Localized Enlargement of the Frontal Cortex
in Early Autism
Ruth A. Carper and Eric Courchesne
Background: Evidence from behavioral, imaging, and postmortem studies indicates that the frontal lobe, as well as other brain
regions such as the cerebellum and limbic system, develops abnormally in children with autism. It is not yet clear to what extent the
frontal lobe is affected; that is, whether all regions of frontal cortex show the same signs of structural maldevelopment.
Methods: In the present study, we measured cortical volume in four subregions of the frontal cortex in 2-year-old to 9-year-old boys
with autism and normal control boys.
Results: The dorsolateral region showed a reduced age effect in patients when compared with control subjects, with a predicted 10%
increase in volume from 2 years of age to 9 years of age compared with a predicted 48% increase for control subjects. In a separate
analysis, dorsolateral and medial frontal regions were significantly enlarged in patients aged 2 to 5 years compared with control
subjects of the same age, but the precentral gyrus and orbital cortex were not.
Conclusions: These data indicate regional variation in the degree of frontocortical overgrowth with a possible bias toward later
developing or association areas. Possible mechanisms for these regional differences are discussed.
Key Words: Gray matter, white matter, MRI, orbital cortex, dorso-
the frontal lobe, may develop abnormally in this disorder.
Patients with autism show deficits in joint attention (McEvoy et al
1993), set shifting (Hughes et al 1994; McEvoy et al 1993; Ozonoff
et al 1991), and cognitive planning (Hughes et al 1994), functions
believed to involve areas of dorsolateral prefrontal cortex. Ab-
normalities in motor function are also present (Müller et al 2001;
Teitelbaum et al 1998) and suggest the involvement of motor
regions, possibly primary motor cortex.
Neuroanatomic examinations also support the likelihood of
frontal lobe and other cerebral maldevelopment. Cerebral struc-
ture has now been examined in at least 21 postmortem cases by
seven different labs (Bailey et al 1998; Belichenko et al 1997;
Coleman et al 1985; Fehlow et al 1993; Guerin et al 1996; Kemper
and Bauman 1998; Williams et al 1980). Although the reported
type and location of cerebral abnormality vary from case to case,
several cases have shown defects in the frontal lobe. These have
included mild disruptions of laminar organization (Bailey et al
1998); thickened cortex (Bailey et al 1998); increased cell pack-
ing density, smaller cells, and a “less distinct laminar structure” in
the anterior cingulate (Kemper and Bauman 1998); a minor
malformation of the orbitofrontal cortex (Kemper and Bauman
1998); patches of decreased pyramidal cell density (Belichenko
et al 1997); and reduced dendritic spine density (Williams et al
1980; note that only case 3 in this study fits criteria for autism). In
the imaging literature, two papers have examined cerebral
cortical volume at the lobar level in autism. One found increased
volume throughout the cerebrum that was maximal (13% in-
crease) in the frontal lobe of autistic toddlers (Carper et al 2002).
In contrast, a study of older children and adults (aged 12 to 30
he characteristic symptoms of autism–communication im-
pairments, social deficits, and restricted or repetitive be-
haviors–suggest that association cortex, particularly that of
years) detected enlargement in more posterior lobes but not the
frontal lobe (however, see Discussion for a recent reanalysis)
(Piven et al 1996). Finally, patients with autism also show
metabolic (George et al 1992; Ohnishi et al 2000; Sherman et al
1984; Zilbovicius et al 1995) and electrophysiological (Ciesielski
et al 1990; Dawson et al 1995; Townsend et al 2001) abnormal-
ities in the frontal lobe.
Collectively, these behavioral, metabolic, neurophysiologic,
and neuroanatomic studies indicate that frontal lobe structure is
often abnormal in autism. However, they do not address the
localization of these defects in a systematic way. Behavioral and
neurophysiologic studies do not allow precise localization of
abnormality, and out of necessity, neuropathological studies
generally only sample select areas of the frontal lobe rather than
survey the entire region. Considering the large number of
different cytoarchitectonic regions included in the frontal lobe
(14 Brodmann’s areas), further localization of the abnormality is
necessary to better characterize the neural bases of autism.
Specifically, it is important to determine if the structure of the
entire frontal lobe is affected or if abnormality is restricted to
particular areas, such as association regions or motor regions.
