Table 2. DFC differences between childhood and adolescence and between adolescence and adulthood

Positive effects (growth) Negative effects (shrinkage)

Left hemisphere Right hemisphere Left hemisphere Right hemisphere
x, y, z Area x, y, z Area x, y, z Area x, y, z Area
Childhood to adolescent DFC
Frontal ⫺27, 29, ⫺21 19 56, 30, 17 4 ⫺5, 41, ⫺27 402 5, 27, ⫺28 306
⫺47, 45, ⫺7 41 12, 37, 58 4 ⫺32, 15, ⫺21 59 14, 65, ⫺20 4
⫺49, 31, 26 26 6, ⫺12, 75 148 ⫺2, 66, ⫺1 7 8, 68, ⫺15 14
⫺44, 38, 30 8 54, 12, ⫺5 6 ⫺12, 72, 8 6 17, 71, ⫺6 23
⫺42, 33, 38 18 ⫺4, 67, 8 4 30, 65, ⫺11 4
⫺26, 43, 44 10 ⫺3, 63, 24 4 7, 72, 5 12
⫺40, 27, 47 25 ⫺1, 51, 44 9 19, 72, 3 12
⫺8, ⫺5, 74 65 ⫺3, 46, 50 9 30, 65, 5 13
⫺52, 12, ⫺2 15 ⫺12, 32, 59 14 36, 64, ⫺2 7
⫺21, 23, 62 24 20, 70, 16 22
⫺12, 23, 64 6 4, 67, 18 40
⫺36, 2, 60 19 27, 54, 36 19
12, 47, 50 20
6, 22, 65 11
22, 19, 65 11
42, 1, 59 4
48, ⫺7, 57 23
Total 227 162 563 545
Parietal ⫺8, ⫺38, 76 49 15, ⫺27, 77 3 ⫺40, ⫺51, 59 174 65, ⫺20, 34 6
⫺7, ⫺45, 73 5 59, ⫺7, 46 4 ⫺28, ⫺69, 57 6 41, ⫺44, 63 15
⫺12, ⫺51, 72 9 38, ⫺66, 53 14
Total 63 7 180 35
Temporo-occipital ⫺6, ⫺83, 42 6 20, ⫺81, 49 16 ⫺58, ⫺65, 11 4 69, ⫺27, 1 69
⫺9, ⫺89, 37 3 11, ⫺82, 46 7 ⫺54, ⫺71, 15 13 51, ⫺77, 13 43
⫺7, ⫺97, 26 15 37, ⫺83, 34 5
19, ⫺96, 26 19
11, ⫺99, 21 46
66, ⫺26, ⫺21 19
59, ⫺58, ⫺16 7
43, ⫺80, ⫺17 5
Total 24 124 17 112
Adolescent to adult DFC
Frontal ⫺15, 57, ⫺21 299 9, 67, ⫺15 3 ⫺47, 47, ⫺7 10 30, 56, 30 9
⫺17, 27, ⫺25 3 34, 58, 23 3 ⫺49, 40, 11 4 29, 32, 52 15
⫺14, 62, 29 12 12, 47, 50 9 ⫺26, 62, 21 3 56, 29, 14 31
⫺32, 41, 40 13 33, 43, 41 27 ⫺33, 55, 25 3 58, 20, 23 144
⫺13, 41, 54 13 46, 38, 31 3 ⫺25, 39, 46 33 53, ⫺4, 51 26
⫺30, 38, 45 10 32, 35, 49 24 ⫺40, 40, 33 16 60, ⫺5, 39 29
⫺21, 32, 57 129 19, 35, 57 19 ⫺42, 34, 38 26 64, ⫺4, 18 78
⫺30, 23, 57 4 25, 19, 64 46 ⫺49, 35, 23 20 57, 8, ⫺1 20
⫺7, 17, 69 95 4, 19, 66 196 ⫺53, 31, 14 11
⫺47, 6, 50 7 13, ⫺2, 74 51 ⫺55, 23, 16 34
11, ⫺17, 79 20 ⫺50, 24, 30 5
43, 7, 55 41 ⫺49, 14, 42 58
⫺57, 15, 16 148
⫺55, 11, 1 4
⫺37, ⫺15, 67 3
Total 585 442 378 352
Parietal ⫺39, ⫺52, 59 5 ⫺62, ⫺4, 24 331 17, ⫺37, 77 24
⫺49, ⫺49, 54 4 ⫺59, ⫺6, 39 14 21, ⫺42, 72 21
⫺58, ⫺25, 46 91 43, ⫺51, 57 4
⫺8, ⫺48, 71 9 13, ⫺62, 65 18
⫺19, ⫺58, 67 12 3, ⫺73, 47 274
⫺2, ⫺73, 48 14 14, ⫺86, 41 11
⫺18, ⫺78, 50 9
⫺30, ⫺78, 47 5
Total 9 0 485 352
Temporo-occipital ⫺54, ⫺69, ⫺13 19 55, ⫺2, ⫺37 19 33, ⫺84, 36 24
⫺65, ⫺55, 10 84 69, ⫺46, 5 116 12, ⫺91, 32 39
52, ⫺76, 17 199 10, ⫺97, 22 65
Total 103 334 0 128

