COGNITIVE NEUROSCIENCE, 2010, 1 (4), 277288

Enhanced brain connectivity in math-gifted adolescents:

An fMRI study using mental rotation
PCNS

James Prescott1, Maria Gavrilescu2, Ross Cunnington3,

Michael W. OBoyle4, and Gary F. Egan5

Enhanced Connectivity

Howard Florey Institute, University of Melbourne, Victoria, Australia

2
Mental Health Research Institute of Victoria, Melbourne, Australia
3
Queensland Brain Institute, University of Queensland, Brisbane, Australia
4Department of Human Development and Family Studies, Texas Tech University, Lubbock, TX, USA
5Center for Neuroscience, University of Melbourne, Victoria, Australia

Mathematical giftedness is a form of intelligence related to enhanced mathematical reasoning that can be tested
using a variety of numerical and spatial tasks. A number of neurobiological mechanisms related to exceptional
mathematical reasoning ability have been postulated, including enhanced brain connectivity. We aimed to further
investigate this possibility by comparing a group of mathematically gifted adolescents with an average math ability
control group performing mental rotation of complex three-dimensional block figures. Functional magnetic resonance imaging (fMRI) data were collected and differences in intrahemispheric and interhemispheric connectivity
between the groups were assessed using structural equation modeling (SEM). The math-gifted showed heightened
intrahemispheric frontoparietal connectivity, as well as enhanced interhemispheric frontal connectivity between
the dorsolateral prefrontal and premotor cortex. These enhanced connectivity patterns are consistent with previous
studies linking increased activation of the frontal and parietal regions with high fluid intelligence, and may be a
unique neural characteristic of the mathematically gifted brain.

Keywords: Math gifted; Mental rotation; fMRI; Brain connectivity.

INTRODUCTION
Is high intelligence attributed to enhanced cognitive
engagement of the frontal regions of the brain? A
number of studies have reported that individuals with
high fluid intelligence show stronger recruitment of
the anterior cingulate, lateral prefrontal cortex, and
associated anterior/frontal brain structures during
mental processing (Duncan, 2003; Geake & Hansen,
2005; Gray, Chabris, & Braver, 2003). Additionally,
Dempster (1991) concluded that the frontal regions
mediate inhibitory control, which is crucial to the
development of high intelligence. And in their review
of brain imaging and related neuropsychological
research, Kane and Engle (2002) conclude that the

prefrontal cortex plays an essential executive attentional role in a variety of higher-order cognitive
domains. These views are reinforced by Pesenti et al.
(2001), who reported that calculation in a mathematical
prodigy (adult) was related to highly efficient episodic
memory encoding and retrieval, and was primarily
mediated by prefrontal areas in connection with the
anterior cingulate. Jung and Haier (2007), in a metaanalysis of the neuroimaging literature, theorized that
significant parietofrontal integration of the brain
(their P-FIT model) is a key component of the biology
of high intelligence.
Mathematical giftedness is a special form of intelligence pertaining to a highly developed capacity for
mathematical reasoning (though not necessarily

Correspondence should be addressed to: Michael W. OBoyle, Department of Human Development and Family Studies, Texas Tech
University, Lubbock, TX 79424-1230, USA. E-mail: michael.oboyle@ttu.edu

2010 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business
www.psypress.com/cognitiveneuroscience
DOI: 10.1080/17588928.2010.506951

278

PRESCOTT ET AL.

arithmetic calculation per se), and is often exhibited

at an early age, and typically emerges in the absence
of formal training or mathematical instruction (Benbow,
1988). Several characteristics have been identified
that distinguish math-gifted (MG) individuals from
those of average math ability. For example, Benbow
(1986) found that extremely math-precocious youth,
compared to their parents, siblings, and an averageability control group, exhibited more than double the
frequency of left-handedness, allergies, and immune
disorders, as well as myopia. MG individuals are also
far more likely to be male than female, with a ratio of
nearly 13/1 at the highest end of the math reasoning
ability spectrum (Benbow, 1988).
As suggested by Geschwind and Behan (1982),
each of the above characteristics may be a reflection
of enhanced right (as compared to left) hemisphere
development, and may result in heightened spatial
reasoning abilities in those who are MG. In their
model, the abovementioned brain pattern is hypothesized to be a by-product of increased prenatal testosterone exposure and/or sensitivity during the second
trimester of fetal development (see Geschwind &
Galaburda, 1984 for details concerning the impact of
fetal testosterone on the brain and the subsequent
development of cerebral dominance). Notably, it has
recently been proposed that this enhanced development of the right hemisphere leads the MG to rely on
primarily spatial capacities when processing information, and may bias them towards the use of mental
imagery as a preferred mode of mental representation,
each of which may be related to (or perhaps they are
the bases of) their precocious mathematical reasoning
ability (OBoyle, 2008).
Studies investigating exceptional math ability have
postulated a number of neurobiological mechanisms
related to this cognitive characteristic. These mechanisms include enhanced intrahemispheric processing
of the right hemisphere and heightened bilateral
engagement of the cerebral hemispheres (Butterworth,
1999; Dehaene, 1997; OBoyle et al., 1995). OBoyle
and colleagues, using a variety of behavioral measures, reported enhanced right-hemisphere involvement and heightened interhemispheric cooperation in
MG adolescents during dichotic listening, free-vision
chimeric face processing, and the performance of a
variety of concurrent finger-tapping tasks (OBoyle &
Benbow, 1990; OBoyle, Gill, Benbow, & Alexander,
1994). Furthermore, OBoyle et al. (1991), using
electroencephalography, demonstrated increased
right-hemisphere activation and coordinated interhemispheric switching of activated regions in the MG
during chimeric face processing, each of which supports the notion that the MG are better able to access