Such localization will help with the development and evaluation
of hypotheses regarding possible causal factors such as abnormal
We used magnetic resonance imaging (MRI) to examine the
volumes of four subregions of the frontal lobe in young children
with autism and in young normal control subjects. Neuroana-
tomic landmarks were used to designate the boundaries of the
regions, which were the precentral gyrus (PCG), dorsolateral
prefrontal cortex (DFC), orbitofrontal cortex (OFC), and the
medial frontal cortex (MFC).
Methods and Materials
Parents of all subjects gave written informed consent for their
child’s participation. Experimental procedures were approved by
the Institutional Review Board of the San Diego Children’s
Hospital Research Center. All patients and subjects were paid for
Patients with Autism
Twenty-five male patients with autism, aged 2.7 to 9.0 years
(mean ? SD: 5.3 ? 1.6 years), were examined. Cerebral and
frontal lobe volumes for all of these were included in previous
From the Center for Autism Research, Children’s Hospital Research Center,
and Neurosciences Department, University of California at San Diego,
San Diego, California.
Jolla, CA 92037; E-mail: email@example.com.
BIOL PSYCHIATRY 2005;57:126–133
© 2005 Society of Biological Psychiatry
reports (Carper and Courchesne 2000; Carper et al 2002;
Courchesne et al 2001). Frontal measures for eight subjects were
also included as part of a report on possible neuroanatomic
contributions to orienting deficits in children with autism (Harris
et al 1999).
All subjects were assessed by a
trained psychologist and met criteria for the diagnosis of autism
according to all of the following: DSM-IV (American Psychiatric
Association 1994), Childhood Autism Rating Scale (CARS)
(Schopler et al 1988), Autism Diagnostic Interview-Revised
(ADI-R) (Lord et al 1994), and Autism Diagnostic Observation
Schedule (ADOS) (Lord et al 1999) (Table 1). All subjects who
were scanned prior to the age of 5 years met clinical criteria at
that time and were also given a second diagnostic evaluation by
Dr. Cathy Lord (an expert in the diagnosis of autism who was
blind to the MRI measures) when they reached 5 years of age or
older. These patients were included only if they met all of the
above criteria after the age of 5. Patients diagnosed with perva-
sive developmental disorders other than autistic disorder were
excluded. A complete neurological exam was given, including
electroencephalogram (EEG) and brain stem auditory evoked
response (BAER) testing. All who met diagnostic criteria were
negative for Fragile X syndrome. Five of the patients showed
seizurelike activity on EEG, although only one of these had
known seizures (7-year-old with brief tonic-clonic episodes).
That individual had been treated with Tegretol (carbamazepine)
for approximately 8 months prior to imaging and behavioral
Intelligence Estimates. Subjects were administered one or
more standardized tests of intelligence, depending on the child’s
level of cognitive functioning and cooperation. These included
the Arthur adaptation of the Leiter International Performance
Scale (Arthur 1980), the Stanford Binet Intelligence Scale (SBIS)
(Thorndike et al 1986), and the Wechsler Intelligence Scale for
Children, Third Edition (WISC-III) (Wechsler 1991). Subjects
were also administered the Peabody Picture Vocabulary Test-
Revised (PPVT-R) (Dunn and Dunn 1981), a measure of recep-
tive language ability. Nearly all of the subjects performed better
on nonverbal portions of the tests than on the verbal portions,
which is typical of patients with autism (Lincoln et al 1995).
Because of this, the child’s highest score from among the Leiter
International Performance Scale, WISC-III performance intelli-
gence quotient (IQ), or Stanford Binet Abstract Reasoning test
was used for intelligence estimates.
Normal Control Subjects
Eighteen normal healthy male control subjects, aged 2.2 to 8.7
years (mean ? SD: 5.1 ? 1.8 years), were examined. Cerebral
and frontal lobe volumes for all were included in previous
reports (Carper et al 2002; Courchesne et al 2001) and frontal
lobe measures for all but four were reported in a study of
correlations between frontal lobe size and cerebellar size in
autism (Carper and Courchesne 2000).