The central coordinate (x, y, z) and area in square millimeters is shown for each cluster. The total surface area of significant surface points is presented for each lobe.
Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter J. Neurosci., November 15, 2001, 21(22):8819–8829 8825

Figure 3. A, Composite statistical map (top view) showing the correspondence in age effects for changes in DFC and changes in gray matter in the child
to adolescent contrast. Shown in green is the Pearson’s correlation map of all positive correlation coefficients for DFC (also seen in Fig. 2), and in blue
is the probability map of all regions of significant gray matter loss (surface point significance threshold, p ⫽ 0.05, as shown in Fig. 1). In red are regions
of overlap in the gray and DFC statistical maps. B, Similar composite map for the adolescent to adult age effects. Note the highly spatially consistent
relationship between brain growth and reduction in gray matter density. The shapes of the regions of greatest age-related change for the two maps (gray
matter and DFC) are nearly identical in many frontal regions in the adolescent to adult contrast. Very few regions of gray matter density reduction fall
outside regions of increases in DFC. C, D (left, right, and top views), Difference between Pearson’s correlation coefficients for the age effects for gray
matter density and the age effects for DFC between childhood and adolescence ( C) and between adolescence and adulthood ( D). These maps are similar
to those of the difference between correlation coefficients for age effects of gray matter and DFC shown in Figures 1 and 2 but instead highlight the
correlation between regions of greatest change in the two separate features of brain maturation measured here (DFC and gray matter density). The color
bar represents corresponding Z scores ranging from ⫺5.0 to 5.0 for the difference between correlation coefficients for DFC and gray matter. Highlighted
in red are regions of significant negative correlations between DFC and gray matter density ( p ⫽ 0.05), showing that the relationship between regions
of greatest gray matter density reduction are statistically the same as the regions with the greatest brain growth, particularly in the adolescent to
adulthood years. Highlighted in white are the regions where the difference between correlation coefficients for the gray matter and DFC maps is positive,
indicating that the change with age is in the same direction for both variables (i.e., increased DFC change goes with increased gray matter density
change).