and integrate the cortical resources of the right

cerebral hemisphere. The MG also exhibit the capacity
to exchange information between the hemispheres in
a highly efficient manner without the usual processing
penalties or costs (OBoyle & Gill, 1998). The relative importance of coordinating neural activity interhemispherically has led to the suggestion that
enhanced connectivity of the hemispheres may contribute to the special attributes of the MG (OBoyle,
Benbow, & Alexander, 1995). Recently, this hypothesis of heightened connectivity as a key characteristic
of the MG brain has undergone further exploration
using both behavioral (Singh & OBoyle, 2004) and
functional magnetic resonance imaging (fMRI) methods (OBoyle et al., 2005), and received support.
Mathematical ability can be tested by assessing
performance on spatial as well as quantitative/numerical reasoning tasks. Mental rotation is one spatial
task, though not inherently numerical or quantitative
in nature, that is thought to correlate highly with
mathematical reasoning ability (Battista, 1990; Benbow,
1988; Hegarty & Kozhevikov, 1999; McGee, 1979;
Reuhkala, 2001). An early review by Corballis (1997)
of hemispheric specialization in mental rotation studies concluded that the right hemisphere performs
mental rotation more efficiently than the left hemisphere. Furthermore, Corballis suggests that bilateralism increases with task demands and that integration
between the hemispheres is essential for successfully
performing complicated spatial tasks, including mental rotation. This viewpoint is supported by recent
functional imaging studies involving mental rotation
of various types of stimuli. For example, at the neural
level the mental rotation of Shepard and Metzler
(1971) 3D objects has been previously shown to activate a bilateral frontoparietal network with right
hemispheric dominance (Cohen et al., 1996; Gill,
OBoyle, & Hathaway, 1998; Kosslyn, DiGirolamo,
Thompson, & Alpert, 1998; Tagaris et al., 1996,
1997). In light of such findings, the mental rotation of
complex 3D objects seems particularly well-suited for
assessing both intrahemispheric and interhemispheric
connectivity in the brain of MG adolescents.
To formally quantify potential differences in neural connectivity between the MG and an age-matched
average math ability control group, a network modeling approach was adopted. The specific technique
employed was structural equation modeling (SEM).
SEM was first applied to human neuroimaging data
by McIntosh and Gonzalez-Lima (1994) and since has
had numerous applications, including studies by
Goncalves and Hall (2003) modeling the visual areas
of the brain, by Rowe et al. ( 2002), who developed a
model for frontal lobe function using SEM, and by

ENHANCED CONNECTIVITY

Kondo et al. (2004), who investigated connectivity in

spatial working memory. Buchel and Friston (1997)
have also published a classic study investigating neural connectivity within and between the ventral and
dorsal streams of the brain during visual processing.
Our SEM analysis is based upon an a priori
anatomical network model. To investigate intrahemispheric and interhemispheric connectivity, we
developed a mental rotation network consisting of
nine brain regions (nodes) including the dorsolateral
prefrontal cortex, premotor cortex, inferior and superior parietal regions bilaterally, and the midline anterior cingulate region. This cortical model was
developed in light of the information processing
stages required for 3D mental rotation as proposed by
Cooper and Shepherd (1973) and developed by Gill et al.
(1998), and involves several component cognitive
processes deemed essential for successful completion
of this task. These substages of processing include:
(1) stimulus registration and encoding; (2) determination of object identity and orientation; (3) imaging and
rotation of the object in the minds eye; (4) comparison of the mentally rotated object with the reference
object; and (5) subsequent response selection and
execution. Note that our network model incorporates
several neural assumptions about such processing,
including the notions that registration and encoding of
the input into spatial working memory primarily
involves the occipital and parietal lobes; that the
imagined rotation of the input arises from the neural
linkage between the frontal and parietal regions; and
that decision and response execution processes are
mediated by the frontal lobes in conjunction with the
anterior cingulate and motor/premotor cortex (Gill et al.,
1998; Jordan, Heinze, Lutz, Kanowski, & Jancke, 2001).
Importantly, in nonhuman primates the brain
connections constituting our proposed mental rotation
network model are constrained by known corticocortical connections. For example, it is well documented that bilateral premotor and dorsolateral
prefrontal regions are reciprocally interconnected and
both project to the anterior cingulate, with the latter
being highly interconnected with neocortical frontal
regions (Fuster, 1997). Electrophysiological (Rizzolatti,
Luppino, & Matelli, 1998) and cytoarchitectonic studies (Petrides & Pandya, 1999) of the cortical motor
system of the macaque monkey reveal that the superior frontal gyrus receives primary inputs from the
superior parietal areas as well as connections from the
inferior parietal regions. Extensive evidence also
exists for connections in nonhuman primates between
the inferior parietal lobule and the dorsolateral
prefrontal cortex (Fuster, 1997; Petrides & Pandya,
1984, 1999), as well as between the superior parietal

279

region and dorsolateral prefrontal cortex (Cavada &

Goldman-Rakic, 1989; Chafee & Goldman-Rakic,
1998). In addition, the superior parietal and inferior
parietal lobules are known to be highly interconnected
across the intraparietal sulcus (Pandya & Seltzer,
1982).
In the present study we propose a mental rotation
(anatomical) network model to examine the interregional covariance structure of the brain processing
required for mental rotation of 3D block figures. Our
main aim was to determine whether mathematical
giftedness was associated with enhanced intrahemispheric and interhemispheric connectivity as exhibited when performing complex spatial processing. We
hypothesized, based on previous research on the MG
brain, that increased interhemispheric connectivity
would be evident in the MG as compared to agematched controls: firstly, between a subnetwork of
anterior regions (i.e., dorsolateral prefrontal cortex,
premotor cortex, and the anterior cingulate), areas that
are associated with attentional and executive functions. And secondly, between a subnetwork of posterior regions (i.e., the inferior and superior parietal
areas) known to mediate image generation and mental
rotation. We also predicted enhanced intrahemispheric frontoparieto connectivity in the MG compared to controls, specifically between the premotor
and dorsolateral prefrontal cortex and the inferior/
superior parietal regions within a given hemisphere of
the brain, the latter being highly interconnected
regions that are thought to mediate high intelligence
in general, and fluid intelligence (a critical ability for
successful mental rotation and potentially math reasoning) in particular. Moreover, we anticipated that a
portion of this increased connectivity would be right
hemispheric given the hypothesis of enhanced development of the right cerebral hemisphere and its
specialized spatial capacities in those who are MG.