Control subjects were recruited through advertisements in the
community and showed no evidence of developmental, educa-
tional, medical, or psychiatric abnormalities on a pre-MRI screen-
Intelligence Estimates. Control subjects were administered
the PPVT-R and either the SBIS or the WISC-III, depending on
their age at the time of testing. Nonverbal scores are shown in
Imaging and Image Processing
Autistic patients were anesthetized with propofol by a li-
censed, board certified anesthesiologist prior to scanning. Con-
trol subjects were typically scanned during normal sleep, al-
though some remained awake during scanning. All subjects were
scanned on the same 1.5-T GE MRI scanner (Signa, General
Electric, Milwaukee, Wisconsin) using a double-echo, proton
density (PD) and T2-weighted axial protocol (repetition time
[TR] ? 3000 milliseconds, echo time [TE] ? 30 and 80 millisec-
onds, 1 number of excitations [NEX], field of view [FOV] ? 20 cm,
matrix ? 256 x 256, 3 mm slices, no gaps). Data were transferred
to Silicon Graphics workstations (Mountain View, California) for
analysis. Image sets from both subject groups were coded and
intermixed to ensure experimenter blindness to group.
Axial image sets were processed using an automated tissue
classification program (SEGMENT) designed in our laboratory.
The algorithms were similar to those described by other re-
searchers in the semiautomated segmentation of nearly identical
PD/T2 imaging protocols (Jackson et al 1994; Matsumae et al
1996). Skull and extracranial structures were removed from the
T2-weighted images using a combination of thresholding and
manual tracing. These images were then used as a mask on the
tissue-classified images to create a data set containing only
intracranial gray matter, white matter, and cerebrospinal fluid
(CSF). Additional details regarding these algorithms and their
validation are given in Courchesne et al (2000).
The volumes of individual brain structures were determined
using a combination of manual tracing and computer algorithms.
The software AREA (developed in our laboratory) allows the user
to refer to the T2, PD, and tissue-classified axial images while
tracing a structure, thereby maximizing the anatomical informa-
tion available. The programs VoxelMath and VoxelView (Vital
Images, Inc., Minneapolis, Minnesota) were used to create
three-dimensional (3-D) reconstructions of the brain surface
from the T2-weighted images, thereby allowing identification of
surface landmarks. VoxelMath allows mathematical processing
of images to maximize visualization of surface landmarks.
VoxelView automatically displays landmarks and manual trac-
ings in all orthogonal slice planes. Finally, LobeWorks (devel-
Table 1. Subject Characteristics
n ? 25
n ? 18
Age at MR Scan (years)
Mentally Retarded (n)
ADI Scoresb: Social
Verbal subjects (n ? 13)
Nonverbal subjects (n ? 11)
5.24 ? 1.63
41.26 ? 4.26
24.38 ? 3.56
17.92 ? 3.71
5.07 ? 1.81
12.45 ? 1.37—
7.75 ? 1.92
79.14 ? 22.01
112.76 ? 14.35
IQ, intelligence quotient; MR, magnetic resonance.
aCARS scores were not available for two subjects.
bADI scores were not available for one subject.
cThree autistic subjects were unable to complete the nonverbal IQ test.
R.A. Carper and E. Courchesne
BIOL PSYCHIATRY 2005;57:126–133 127
oped in our laboratory) translates landmarks made in VoxelView
into the AREA software for calculation of regional volumes.
All steps of volume measurement–including 3-D rendering,
landmarking, manual tracing, and evaluation of automated pro-
cesses–were performed by experienced neuroanatomists using
neuroanatomical atlases (Carpenter 1976; Duvernoy 1988; Duver-
noy 1991; Mai et al 1997; Nieuwenhuys et al 1988; Ono et al 1990).
Five cases were recoded and measured a second time.
Intraclass correlations were .88 for OFC, .99 for MFC, .98 for DFC,
and .99 for PCG. Maximum absolute volume differences were
OFC ? 6%, MFC ? 4%, DFC ? 3%, and PCG ? 2%.