Table 3. DFC permutation analyses within gray matter ROIs

Child to adolescent DFC Adolescent to adult DFC

Left Right Left Right
Positive Negative Positive Negative Positive Negative Positive Negative
Gray matter increase 0.9790 0.0184 0.8940 0.0003 0.8819 0.0140 0.7414 0.0163
Gray matter decrease 0.1112 0.8056 0.0634 0.4570 0.0114 0.9506 0.0177 0.5419

The p values from permutation tests assessing positive (increased) and negative (decreased) age effects on DFC within regions of significant gray matter density decrease and
separate regions of gray matter density increase are shown.

frontal lobe brain growth in the correlation map for DFC. It is gray and DFC maps. A reverse pattern is observed in the left and
interesting to note the correspondence in the distributions of right perisylvian cortex in the comparison between gray matter
these two features of brain development (brain growth and gray changes and DFC changes for the adolescent to adult compari-
matter density reduction) despite their irregular shapes and pat- son. In these regions, significant positive correlations are ob-
terns over the brain surface. Similar effects are observed in the served such that increased gray matter density is statistically
child to adolescent comparison composite map, in which signifi- associated with increased DFC.
cant gray matter loss tends to be seen primarily in regions where New permutation analyses were conducted in which we as-
growth is observed, although these effects are in different regions sessed for group effects in DFC within ROIs created from the
than those in the adolescent to adult age range. The difference gray matter statistical maps (positive and negative). Results from
between correlation coefficients for gray matter and DFC age these analyses are presented in Table 3. In the child to adolescent
effects indicate that the two phenomena are quite closely linked analyses, trend-level significant increases in DFC are observed in
primarily in the frontal lobes in the older age range and more regions of significant gray matter loss. An opposite pattern is
distributed over the frontal and the parietal lobes in the younger observed in regions of gray matter increase, where DFC is sig-
age range, statistically quantifying the similar appearance of the nificantly reduced in the adolescents relative to the children.
8826 J. Neurosci., November 15, 2001, 21(22):8819–8829 Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter

Figure 4. Statistical map of the corre-

lation between gray matter density and
DFC across all subjects studied. Ana-
tomically, the central sulcus and sylvian
fissure are highlighted. Shown in shades
of blue, purple, and pink are regions
where the correlation is positive (i.e.,
greater gray density associated with
greater DFC), and in shades of green to
yellow are regions where the correlation
between DFC and gray matter density is
negative. Highlighted in red are regions
where the negative relationship is highly
statistically significant ( p ⫽ 0.000001). Note that none of the positive correlations between DFC and gray matter density was significant, even when p ⫽
0.01 was used as a threshold.