METHOD
Participants
A group of 8 mathematically gifted male participants
(mean age 14.3, range 1315; numerical IQ 133 (max
=135), with no known neurological or psychiatric
disorders) were selected for participation in the study
on the basis of their performance on the psychometric
School and College Ability Test (SCAT 111-intermediate
Level-Form Y, Career Wise, Pty Ltd, Melbourne, an
Australian equivalent to the SAT; see Domino &
Domino, 2006). Two sections of the SCAT were used
for classification: one assessing verbal (VIQ) and the

280

PRESCOTT ET AL.

other quantitative (PIQ) reasoning ability. The MG

group comprised participants with PIQ scores above
the 98th percentile for their age (average PIQ = 131.6;
VIQ = 120). Note that for the MG, their PIQ scores
were significantly higher than their VIQ scores, giving them an unbalanced intellectual profile, one that is
biased toward high quantitative reasoning compared
to only moderately high verbal reasoning ability. The
control group (mean age 14.2 years, range 1216;
numerical IQ = 111 (max = 135)) consisted of 8 individuals with PIQ scores placing them in the 4555th
percentile range for their age (average PIQ = 111;
VIQ = 116.6), and their VIQ scores were roughly
equivalent to their PIQ scores (thus reflecting a
balanced intellectual profile). Importantly, the control
VIQ scores were not significantly different from those
of the MG group. Thus, the MG were specifically
math-gifted as the two groups differed solely in their
SCAT-PIQ levels. Note that 4 left-handed and 4 righthanded participants, defined by their performance on
the Edinburgh Handedness Inventory, were intentionally selected and counterbalanced across the groups.
This matching of left- and right-handed subjects,
along with the equivalence of the VIQ scores for each
group, was done to ensure that any observed differences in brain activity were attributable to their math
giftedness rather than their hand preference or intellectual ability per se.

Stimulus materials and procedure

Stimuli were displayed using E-prime 1.0 software
(Schneider, Eschman, & Zuccolotto, 2002) via a
laptop computer. Each acquisition consisted of 18
baseline trials and 18 mental rotation trials, with the
stimuli projected onto a 1.6 1.2 m2 screen at a
distance of approximately 3.5 m. Participants viewed
the stimuli with the assistance of an angled viewfinder
fastened to the MRI head coil, and positioned immediately in front of the eyes of the participant. A single
trial consisted of the simultaneous presentation of a
target stimulus (above) and four test stimuli (below)
for a period of 10 s (1 s interstimulus interval). The
baseline and rotation conditions were presented alternately in blocks of three trials, which were repeated
six times during a scanning session of 6 min 36 s. EPI
scans were acquired on a General Electric 3T MRI
scanner (TR = 3000 ms, TE = 40 ms, FA = 60, voxel
= 1.875 1.875 mm2, slice thickness = 4.5 mm,
interslice gap = 0.5 mm) with one acquisition session
per participant which comprised 132 brain images.
Before commencing the experiment, participants
were given one practice trial of the baseline task and

the rotation task. In the rotation condition, participants were required to select the test stimulus that
could be rotated to match the target, and instructed to
press one of four fiber-optic buttons using the index
or middle finger of either hand. Note that button
position was spatially concordant with the display
position of the four test stimuli (i.e., the index and
middle finger of the left hand mapped on to the spatial
position of the two test figures on the left, and the
index and middle finger of the right hand mapped on
to the spatial position of the two test stimuli on the
right). Each trial in the rotation condition comprised a
unique set of stimuli, rotated three-dimensionally with
angles of rotation between 45 and 180 (see upper
panel of Figure 1).
The baseline condition (see bottom panel of Figure 1)
was similar to the rotation condition with the major
difference being that the block stimuli employed were
Fourier (blurred) transforms of the original rotation
stimuli. These Fourier transforms were used in the
baseline condition to isolate the cognitive processes
involved in mental rotation, by providing contrasting
stimuli that were identical in visual frequency,
luminance and contrast but had no identifiable shape
per se and required no mental rotation. Participants
were told that baseline trials had no correct answer
and that they were to simply choose the test stimulus
they considered to be the best visual match for the
target stimulus. Debriefing of participants after
completion of the study anecdotally confirmed their
efforts to perform the required, albeit somewhat
ambiguous, baseline matching task.

Preprocessing of the fMRI data

Head motion was corrected using FLIRT motion
correction (FMRIB, Release 3.2). Following motion
correction an ICA decomposition (FMRIB, Release
3.2) of the dataset into spatially independent components
was performed. Motion artifacts were identified by
calculating and ranking the correlation coefficient
between the time course of each component and the
motion parameters from FLIRT. A component having
a correlation coefficient greater than .30 with the
motion parameters was deemed artifactual, as were
components dominated by spatial and temporal characteristics localized in one slice or within a small time
frame that resulted from slice dropout and/or residual
motion. Components identified as artifacts were
removed from further analysis, with the resulting
images then realigned and normalized to the SPM5
EPI template, resliced with a voxel size of 2 2 2 mm3
and spatially smoothed with an 8 mm full width, half

ENHANCED CONNECTIVITY

281

Figure 1. A sample rotation trial (upper panel) with the target stimulus appearing above four test stimuli; a sample baseline trial (lower
panel) that uses Fourier (blurred) transforms of the same rotation stimuli.

maximum smoothing function. Statistical analysis

was performed using SPM5 with each stimulus represented as a discrete event using the SPM5 canonical
hemodynamic response function including both temporal and dispersion derivatives. High pass filtering
was applied and an AR (1) + white noise model was
employed to correct for temporal autocorrelation. A
fixed-effects group analysis was used to identify
significant activations.

RESULTS
The MG and control participants performed equivalently on the mental rotation task (mean standard
deviation, MG 43% 11%; control 40% 13%), with
both groups performing significantly better than
chance (25%). Average reaction times to solve the
MR problems were also equivalent for the two groups
(MG = 6.35 0.78 s; control 6.13 0.46 s). These
accuracy levels were somewhat lower than anticipated, and the fact that the MG did not outperform
controls was unexpected given that mental rotation

has been reported to correlate with high mathematical

reasoning ability. In this instance, however, we
believe that having adolescents perform a complex
task such as mental rotation in such a novel,
somewhat intimidating and noisy brain-scanning
environment may have reduced the overall level of
mental rotation performance, which in effect artificially brought the accuracy level of the two groups
closer together. This equivalence, however, was
actually fortunate as subsequent differences in brain
activation profiles between the groups cannot be
attributed to differences in task difficulty or performance
level confounds.