Frontal Cortex Subregions
Overview. The boundary of the frontal lobe was delimited
using the method described in Carper and Courchesne ( 2000),
which uses the central sulcus as the primary posterior boundary
of the frontal lobe. After the entire frontal lobe was identified in
this way, the frontal gray matter was further divided into four
subregions per side by one of the authors (R.A.C.) based on the
method described by Semendeferi et al (1997). The subregions
were orbitofrontal cortex, medial frontal cortex, precentral gyrus,
and the remaining dorsolateral convexity. White matter was not
subdivided, as it was considered unlikely that a reliable method
could be devised. Boundary sulci were identified by viewing the
3-D surface display and sagittal, axial, and coronal slice displays
in VoxelView (Figure 1). Note that the processing used to maxi-
mize surface image quality also reverses pixel intensity so that white
matter appears brighter than gray matter on the T2 images and
reduces the apparent thickness of gray matter (Figure 1). These
processed images are used only for identification of boundary
sulci. Once these were identified, their full extent was traced
manually on the original T2 axial slices.
Separating Orbitofrontal Cortex from Dorsal Convexity.
The orbital cortex was separated from the dorsal region by a
boundary passing through the frontomarginal sulcus (FMS) and
the lateral orbital sulcus (LOS, alternatively named the fronto-
orbital sulcus by Ono et al 1990). The FMS is easily identified in
coronal slices, while the LOS is best visualized sagittally. If the
FMS and LOS did not connect directly, the surface endpoints of
the two sulci were connected by a straight line, which was then
used to mark boundary points on the axial slices. Similarly, if the
LOS connected directly with the Sylvian fissure or with the
horizontal ramus of the Sylvian fissure, the boundary was
complete. Otherwise, a straight line was drawn from the surface
endpoint of the LOS to the Sylvian fissure at the vertex of the
inferior frontal gyrus pars orbitalis. This line was then used to
mark boundary points on the axial slices. If the LOS was entirely
absent, then the lateral most endpoint of the FMS rather than the
LOS was connected to the Sylvian fissure.
Separating Precentral Gyrus from Dorsal Convexity. The
precentral gyrus was separated from the remainder of the dorsal
convexity by a boundary passing through the precentral sulcus
(PCS). The PCS generally has two or more separate segments.
Since these segments frequently intersect with the superior and
inferior frontal sulci, only branches that traversed the brain
surface in a primarily inferior to superior direction, rather than a
posterior to anterior direction, were used. If two segments of the
PCS could be visualized on a single axial slice, the boundary line
was drawn through both segments so that the posterior aspect of
each was classified with the precentral gyrus region of interest
Figure 1. Parcellation of frontal cortex. The four subregions of frontal cortex are identified on processed T2 images. Upper panels illustrate the 3-D surface
renderings used for identification of landmarks; lower panels illustrate single slice views. Image processing reverses pixel intensity so that white matter
(See text for details). FMS, frontomarginal sulcus; LOS, lateral orbital sulcus; PCS, precentral sulcus; CS, central sulcus; OFC, orbitofrontal cortex; DFC,
dorsolateral prefrontal cortex; PCG, precentral gyrus; MFC, medial frontal cortex.
128 BIOL PSYCHIATRY 2005;57:126–133
R.A. Carper and E. Courchesne
(ROI), while the anterior aspects remained with the dorsal frontal
axial image, a line was drawn through the white matter to
separate the medial surface from the other ROIs. At the anterior
and posterior limits of the brain, a straight line was drawn
through the gray matter. In the most inferior slices, the boundary
passed through the gyrus rectus; that is, through the white matter
medial to the olfactory sulcus.
Statistical analyses were performed using the SPSS software
package (v. 10.0 for the Macintosh; SPSS Inc., Chicago,
Illinois). The reader will note that we did not adjust for overall
brain size or cerebral size by using a ratio measure, as is
sometimes done in brain volumetry studies. Such methods
apply a single linear transformation to each regional volume,
but there is no empirical basis on which to assume that
variability in overall volume would affect each region within
that volume in the same way. In addition, at least one group
(Arndt et al 1991) has demonstrated that the use of ratios or
proportional volumes can reduce overall reliability by com-
bining multiple sources of error, i.e., those of the numerator
and of the denominator. A similar argument can be made
against using an overarching volume as a covariant in assess-
ment of smaller regions within that volume. By including all
the regions of interest in a within-subjects analysis of variance
(ANOVA) design, we are able to assess the regional specificity
of enlargement, the primary goal of ratio and proportion
measures, without compounding measurement error.
Growth of Frontal Cortical Regions. The relationships be-
tween ROI volumes and subject age were compared between
the two subject groups using multiple regression analyses.