Significantly increased DFC is observed in the regions of signif- DISCUSSION

icant gray matter loss in the adolescent to adult group compari- In this report, for the first time we have mapped the spatial
son, and significantly decreased DFC is observed in regions of distribution of late brain growth and demonstrate that it does
gray matter gain during the same age range. These results confirm indeed continue in the frontal and posterior temporal lobes
that gray matter density loss is observed in spatial and temporal during the postadolescent years regardless of whether individual
conjunction with regions of brain growth. differences in global brain size are controlled. Interestingly, the
Across the entire age range between 7 and 30 years, DFC and anatomical regions within the frontal lobes where we see the most
gray matter density are highly negatively correlated such that robust accelerated gray matter density loss are in precisely the
those with a cortical surface that shows the most relative growth same locations where we see the most robust continued postado-
also have the least dense gray matter in the dorsal frontal and lescent brain growth. This effect was confirmed with permutation
posterior parietal cortices (Fig. 4). No significant positive corre- tests. The strong correspondence in the age effects for gray matter
lations between DFC and gray matter density were observed density reduction and increased brain growth in frontal cortex
anywhere on the surface of the brain, suggesting that different may provide new insight for making inferences about the cellular
biological processes may be associated with the shrinkage of the processes contributing to postadolescent brain maturation. Re-
brain that occurs during this age range. Notably, growth in the gressive (i.e., synaptic pruning) and progressive (i.e., myelination)
posterior aspects of the temporo-occipital junction is not associ- cellular events are known to occur simultaneously in the brain
ated with gray matter density reduction, again suggesting distinct during childhood, adolescence, and young adulthood, both of
biological phenomena in these regions. which could result in the appearance of gray matter density
Post hoc analyses of brain growth in reduction or cortical thinning on MRI.
nonscaled images A reduction in the number of synapses in the cortex could
Post hoc analyses of brain growth in nonscaled images confirm the result in our observations of reduced gray matter density. On its
local or relative group differences in DFC observed in the scaled own, this process would seem to have to result in a net brain
images (Fig. 5), particularly for the adolescent to adult compari- volume loss (along with an increase in CSF). Notably, however,
son. The nonscaled maps show that continued brain growth does we have now shown local brain growth in the same regions where
occur between adolescence and adulthood in the very dorsal-most gray matter density reduction is occurring rather than brain
aspects of the posterior frontal lobes bilaterally and in the poste- shrinkage. An increase in the amount of myelin could also result
rior inferior temporal lobes bilaterally whether brain size differ- in a reduction in the amount of brain tissue that has a gray matter
ences are controlled or not. These results are quite robust despite appearance on MRI, given that nonmyelinated peripheral axonal
the relatively large interindividual variability in total brain vol- and dendritic fibers do not have normal white matter signal values
ume (Jernigan et al., 1991; Pfefferbaum et al., 1994) that could on T1-weighted MRI (Barkovich et al., 1988). Increased myeli-
potentially obscure results in nonscaled data sets. As shown in nation would seem to necessarily result in a net brain volume
Figure 5, between adolescence and adulthood, large, diffuse re- increase, given that myelin consists of space-occupying glial cells
gions of shrinkage or decreased DFC are observed in frontal and (Friede, 1989). This would be consistent with the new data
parietal regions surrounding the frontal and temporal growth presented here of late growth in frontal cortex concomitant with
areas, probably resulting from large increases in cortical CSF the cortical gray matter density reduction. It is also possible that
known to occur more prominently between adolescence and gray matter density reduction attributable to regressive factors
adulthood (Pfefferbaum et al., 1994). This is in contrast to the and growth is occurring simultaneously such that the late growth
large regions of growth in frontal cortices between childhood and presumably attributable to increased myelination fills in the space
adolescence, with shrinkage occurring only in parietal and infe- vacated by the reduction in synaptic density. Recent animal
rior temporal cortices bilaterally. These regions of growth and research has suggested that increased myelin is associated with
shrinkage were not as prominent in the analyses of scaled image neurite growth-inhibiting factors during critical periods for cor-
data sets when overall differences in brain size were corrected. tical plasticity (Schoop et al., 1997). This close temporal linkage
The analyses of nonscaled images do suggest that much of the between dendritic arborization–synaptic density changes and in-
progressive maturational change that leads to the subtle increase creased myelination could be consistent with our in vivo findings
in total brain size occurs during the years between childhood and of cortical gray matter density reduction spatially concomitant
adolescence, with only relatively subtle growth yet to occur after with late brain growth.
adolescence in the dorsal frontal and posterior temporal cortices. Notably, significant decreases in DFC are also observed in
Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter J. Neurosci., November 15, 2001, 21(22):8819–8829 8827

Figure 5. Differences between groups

in DFC shown in millimeters in color
(according to the color bar) between
childhood and adolescence in both non-
scaled ( A) and scaled ( C) image data
sets. Differences between adolescents
and adults are also shown in nonscaled
( B) and scaled ( D) images. Anatomi-
cally, the central sulcus and sylvian fis-
sure are shown in black. The maps in
the scaled image space allow an assess-
ment of the magnitude (in millimeters)
of differences in DFC shown as statisti-
cal maps in Figure 2. The same color bar
applies to both nonscaled and scaled
images; regions of brain growth be-
tween the younger and older age group
tested are shown in dark blue, purple,
and pink, and regions of shrinkage be-
tween the younger and older groups
tested are shown in red, yellow, green,
and light blue. Note that whether or not
brain size correction is made with scal-
ing, dorsal frontal and posterior tempo-
ral lobes show evidence for continued
growth after adolescence. Other less ro-
bust regions of brain growth or shrink-
age are “scaled” out when brain size
correction is used to control individual
differences.