Time course and ROI selection of the

fMRI data
Activation time courses were extracted from regions
of interest (ROIs) including all voxels in an 8 mm
radius sphere centered on the voxel with the largest
Z-score within each activated but non-overlapping
brain region. The covariance matrix was estimated

282

PRESCOTT ET AL.

within condition and across participants. A correction

for the effective degrees of freedom was applied to
the time courses (Worsley & Friston, 1995) to account
for the temporal autocorrelation of these observations.
Homogeneity plots were used to test the variance
across the two groups to provide confidence that any
differences obtained were not due to sample heterogeneity. The standard deviations of the covariance across
200 bootstrapped samples were in the same range for
the MG and controls. To test for significant handedness
effects, a scatter plot of the time courses of the lefthanded participants was plotted against the time
courses of right-handed participants for each combination of brain connections (36 in all for the 9 modeled
regions/nodes) and groups. These different combinations were indistinguishable, indicating that handedness had no significant impact on the results obtained.

Structural equation modeling analysis

SEM as implemented in AMOS 4.0 was used to fit the
estimated covariance matrix of activations to the
anatomical model for the MG and control groups.
Initial residual variances were set to 50% of the
variance for each brain region. Path coefficients less
than .05 were removed from the model. Following the
removal of a given path, the residual variance was
then modified to account for changes to the model.
Modification indices were used to vary the amount of
residual variance accounted for by each variable to

improve model fit. Assessment of model fit to the data

was done using c2.
When network models having good fits were
obtained, group differences between the models were
compared in two ways: (1) identifying differences in
the absolute path coefficients between the MG and
controls, and (2) identifying significant differences in
connectivity between the two groups assessed by the
nested (stacked) models approach (McIntosh &
Gonzalez-Lima, 1994). The latter involves testing a
model in which path coefficients are set to be equal
across groups, while in the former they are allowed to
be different. The two groups were then compared to
identify differences in the strength and patterns of
connections for each of the three submodels of our
anatomical network: (1) anterior regions (i.e., connections among dorsolateral frontal cortex, premotor
cortex, and the anterior cingulate), (2) posterior
regions (i.e., connections between the inferior and
superior parietal lobes), and (3) frontoparieto intrahemispheric (i.e., connections among dorsolateral
prefrontal cortex, premotor cortex, and the inferior/
superior parietal lobes within the same hemisphere).

FMRI data analysis

SPM 5 was used to identify significant clusters of
activation during the mental rotation task as compared
to baseline at PFWE < 0.05 with a cluster threshold of
10 voxels (see Table 1 and Figure 2). Note that in a

TABLE 1
Z-scores and brain coordinates in standard stereotactic space (Talairach & Tournoux, 1998) for
significant activation clusters at p (uncorrected) < 0.001
Math gifted

Average ability

Region

Coordinates

Z-score

Right premotor
Left premotor
Right dorsolateral prefrontal
Left dorsolateral prefrontal
Anterior cingulated
Left superior parietal
Right superior parietal
Left inferior parietal
Right inferior parietal
Right cerebellum
Left cerebellum
Right precuneus
Left precuneus
Right occipital
Left occipital
Right ventrolateral prefrontal
Left ventrolateral prefrontal
Right middle frontal

30 9 54
27 15 57
48 12 18
48 12 27
6 12 51
24 69 54
27 75 54
42 54 42
39 48 42
45 63 39
42 60 30
21 81 45
15 84 42
30 87 21
24 87 15
36 30 6
33 306
48 45 21

>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
>7.31
7.31
>7.31
6.42
6.93
6.07

Coordinates
27 3 48
24 15 63
51 3 27
48 3 27
9 15 39
30 66 48
30 66 48
39 54 45
39 55 45
45 66 30
45 72 21
21 78 45
15 87 42
30 90 12
21 90 24
21 33 15
30 33 6
54 36 21

Z-score
7.75
6.26
5.87
6.29
3.82
>7.75
>7.75
>7.75
>7.75
>7.75
6.87
>7.75
>7.75
7.35
>7.75
6.30
6.30
6.07

ENHANCED CONNECTIVITY

283

Figure 2. Sagittal and coronal slices showing significant activations for the MG (A) and the control group (B) at a probability threshold of
FEW < 0.05 and a cluster threshold of 10 voxels.

preliminary analysis left- vs. right-handedness was

included as an independent variable, but was not
significant in the main, nor did it reliably interact with
any other variable. As such, handedness was excluded
from further analyses.
For the MG, the fixed effects group analysis revealed
significant dorsal stream activations (Figure 2A). Such
activation spread from the occipital lobes superiorly
via the occipito-parietal junction to the precuneus and
parietal lobes. In the parietal lobe, the superior
parietal lobule (BA 7) showed significant activation
bilaterally. This activation extended from the superior
parietal lobe through the intraparietal sulcus, with
significant activation extending into the inferior
parietal lobule (BA 40) bilaterally. Activation in the
cerebellum was also significant. In the anterior region
of the brain, activation was observed bilaterally in
premotor cortex (BA 6), through the medial frontal
gyrus (anterior cingulate region; BA 32), dorsolateral
prefrontal cortex (BA 9, 46), ventrolateral prefrontal
cortex (BA 47), and the right middle frontal gyrus
(BA 10). The control group showed significant
activations in most of the same regions as the MG
(Figure 2B) albeit to a lesser extent, excluding the
anterior cingulate, which was found to be more active
in the MG (Puncorrected < 0.001).

Structural equation modelling

Good model fits were achieved for the MG (c2 =
16.89, df = 14, p = .26) and the controls (c2 = 18.95,
df = 17, p = .33). A comparison between the groups
using the nested models approach revealed an overall significant difference between the two models
(c2diff = 108.7, df = 40, p < .05). When specifically
tested as coefficients of interest, there was an intrahemispheric frontoparietal connectivity difference
(c2diff = 38.1, df = 16, p < .001) between the groups,
with 9 of 14 connections being stronger in the MG
(see the blue arrows in Figure 3). The remaining 5
connections were not significantly different between
the groups. When further examining hemisphere
specific effects in the frontoparietal submodel, significant differences in connectivity between the groups
were observed in both cerebral hemispheres (c2diff =
18.3, df = 8, p < .007 for the left hemisphere and
c2diff = 21.3, df = 8, p < .002 for the right hemisphere); 4 of 8 frontoparietal connections within the
left hemisphere were stronger in the MG group, as
were 5 of 6 frontoparietal connections in the right
hemisphere. The remaining connections in each
hemisphere were not significantly different between
the groups.