Bilateral volumes were used in these first analyses to simplify;
laterality was included as a variable in a separate set of
analyses described below. For each volumetric anatomic
measure, five independent variables were included in a
multiple regression analysis–linear age, quadratic age, subject
group (effect coded), and interactions of study group with
linear age and with quadratic age. Significant results for either
of the interaction variables would indicate that the two subject
groups showed different patterns of volume change across
age. Use of effect coding with multiple regression to examine
relationships in continuous data are described in Cohen and
Cohen (1983) and Darlington (1990).
Comparing Frontal Cortical Regions Between Groups. Re-
peated measures ANOVA was used to compare regional volumes
between the two subject groups in more restricted age ranges.
Both groups were separated into two equal age bins by a
cutpoint at 5.0 years. This age cutpoint was chosen because it
separated subjects into equal-sized bins. This minimized the
age-related variability within groups, thereby increasing statisti-
cal power. This cutpoint differs from the 4.0-year cutpoint used
in our previous report. There is no single exact age at which the
rate of brain growth suddenly changes for all individuals (see
Figure 2). Therefore, the age cutpoint is somewhat arbitrary.
With the current sample of subjects, a cutpoint of 5.0 years
provided the largest possible sample size in each age bin. Mean
values of age and ROI volume for each of these age bins are
shown in Table 2.
The first analysis for each age bin was a three-factor repeated
measures ANOVA with ROI (DFC, OFC, MFC, PCG) and hemi-
sphere (left, right) as repeated measures and subject group as the
between-subjects variable. If this first analysis did not show
effects of laterality (i.e., an interaction between subject group
and hemisphere), volumes for the left and right were added
together and a second ANOVA was performed using the bilateral
volumes. This was a group by ROI analysis. As would be
expected, the ROI variable showed significant main effects in all
comparisons (e.g., DFC volume is 2.5 to 3 times larger than OFC,
so that there is always a main effect of ROI); therefore, main
effects of ROI are not described. Results listed below used the
Huynh-Feldt adjustment (Box adjustment) to degrees of freedom
to correct for failures of sphericity.
Growth of Frontal Cortical Regions
Regression analyses were significant for the DFC [R ? .539;
F(5,37) ? 3.03; p ? .02] and MFC [R ? .570; F(5,37) ? 3.56; p ?
.01] but not for the other ROIs. Both of these regions showed
significant linear relationships with age (DFC: contribution to
R2? .18; p ? .004; MFC: contribution to R2? .26; p ? .001), but
only DFC showed a significant difference in age effects between
groups (i.e., a group by linear age interaction: contribution to
R2? .08; p ? .04). In normal control subjects, a DFC volume
increase of 48 % was predicted from 2 years of age to 9 years of
age, but only a 10% increase was predicted in autistic patients
(Figure 2). As seen in Figure 2, before age 5, DFC volumes for the
two subject groups are strikingly different with very little overlap
between the two groups. However, because of the different age
effects between the two groups, volumes are fairly similar after
about 5 years of age. We performed additional tests to assess
whether similar effects might be present, but more subtle, in the
other regions of interest. In other words, perhaps the other
regions also show enlargement in the youngest subjects but not
of sufficient magnitude to be detected in a regression analysis
across a broad age range.
Comparing Frontal Cortical Regions Between Groups
Subjects Aged 2.2 to 5.0 Years. Three-way ANOVA for the
youngest subjects showed a significant main effect of hemi-
sphere [right ? left; F(1,19) ? 4.29; p ? .05] but no significant
Figure 2. DFC volume as a function of subject age. Autism patients with
mental retardation are indicated by an X, all other autism patients are
for all of these subjects with autism. Control subjects are indicated by a
closed circle and a dashed line. DFC, dorsolateral prefrontal cortex.
R.A. Carper and E. Courchesne
BIOL PSYCHIATRY 2005;57:126–133 129
interactions between hemisphere and any other variable. A
two-way ANOVA of bilateral volumes revealed a significant
group by ROI interaction [F(3,57) ? 6.89; p ? .003] and a
significant main effect of group [F(1,19) ? 8.54; p ? .009],
suggesting that the ROIs were differentially affected in autism.