regions of gray matter density increase whether these effects are tion with increased gray matter density. Again, it is possible that
assessed between childhood and adolescence or between adoles- some relatively complex combination of progressive and regres-
cence and adulthood. Gray matter gain has been reported by sive cellular changes are occurring simultaneously during the
another research group (Giedd et al., 1999), although the age adolescent years, accounting for our observations. Postmortem
range was younger than those reported here. Increased gray and animal studies are needed to best interpret the cellular
matter density could result from increased synaptic density changes that might be associated with gray matter density in-
(Kleim et al., 1996), increases in somal size, or perhaps even new crease during this age range.
cell generation (Gould et al., 1999). Animal studies have long In this report, we show a strong negative correlation between
shown increased dendritic arborization and cortical thickening as brain growth and gray matter density, particularly in the frontal
a result of enriched environmental experience (for review, see and parietal lobes when all subjects between 7 and 30 years are
Diamond, 2001). However, none of these cellular events has been examined. During this age range, the greater the brain growth in
shown in postmortem or animal studies to occur as part of these regions, the less dense the gray matter in the cortex. Note
normative maturation during the adolescent years. It is not clear the regional specificity of this effect given that similar phenomena
why brain shrinkage or decreased DFC would occur in conjunc- are not observed in the perisylvian region or the posterior tem-
8828 J. Neurosci., November 15, 2001, 21(22):8819–8829 Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter

poral and inferior parietal lobes. Together, these results suggest the local diffusion anisotropy, whereas synaptic pruning should
that different factors, perhaps variable combinations of regressive not increase the directional preference of water diffusion.
and progressive cellular events occurring simultaneously, influ- Table 1 shows that gray matter loss at the cortical surface is
ence regional patterns of brain growth and shrinkage at different most prominent in the frontal lobes between childhood and
stages of development. At some point in the developmental adolescence as well as between adolescence and adulthood,
trajectory between birth and death, we would expect the negative somewhat in contrast to our previous report, in which we showed
relationship between brain growth and cortical gray matter den- gray matter density reduction to be most prominent in the pari-
sity reduction to reverse, in which cortex that is reducing in etal lobes between adolescence and adulthood (Sowell et al.,
density because of degenerative changes (i.e., cell death) would 1999b). In that report, we looked at gray matter density through-
result in brain shrinkage (decreased DFC). The changes we out the entire brain to the depths of each sulcus, not just at the
observe here are likely maturational in nature and may be related cortical surface as we have here. We also used anatomical land-
to pubertal and hormonal changes that occur during the adoles- marks on the brain surface to match anatomy across subjects in
cent years. We might expect different factors to affect relation- the present report. Perhaps poorer intersubject image registration
ships between brain size and tissue density at different stages of in the previous report was masking some of the frontal lobe gray
development, but because we did not measure pubertal status in matter density reduction at the cortical surface observed here
our child and adolescent subjects, we are unable to directly between childhood and adolescence. Nonetheless, results re-
measure its potential influence on brain changes. Additional ported here do show a robust increase in gray matter density
studies with older participants and careful assessment of hor- reduction within the frontal lobes and a decrease in parietal lobe
monal status will be needed to address these issues. gray matter density reduction occurring during the postadoles-
Examination of the spatial and temporal patterns of brain cent years, consistent with our previous reports (Sowell et al.,
growth, brain shrinkage, and cortical thinning over time may help 1999a,b).
explain the cognitive and behavioral changes that occur during Using the methods described here, it is possible to find smaller
this age range in addition to helping us further understand regions of gray matter density gain and loss (or growth and
relationships between different cellular maturational events. Sta- shrinkage) within the same larger lobar regions typically mea-
bilization of changes in the parietal cortex appears first, where sured in volumetric studies (Jernigan et al., 1991; Reiss et al.,
1996; Giedd et al., 1999; Sowell et al., 2001b). For example, there
more shrinkage and greater gray matter thinning are occurring
are regions of gray matter density gain (Fig. 1, blue regions in the
between childhood and adolescence than between adolescence
frontal lobe) distributed in close proximity to regions of gray
and adulthood. Later in development, there is brain growth
matter density loss (Fig. 1, green and red regions in the frontal
corresponding to cortical thinning in the dorsal frontal region, the
lobe) over the frontal cortex between childhood and adolescence.
region of the brain known to develop latest in terms of myelina-
Generally the regions of gray matter density gain do not reach
tion and synaptic density (Yakovlev and Lecours, 1967; Hutten-
statistical significance on a point-by-point basis as do the regions
locher and de Courten, 1987). As described in our previous report
of significant gray matter loss, and permutation tests suggest only
of postadolescent changes in gray matter density (Sowell et al.,
significant gray matter density loss during the age range studied.
1999a), it is likely that the visuospatial functions typically asso-
As discussed above, this may be inconsistent with volumetric
ciated with parietal lobes are operating at a more mature level
results from another group, which show gray matter volume
earlier than the executive functions typically associated with
increases in some brain regions between 4 and ⬃12 years (Giedd
frontal brain regions. The new findings described here may sug- et al., 1999). However, in our own volumetric studies with the
gest that cortical thinning or reduction in gray matter density is same children and adolescents studied here, we did not see
first accompanied by growth (as seen the frontal lobes in the evidence for a gray matter volume increase (Sowell et al., 2001b).
postadolescent years) and later by brain shrinkage as the regres- Notably, our youngest subjects were 7 years old, and it is possible
sive changes overtake the progressive changes (as seen earlier on that the most robust gain in gray matter observed by the other
in the parietal lobes). Perhaps gray matter density reduction research group occurred before 7 years, which could account for
associated with growth (presumably increased myelination) is the discrepancy in results. It is also possible that, in sum, the small
associated with different aspects of improved cognitive function- regions of gray matter density gain could overtake the gray matter
ing than the cortical thinning associated with brain surface con- density loss in magnitude to result in a subtle net volume gain.
traction (presumably synaptic pruning). It may not be unreason- This effect would be difficult to measure quantitatively (given that
able to hypothesize that improved accuracy (i.e., improved gray matter is averaged over several millimeters of cortex at a
cognitive task performance) may result from regressive changes time), but it is important to note that volume changes in particular
such as synaptic pruning, given that unused or less efficient tissue compartments within specified ROIs will not necessarily be
synaptic connections are being pruned away during this age range directly reflected in the tissue density measures over the entire
(Huttenlocher, 1979). On the other hand, increased efficiency cortical surface of the same region. It is primarily this increased
(i.e., reduced reaction times) might result from increased myeli- spatial resolution for detecting age effects that motivated the use
nation observed as brain growth, given that myelinated fibers of surface-based rather than the more traditional volumetric
improve the conduction speed of electrical impulses between methods in the present study.
various brain regions. By looking at brain growth and gray matter The results presented in this report of local brain growth and
density at the cortical surface simultaneously, we can test these gray matter density are clearly quite complex and are primarily
hypotheses and parse out the relative contributions of these interpreted in relation to previous findings in the same subjects.
various factors to functional and structural brain maturation. Given the nature of the methods used, with numerous multiple
Future studies should also use diffusion tensor imaging (Pierpaoli comparisons for each statistical map presented, great care must
et al., 1996) as an approach to disentangle increased white matter be taken in interpreting results shown in statistical maps. Permu-
from synaptic loss, because increased myelination would increase tation tests confirm the overall significance of the gray matter and
Sowell et al. • Postadolescent Brain Growth and Frontal Gray Matter J. Neurosci., November 15, 2001, 21(22):8819–8829 8829