284

PRESCOTT ET AL.

.32*
.00

LDL
.51*
.12
.25*
.00

.25*
.00

RDL

.46*
.32

.53*
.35

AC

.23*
.07

.37*
.00
.15*
.00

LPM

.82*
.66

.34*
.22

.24*
.00

.30*
.00

LSP

DL = Dorsolateral Prefrontal
PM = Premotor Cortex
IP = Inferior Parietal
SP = Superior Parietal
AC = Anterior Cingulate

.40*
.11

RPM

.67*
.00

LIP

.41*
.15 .28*
.13

.38*
.26
.24*
.00

.39*
.00

RIP

RSP

Blue = Fronto-Parietal Connections

Green = Anterior Connections
Yellow = Posterior Connections
L = left hemisphere
R = right hemisphere

Red # = Path coefficient for MG

Black # = Path coefficient for controls
* = path coefficients that are
significantly larger at p < .05.

Figure 3. Schematic diagram of the brain connections that are significantly greater in the MG when performing mental rotation. The colour
scheme relates to the three hypothesized subnetworks involved in performing the mental rotation task: (yellow) encoding of the input via
posterior connections; (blue) the processing and manipulation of the encoded spatial information mediated by frontoparietal connections;
(green) central executive functioning, including reasoning/decision processes and response execution, as mediated by anterior connections.

A statistically reliable difference was also obtained

among anterior connections in the MG as compared to
controls (c2diff = 45.1, df = 16, p < .001). Of the 16
connections in this submodel, 9 were stronger in the
MG (see the green arrows in Figure 3) with the
remaining 7 connections not significantly different
between MG and controls. For the 8 connections
between dorsolateral prefrontal cortex, premotor
regions, and the anterior cingulate, there were no reliable group differences (c2diff = 7.6; df = 8, p < .10);
however, the MG and controls did differ in connectivity between the dorsolateral prefrontal and premotor
regions (c2diff = 33.6, df = 8, p < .001). Specifically, 5
of these 8 connections were stronger in the MG group,
with the remaining 3 connections not significantly
different. A statistically reliable difference was also
observed between groups in the posterior parietal
regions (c2diff = 27.8, df = 8, p < .001). In this
submodel, only 2 of the 8 connections between the
superior parietal and inferior parietal regions were
found to be significant and both were in the left

hemisphere, and both were stronger in the MG group

(see the yellow arrows in Figure 3). The remaining 6
connections in this posterior submodel were not
significantly different.

DISCUSSION
In the present study, we examined the covariance
structure of the functional activation maps for MG
and controls performing 3D mental rotation. We used
SEM to estimate the effective connectivity between
brain regions of an anatomical network model of
mental rotation proposed to underlie performance in
this task. We also examined three processing subnetworks by grouping a number of path coefficients into
three regional submodels. Overall, the activation
maps for the MG and the controls exhibited similar
cortical activation patterns, suggesting that the two
groups employ similar strategies when performing
mental rotation; this is also suggestive of a quantitative

ENHANCED CONNECTIVITY

(rather than qualitative) difference between the two

groups in terms of their characteristic brain processes.
The activation maps and behavioral results suggest
that the current mental rotation task was a difficult
one, involving spatial perception, imagery formation,
spatial attention, working memory, and decision processes. To reduce this processing complexity, we modeled a subset (Buchel & Friston, 1997; Bullmore et al.,
2000; McIntosh, 1999) of nine brain regions thought
to underlie performance of our mental rotation task,
and focused on three major subnetworks of connections relating to the cognitive components generally
involved in mental rotation.
When intrahemispheric frontoparietal connections
were examined using SEM, a significant difference
was observed between the two groups: specifically
the MG had higher effective connectivity, with a
slightly larger number of these frontoparietal connections found within the right as compared to the
left hemisphere. Note that the right hemisphere is
considered the dominant hemisphere for performing
complex spatial analysis and is also highly involved
in mathematical reasoning (Butterworth, 1999,
OBoyle et al., 1995). Our anatomical model conceptualizes these frontoparietal connections as a link
between the imagery-based spatial representation of
the object to be rotated (i.e., image formation), the
manipulation of this image (i.e., attempted rotation in
the minds eye), and the subsequent decision-making
processes (i.e., match or no match), as mediated by
the parietal, prefrontal, and premotor regions of the
brain, respectively. The enhanced connectivity of
these frontoparietal pathways provides evidence for
the importance of this processing stream for mental
rotation, and its heightened activation and connectivity
represent unique characteristics of the MG brain.
Dehaene et al. (1999) identified greater activation
in the anterior region of the brain for exact calculation
encoded in a language specific format, and more posterior (parietal lobe) activation for mathematical
approximation, suggesting that the integration of
these brain regions may serve as the neural basis for
advanced mathematical abilities. Our results are
consistent with these ideas and further suggest that
gifted mathematical reasoning ability is related to the
enhanced connectivity between these anterior/posterior
brain regions. Interestingly, a mental rotation study
conducted by Silberstein, Danieli, and Nunez (2003)
that used steady-state event-related partial coherence
(ERPC) to examine synchronization between prefrontal
and posterior electrodes reported increased prefrontal
to posterior ERPC with increasing angle of rotation.
Their observation is consistent with other conventional EEG studies showing increased coherence