T tests showed that patients with autism had significantly larger
DFC (t ? 3.22; p ? .005) and also MFC volumes (t ? 2.66;
p ? .02) but did not differ significantly on OFC or PCG volumes
Finally, to determine if DFC and MFC volumes were differen-
tially affected, the two regions were compared directly. Since the
range of volumes differs between these two regions regardless of
subject group, volumes were first transformed into z scores
based on the mean values and standard deviations of the control
subjects. A t test performed in the autism group then revealed a
significant difference (t ? 2.25; p ? .05), indicating greater
abnormal enlargement of DFC than of MFC (Figure 3a).
Subjects Aged 5 to 9 Years. Three-way ANOVA for the older
children did not reveal significant main effects or interactions
related to hemisphere. Subsequent two-way ANOVA of bilateral
volumes showed neither a main effect of group nor a group by
ROI interaction. Additional analyses were not performed (Figure
Removal of Mentally Retarded Patients and Patients with
Many patients with autism have IQs in the mentally retarded
range. Some researchers prefer to restrict their analyses to either
mentally retarded or nonretarded autism patients to simplify. We
therefore repeated our comparisons with the mentally retarded
subjects (n ? 6) removed to make our results more easily
comparable to a broader range of other studies. (There were not
enough mentally retarded patients to examine this group sepa-
rately.) Age regressions were significant for OFC, MFC, and DFC
[F(5,31) ? 2.85; p ? .03], but none of these regions showed
significant interactions between group and age; that is, there was
no longer evidence of groupwise differences in DFC age effects.
However, four of the six retarded patients were over 6.5 years of
age, so their removal substantially reduced the group represen-
tation at the older end of the age range. This would make it more
difficult to achieve a significant regression. Thus, the absence of
a significant groupwise difference may be due to the decreased
sample size rather than a true lack of effect.
Results of groupwise comparisons were very similar to those
described for the full subject sample. Among subjects under age 5
years, there was a significant group by ROI interaction [F(3,51) ?
4.96; p ? .01] and a significant main effect of group [F(1,17) ? 5.74;
p ? .03]. T tests showed that patients with autism had significantly
larger DFC (t ? 2.79; p ? .01) and also MFC volumes (t ?
2.14; p ? .05) but did not differ significantly on OFC or PCG
volumes. However, direct comparison of DFC and MFC did not
show significant differences. In patients older than 5 years, two-way
ANOVA did not reveal any significant results related to group,
although all mean volumes were smaller than normal.
A similar repetition of analyses was performed after eliminat-
ing patients with seizure disorder or seizure activity on EEG (n ?
5). Results were very similar to those derived previously. Age-
based regression analyses were significant for MFC and DFC
regions [F(5,32) ? 2.59; p ? .05]. The MFC did not show any
group-related effects, while the DFC now showed only a trend
toward a group by age interaction (contribution to R2? .078;
p ? .07). In groupwise analyses of subjects under age 5, there
was a significant group by ROI interaction [F(3,45) ? 3.53; p ?
.05] and a significant main effect of group [F(1,15) ? 5.70; p ?
.03]. T tests again showed that patients with autism had signifi-
cantly larger DFC (t ? 2.35; p? .03) and also MFC volumes
(t ? 2.38; p? .03) but did not differ significantly on OFC or
PCG volumes. However, there was no significant difference in
the degree of abnormality between DFC and MFC. In patients
older than 5 years of age, two-way ANOVA did not reveal any
significant results related to group.
The present analyses demonstrated that DFC volume in-
creased less with age in the autism group than in normal control
subjects. This effect was not detected in the other frontal lobe
regions. In addition, the DFC and MFC regions of the frontal lobe
are larger than normal in young children with autism (those
under age 5). The age effect was not seen when subjects with
mental retardation were removed, but this was likely due to the
decreased statistical power; all other findings remained the same.
Table 2. Mean Volumes and Standard Deviations for Regions of Interest
Age ? 5 YearsAge ? 5 Years
(n ? 12)
(n ? 9)
(n ? 13)
(n ? 9)
DFC, dorsolateral prefrontal cortex; MFC, medial frontal cortex; OFC, orbital frontal cortex; PCG, precentral gyrus.
130 BIOL PSYCHIATRY 2005;57:126–133
R.A. Carper and E. Courchesne