DFC statistical maps. To test regional hypotheses, we used per- and matching in magnetic resonance images. In: Proceedings of the
mutation tests within regions of interest based on our a priori SPIE Conference on Visualization in Biomedical Computing (Robb
RA, ed), pp 160 –169. Rochester, MN: SPIE.
hypotheses to ensure that we would not make type I errors in our Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J (1995) A proba-
interpretation. Changes in DFC were predicted within the frontal bilistic atlas of the human brain: theory and rationale for its develop-
lobes, where we saw increased gray matter density reduction in ment. The International Consortium for Brain Mapping (ICBM). Neu-
roImage 2:89 –101.
our previous reports (Sowell et al., 1999a,b). Our results of late Ono M, Kubik S, Abernathey CD (1990) Atlas of the cerebral sulci.
brain growth in the dorsal frontal cortex are also consistent with Stuttgart: Thieme.
Passler MA, Isaac W, Hynd GW (1985) Neuropsychological develop-
what would be expected from the cognitive behavioral literature, ment of behavior attributed to frontal lobe functioning in children. Dev
which shows improved frontal lobe functioning during adoles- Neuropsychol 1:349 –370.
cence (Passler et al., 1985). Notably, regions of late brain growth Paus T, Zijdenbos A, Worsley K, Collins DL, Blumenthal J, Giedd JN,
Rapoport JL, Evans AC (1999) Structural maturation of neural path-
in the lateral posterior temporal lobes were not predicted and are ways in children and adolescents: in vivo study. Science 283:1908 –1911.
interpreted more cautiously. In a previous study, we saw evidence Peterson BS, Vohr B, Staib LH, Cannistraci CJ, Dolberg A, Schneider
for increased sulcal asymmetry in the posterior sylvian fissure, KC, Katz KH, Westerveld M, Sparrow S, Anderson AW, Duncan CC,
Makuch RW, Gore JC, Ment LR (2000) Regional brain volume ab-
which was related to increased gray matter density in the same normalities and long-term cognitive outcome in preterm infants.
region (Sowell et al., 2001c). It is possible that these structural JAMA 284:1939 –1947.
changes are functionally significant and, given their anatomical Pfefferbaum A, Mathalon DH, Sullivan EV, Rawles JM, Zipursky RB,
Lim KO (1994) A quantitative magnetic resonance imaging study of
location, related to improvement in language skills that occurs changes in brain morphology from infancy to late adulthood. Arch
throughout adolescence. Additional studies in new subject sam- Neurol 51:874 – 887.
ples will be required to confirm these results and to assess rela- Pierpaoli C, Jezzard P, Basser PJ, Barnett A, Di Chiro G (1996) Diffu-
sion tensor MR imaging of the human brain. Radiology 201:637– 648.
tionships among cognitive function, late brain growth, increased Reiss AL, Abrams MT, Singer HS, Ross JL, Denckla MB (1996) Brain
sulcal asymmetry, and increased gray matter density in the peri- development, gender and IQ in children. A volumetric imaging study.
sylvian brain region. Brain 119:1763–1774.
Schoop VM, Gardziella S, Muller CM (1997) Critical period-dependent
reduction of the permissiveness of cat visual cortex tissue for neuronal
REFERENCES adhesion and neurite growth. Eur J Neurosci 9:1911–1922.
Barkovich AJ, Kjos BO, Jackson Jr DE, Norman D (1988) Normal Sowell ER, Thompson PM, Holmes CJ, Jernigan TL, Toga AW (1999a)
maturation of the neonatal and infant brain: MR imaging at 1.5 T. In vivo evidence for post-adolescent brain maturation in frontal and
Radiology 166:173–180. striatal regions. Nat Neurosci 2:859 – 861.
Benes FM, Turtle M, Khan Y, Farol P (1994) Myelination of a key relay Sowell ER, Thompson PM, Holmes CJ, Batth R, Jernigan TL, Toga AW