285

between these two brain regions during spatial working

memory tasks (Sarnthein, Petsche, Rappelsberger,
Shaw, & von Stein, 1998), as well as other types of
mental rotation tasks (Bhattacharya, Petsche, Feldmann,
& Rescher, 2001). These findings further highlight
the important role of frontoparietal connections in
performing mental rotation. By way of speculation,
the enhanced frontoparietal connectivity exhibited by
the MG not only may provide them with an advantage
when performing complex spatial tasks such mental
rotation, but may also serve as the basis for their
unusual success at performing mathematical reasoning
tasks.
The anterior subnetwork that includes dorsolateral
prefrontal cortex, premotor regions, and the anterior
cingulate also exhibited heightened activation and
higher effective connectivity in the MG. These findings
provide additional support for the hypothesis that the
frontal regions play a crucial role in executive and
decision-making processes and are an integral component of high intelligence (Duncan, 2003; Geake &
Hansen, 2005; Gray et al., 2003). Note that the difference between the two groups in the frontal subnetwork was primarily due to increased path coefficients
for interhemispheric dorsolateral prefrontal and premotor connections, which were significantly greater
in the MG. This result is congruent with previous
behavioral findings by Singh and OBoyle (2004) and
further implicates enhanced interhemispheric connectivity between the frontal regions as a unique characteristic of the MG brain. The latter may provide the MG
with the opportunity to combine and integrate inputs
from a variety of brain areas without the usual interhemispheric processing penalties and costs, which
may be a particularly useful capacity when it comes to
discovering and formulating novel solutions to complex mathematical reasoning problems.
Posterior connections between the left inferior and
superior parietal regions were also found to be
stronger in the MG group. While the exact reason for
these enhanced connections is unclear, they may
reflect the use of an additional, perhaps object-based
(analytic) representation, to supplement the abovementioned imagery-based (right hemisphere) memory
representation when performing this complex spatial
task (Podzebenko, Egan, & Watson, 2002).

Study limitations
There are a number of limitations of using SEM
modeling to investigate networks of cortical connectivity. While such studies provide anatomically plausible results, it is important to assess the behavioral

286

PRESCOTT ET AL.

significance that can be attributed to them. A major

inherent limitation of all SEM modeling is the nonuniqueness of the solution (MacCallum, Wegener,
Uchino, & Fabrigar, 1993). Note that the nature of the
anatomical model, in conjunction with the estimated
interregional covariances and image preprocessing
techniques, is a significant contributor to an obtained
solution. The complexity of any anatomical model
created via SEM must therefore be limited by
interpretability and simplicity, and should be designed
to avoid computational instabilities (McIntosh &
Gonzalez-Lima, 1994). For these reasons, in our study
only homotopic interhemispheric connections were
included. And due to the large number of neuronal
connections in the human brain, a complete model
was extremely difficult if not impossible to create
(although progress is being made in developing and
applying more extensive SEM modeling techniques,
e.g., ES-SEM; see Hanson et al., 2004). Thus care
was taken to ensure that the proposed model was
physiologically sound and cognitively plausible. In
addition, our SEM brain network was based on
anatomical evidence from primate studies that, while
often in agreement with human studies (Petrides,
1994), cannot always be confidently extended to
Homo sapiens (Bullmore et al., 2000).
It is important to note that in the present investigation,
to ensure that interregional covariance was not due to
common activity and engagement in response to
specific task demands, we removed these factors from
the analysis by calculating a within-subject and
across-condition covariance matrix (Goncalves, Hall,
Johnsrude, & Haggard, 2001; Honey et al., 2002;
McIntosh & Gonzalez-Lima, 1994). In addition,
homogeneity plots were used to ensure that all participants in the group had a similar covariance structure.
Of course, one cannot overemphasize that the manner
in which fMRI data are preprocessed is an extremely
important factor in obtaining any SEM result, and
thus it is imperative to remove or limit the effect of
noise on the covariance estimates (Gavrilescu et al.,
2004). In the present study, we enhanced our fMRI
preprocessing strategy with ICA to further limit the
confounding effects of motion and other noisecontributing factors.

CONCLUSIONS
In this study we compared MG adolescents with a
control group while they performed mental rotation of
complex 3D block figures. We assessed differences in
intrahemispheric and interhemispheric connectivity
between the groups based on an a priori anatomical

model of mental rotation using SEM techniques. The

MG showed enhanced intrahemispheric frontoparietal
connectivity and, consistent with previous literature,
the obtained enhanced frontoparietal connections
were somewhat more numerous in the right as compared to the left hemisphere (OBoyle et al., 1995).
We further speculate that the increased effective
connectivity of these frontoparietal regions provides
the MG with an advantage in performing complex
mathematical reasoning problems, perhaps as a byproduct of their enhanced spatial ability, coupled with
a related propensity to code information via imagerybased representations. Each of the latter would seem
especially useful for solving complex mathematical
reasoning problems. Interestingly such theorizing is
consistent with the findings of Halberda, Mazzoco,
and Feigenson (2008) who found that individual differences in school mathematics are strongly related to
an ancient and unlearned number sense, which, speculatively, may be an evolutionary extension and
expression of ones inherent spatial ability level.
We also found enhanced frontal connectivity
between the dorsolateral prefrontal and premotor
brain areas of the MG, which is consistent with other
findings that correlate activation of these frontal
regions with high fluid intelligence (Geake & Hansen,
2005), the latter being a cognitive capacity known to
be especially important for successful mental rotation,
particularly in terms of interconnections with the
parietal lobes as suggested by Jung and Haier (2007)
in their P-Fit model.
Thus, the current results suggest that enhanced
brain connectivity is a unique brain characteristic of
math giftedness and future studies using diffusion
tensor imaging (DTI) and resting state functional
connectivity (rs-fcMRI) techniques will be particularly important for confirming the enhanced connectivity of the MG brain at the neural level.
Additionally, studies employing other types of task
(e.g., verbal processing) would be useful in determining if the enhanced functional connectivity of the MG
brain is specific to mathematical processing per se, or
is also manifest when engaged in other types of
higher-order cognitive functioning (e.g., enhanced
brain connectivity in the MG during language
processing). Finally, the issue of whether the
enhanced functional connectivity of the MG brain is
found only in those individuals at the most extreme
end of the math ability spectrum (i.e., a qualitative
difference from those who are nongifted) or is a
neural characteristic that is distributed to a greater or
lesser extent across the entire math ability continuum
(i.e., a quantitative difference in connectivity as a
function of ones high/moderate/low mathematical

ENHANCED CONNECTIVITY

ability level) is an interesting question that awaits

further research.
Manuscript received 18 November 2009
Manuscript accepted 23 June 2010
First published online 9 August 2010

REFERENCES
Battista, M. T. (1990). Spatial visualisation and gender differences in high school geometry. Journal of Research
in Mathematics Education, 21, 4760.
Benbow, C. P. (1986). Physiological correlates of extreme
intellectual precocity. Neuropsychologia, 24, 719725.
Benbow, C. P. (1988). Sex differences in mathematical reasoning ability in intellectually talented preadolescents.
Brain and Behavioural Sciences, 11, 169232.
Bhattacharya, J., Petsche, H., Feldmann, U., & Rescher, B.
(2001). EEG gamma-band phase synchronization between
posterior and frontal cortex during mental rotation in
humans. Neuroscience Letters, 311, 2932.
Buchel, C., & Friston, K. J. (1997). Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modeling and
fMRI. Cerebral Cortex, 7(8), 768778.
Bullmore, E., Horwitz, B., Honey, G., Brammer, M.,
Williams, S., & Sharma, T. (2000). How good is good
enough in path analysis for fMRI data? NeuroImage,
11(4), 289301.
Butterworth, B. (1999). The mathematical brain. London:
Macmillan.
Cavada, C., & Goldman-Rakic, P. S. (1989). Posterior parietal cortex in rhesus monkey: II. Evidence for segregated
corticocortical networks linking sensory and limbic
areas with the frontal lobe. Journal of Comparative
Neurology, 287(4), 422445.
Chafee, M. V., & Goldman-Rakic, P. S. (1998). Matching
patterns of activity in primate prefrontal area 8a and
parietal area 7ip neurons during a spatial working
memory task. Journal of Neurophysiology, 79(6),
29192940.
Cohen, M. S., Kosslyn, S. M., Breiter, H. C., DiGirolamo,
G. J., Thompson, W. L., Anderson, A. K., et al. (1996).
Changes in cortical activity during mental rotation: A
mapping study using functional MRI. Brain, 119(1),
89100.
Cooper, L. A., & Shepard, M. E. (1973). Chronometric
studies of the rotation of mental images. In W. G. Chase
(Ed.), Visual information processing (pp. 75176). New
York: Academic Press.
Corballis, M. (1997). Mental rotation and the right hemisphere. Brain and Language, 57, 100121.
Dehaene, S. (1997). The number sense. London: Oxford
University Press.
Dehaene, S., Spelke, E., Pinel, P., Stanescu, R., & Tsivkin, S.
(1999). Sources of mathematical thinking: Behavioral and
brain-imaging evidence. Science, 284(5416), 970974.
Dempster, F. N. (1991). Inhibitory processes: A neglected
dimension of intelligence. Intelligence, 15(2), 157173.
Domino, G., & Domino, M. L. (2006). Psychological testing: An introduction (2nd ed.). New York: Cambridge
University Press.

287

Duncan, J. (2003). Intelligence tests predict brain response to

demanding task events. Nature Neuroscience, 6, 207208.
Fuster, J. M. (1997). The prefronal cortex: Anatomy, physiology, and neurospychology of the frontal lobe (3rd ed.).
Philadelphia: Lippincott-Raven.
Gavrilescu, M., Stuart, G., Waites, A., Jackson, G., Svalbe,
I., & Egan, G. (2004). Changes in effective connectivity
in the presence of task- correlated motion: An fMRI
study. Human Brain Mapping, 21, 4963.
Geake, J., & Hansen, P. (2005). Neural correlates of intelligence as revealed by fMRI of fluid analogies. NeuroImage,
26, 255264.
Geschwind, N. & Behan, P. (1982). Left-handedness: Association with immune disease, migraine, and developmental learning disorder. Proceedings of the National
Academy of Sciences of the United States of America, 79,
50975100.
Geschwind, N., & Galaburda, A. (1984). Cerebral Dominance: The biological foundations. Cambridge, MA:
Harvard University Press.
Gill, H. S., OBoyle, M. W. & Hathaway, J. (1998). Cortical
distribution of neural activity for component processes
in mental rotation. Cortex, 34, 707718.
Goncalves, M. S., & Hall, D. A. (2003). Connectivity analysis with structural equation modeling: An example of the
effects of voxel selection. NeuroImage, 20(3), 14551467.
Goncalves, M. S., Hall, D. A., Johnsrude, I. S., & Haggard,
M. P. (2001). Can meaningful effective connectivities be
obtained between auditory cortical regions? NeuroImage,
14(6), 13531360.
Gray, J. R., Chabris, C., & Braver, T. (2003). Neural mechanisms of general fluid intelligence. Nature Neuroscience, 6,
316322.
Halberda, J., Mazzoco, M., & Feigenson, L. (2008). Individual differences in non-verbal number acuity correlate
with maths acheivement. Nature, 455, 665668.
Hanson, S. J., Matsuka, T., Hanson, C., Rebbechi, D.,
Halchenko, Y., & Zaimi, A. (2004). Structural equation modelling of neuroimaging data: Exhaustive
search and Markov chain Monte Carlo. Paper presented at the 10th annual meeting of the Organization
of Human Brain Mapping, Budapest, Hungary.
Hegarty, M., & Kozhevikov, M. (1999). Types of visual
spatial representations and mathematical problem solving. Journal of Educational Psychology, 91(4), 684689.
Honey, G. D., Fu, C. H., Kim, J., Brammer, M. J., Croudace, T. J., Suckling, J., et al. (2002). Effects of verbal
working memory load on corticocortical connectivity
modeled by path analysis of functional magnetic resonance imaging data. NeuroImage, 17(2), 573582.
Jordan, K., Heinze, H. J., Lutz, K., Kanowski, M., &
Jancke, L. (2001). Cortical activations during the mental
rotation of different visual objects. NeuroImage, 13(1),
143152.
Jung, R. E., & Haier, R. J. (2007). The fronto-parietal integration theory (P-FIT) of intelligence: Converging neuroimaging evidence. Brain and Behavioral Sciences, 30,
135187.
Kane, M., & Engle, R. (2002). The role of the prefrontal cortex
in working memory capacity, executive attention, and general fluid intelligence: An individual-differences perspective. Psychonomic Bulletin and Review, 9, 637671.
Kondo, H., Morishita, M., Osaka, N., Osaka, M., Fukuyama, H.,
& Shibasaki, H. (2004). Functional roles of the

288

PRESCOTT ET AL.