zone in the hippocampal formation occurs in the human brain during (1999b) Localizing age-related changes in brain structure between
childhood, adolescence, and adulthood. Arch Gen Psychiatry childhood and adolescence using statistical parametric mapping. Neu-
51:477– 484. roImage 9:587–597.
Cohen J, Cohen P (1983) Applied multiple regression analysis for the Sowell ER, Delis D, Stiles J, Jernigan TL (2001a) Improved memory
behavioral sciences, Ed 2. London: Erlbaum. functioning and frontal lobe maturation between childhood and ado-
Diamond MC (2001) Response of the brain to enrichment. An Acad lescence: a structural MRI study. J Int Neuropsychol Soc 7:312–322.
Bras Cienc 73:211–220. Sowell ER, Trauner DA, Gamst A, Jernigan TL (2001b) Development
Duvernoy HM, Cabanis EA, Vannson JL (1991) The human brain: of cortical and subcortical brain structures in childhood and adoles-
surface, three-dimensional sectional anatomy and MRI. Vienna: cence: a structural magnetic resonance imaging study. Dev Med Child
Springer. Neurol, in press.
Friede RL (1989) Gross and microscopic development of the central Sowell ER, Thompson PM, Rex D, Kornsand D, Jernigan TL, Toga AW
nervous system. In: Developmental neuropathology, Ed 2 (Friede RL, (2001c) Mapping sulcal pattern asymmetry and local cortical surface
ed), pp 2–20. Berlin: Springer.
Giedd JN, Blumenthal J, Jeffries NO, Castellanos FX, Liu H, Zijdenbos gray matter distribution in vivo: maturation in perisylvian cortices.
A, Paus T, Evans AC, Rapoport JL (1999) Brain development during Cereb Cortex, in press.
childhood and adolescence: a longitudinal MRI study. Nat Neurosci Talairach J, Tournoux P (1988) Co-planar stereotaxic atlas of the human
2:861– 863. brain. New York: Thieme.
Gould E, Reeves AJ, Graziano MS, Gross CG (1999) Neurogenesis in Thompson PM, Woods RP, Mega MS, Toga AW (2000) Mathematical/
the neocortex of adult primates. Science 286:548 –552. computational challenges in creating deformable and probabilistic at-
Huttenlocher PR (1979) Synaptic density in human frontal cortex: de- lases of the human brain. Hum Brain Mapp 9:81–92.
velopmental changes and effects of aging. Brain Res 163:195–205. Thompson PM, Mega MS, Woods RP, Zoumalan CI, Lindshield CJ,
Huttenlocher PR, de Courten C (1987) The development of synapses in Blanton RE, Moussai J, Holmes CJ, Cummings JL, Toga AW (2001)
striate cortex of man. Hum Neurobiol 6:1–9. Cortical change in Alzheimer’s disease detected with a disease-specific
Jernigan TL, Trauner DA, Hesselink JR, Tallal PA (1991) Maturation of population-based brain atlas. Cereb Cortex 11:1–16.
human cerebrum observed in vivo during adolescence. Brain Woods RP, Mazziotta JC, Cherry SR (1993) MRI-PET registration with
114:2037–2049. automated algorithm. J Comput Assist Tomogr 17:536 –546.
Kleim JA, Lussnig E, Schwarz ER, Comery TA, Greenough WT (1996) Wright IC, McGuire PK, Poline JB, Travere JM, Murray RM, Frith CD,
Synaptogenesis and Fos expression in the motor cortex of the adult rat Frackowiak RS, Friston KJ (1995) A voxel-based method for the sta-
after motor skill learning. J Neurosci 16:4529 – 4535. tistical analysis of gray and white matter density applied to schizophre-
Kollokian V (1996) Performance analysis of automatic techniques for nia. NeuroImage 2:244 –252.
tissue classification in magnetic resonance images of human brain. In: Yakovlev PI, Lecours AR (1967) The myelogenetic cycles of regional
Computer science. Masters thesis, Concordia University. maturation of the brain. In: Regional development of the brain in early
MacDonald D, Avis D, Evans A (1994) Multiple surface identification life (Minkowski A, ed), pp 3–70. Oxford: Blackwell.