cingulo-frontal network in performance on working

memory. NeuroImage, 21(1), 214.
Kosslyn, S. M., DiGirolamo, G. J., Thompson, W. L., &
Alpert, N. M. (1998). Mental rotation of objects versus
hands: Neural mechanisms revealed by positron emission tomography. Psychophysiology, 35(2), 151161.
MacCallum, R. C., Wegener, D. T., Uchino, B. N., &
Fabrigar, L. R. (1993). The problem of equivalent
models in applications of covariance structure analysis.
Psychological Bulletin, 114(1), 185199.
McGee, M. G. (1979). Human spatial abilities: Psychometric studies and environmental, genetic, hormonal, and
neurological influences. Psychological Bulletin, 86(5),
889918.
McIntosh, A. R. (1999). Mapping cognition to the brain
through neural interactions. Memory, 7, 523548.
McIntosh, A. R., & Gonzalez-Lima, F. (1994). Structural
equation modeling and its application to network
analysis in functional brain imaging. Human Brain
Mapping, 2, 222.
OBoyle, M. W. (2008). Mathematically gifted children:
Developmental brain characteristics and their prognosis
for well-being. Roeper Review, 30, 181186.
OBoyle, M. W., Alexander, J., & Benbow, C. P. (1991).
Enhanced right hemisphere activation in the mathematically precocious. A preliminary EEG investigation.
Brain and Cognition, 17, 138153.
OBoyle, M. W., & Benbow, C. P. (1990). Enhanced right
hemisphere involvement during cognitive processing
may relate to intellectual precocity. Neuropsychologia,
28, 211216.
OBoyle, M. W., Benbow, C. P., & Alexander, J. E. (1995).
Sex differences, hemispheric laterality and associated
brain activity in the intellectually gifted. Developmental
Neuropsychology, 11, 415443.
OBoyle, M. W., Cunnington, R., Silk, T., Vaughan, D.,
Jackson, G., Syngeniotis, A., & Egan, G. F. (2005).
Mathematically gifted male adolescents activate a
unique brain network during mental rotation. Cognitive
Brain Research, 25, 583587.
OBoyle, M. W., & Gill, H. S. (1998). On the relevance of
research findings in cognitive neuroscience to educational practice. Educational Pyschology Review, 10,
397409.
OBoyle, M. W., Gill, H., Benbow, C. P., & Alexander, J.
(1994). Concurrent finger-tapping in mathematically
gifted males: Evidence for enhanced right hemisphere
involvement during linguistic processing. Cortex, 30,
519526.
Pandya, D. N., & Seltzer, B. (1982). Intrinsic connections
and architectonics of posterior parietal cortex in the
rhesus monkey. Journal of Comparative Neurology,
204, 196210.
Pesenti, M., Zago, L., Crivello, F., Mellet, E., Samson, D.,
Duroux, B., et al. (2001). Mental calculation in a prodigy
is sustained by right prefrontal and medial temporal
areas. Nature Neuroscience, 4, 103107.
Petrides, M. (1994). Frontal lobes and woking memory:
Evidence from investigations of the effects of cortical

excisions in nonhuman primates (Vol. 9). Amsterdam:

Elsevier.
Petrides, M., & Pandya, D. N. (1984). Projections to the
frontal cortex from the posterior parietal region in the
rhesus monkey. Journal of Comparative Neurology,
228(1), 105116.
Petrides, M., & Pandya, D. N. (1999). Dorsolateral prefrontal
cortex: Comparative cytoarchitectonic analysis in the
human and the macaque brain and corticocortical connection patterns. European Journal of Neuroscience, 11(3),
10111036.
Podzebenko, K., Egan, G. F., & Watson, J. D. (2002).
Widespread dorsal stream activation during a parametric
mental rotation task, revealed with functional magnetic
resonance imaging. NeuroImage, 15(3), 547558.
Reuhkala, M. (2001). Mathematical skills in ninth-graders:
Relationship with visuo-spatial abilities and working
memory. Educational Pyschology, 21(4), 387399.
Rizzolatti, G., Luppino, G., & Matelli, M. (1998). The
organization of the cortical motor system: New concepts. Electroencephalography and Clinical Neurophysiology, 106(4), 283296.
Rowe, J., Stephan, K. E., Friston, K., Frackowiak, R., Lees,
A., & Passingham, R. (2002). Attention to action in Parkinsons disease: Impaired effective connectivity among
frontal cortical regions. Brain, 125(2), 276289.
Sarnthein, J., Petsche, H., Rappelsberger, P., Shaw, G. L.,
& von Stein, A. (1998). Synchronization between prefrontal and posterior association cortex during human
working memory. Proceedings of the National Academy of Sciences of the United States of America,
95(12), 70927096.
Schneider, W., Eschman, A., & Zuccolotto, A. (2002).
E-prime computer software (Version 1.0). Pittsburgh,
PA: Pyschology Software Tools.
Silberstein, R. B., Danieli, F., & Nunez, P. L. (2003).
Fronto-parietal evoked potential synchronization is
increased during mental rotation. NeuroReport, 14(1),
6771.
Shepard, R. & Metzler, (1971). Mental rotation of threedimensional objects. Science, 171, 701703.
Singh, H., & OBoyle, M. W. (2004). Interhemispheric
interaction during globallocal processing in mathematically gifted adolescents, average-ability youth, and
college students. Neuropsychology, 18(2), 371377.
Tagaris, G. A., Kim, S. G., Strupp, J. P., Andersen, P.,
Ugurbil, K., & Georgopoulos, A. P. (1996). Quantitative
relations between parietal activation and performance in
mental rotation. NeuroReport, 7(3), 773776.
Tagaris, G. A., Kim, S.-G., Strupp, J. P., Andersen, P.,
Ugurbil, K., & Georgopoulos, A. P. (1997). Mental rotation studied by functional magnetic resonance imaging
at high field (4 Tesla): Performance and cortical activation. Journal of Cognitive Neuroscience, 9, 419432.
Talairach, J., & Tournoux, P. (1998). Co-planar stereotaxic
atlas of the human brain. Stuttgart, Germany: Thieme.
Worsley, K. J., & Friston, K. J. (1995). Analysis of
fMRI time-series revisitedAgain. NeuroImage, 2(3),
173181.