tmp39AC.tmp

Neuroscience Research 94 (2015) 50–61

Contents lists available at ScienceDirect

Neuroscience Research
journal homepage: www.elsevier.com/locate/neures

The effect of education on regional brain metabolism and its
functional connectivity in an aged population utilizing positron
emission tomography
Jaeik Kim a,b , Jeanyung Chey a,b,∗ , Sang-Eun Kim b,c,d,e , Hoyoung Kim a,f
a

Department of Psychology, Seoul National University, Seoul, South Korea
Interdisciplinary Program in Neuroscience, Seoul National University, Seoul, South Korea
c
Department of Nuclear Medicine, Seoul National University Bundang Hospital, Seoul National University College of Medicine, Seoul, South Korea
d
Department of Transdisciplinary Studies, Graduate School of Convergence Science and Technology, Seoul National University, Seoul, South Korea
e
Advanced Institutes of Convergence Technology, Suwon, South Korea
f
Department of Psychology, Chunbuk National University, Cheonju, South Korea
b

a r t i c l e

i n f o

Article history:
Received 25 October 2014
Accepted 2 December 2014
Available online 31 December 2014
Keywords:
Glucose metabolism
Memory
Language
Connectivity
Graph analysis
Reserve effect

a b s t r a c t
Education involves learning new information and acquiring cognitive skills. These require various cognitive processes including learning, memory, and language. Since cognitive processes activate associated
brain areas, we proposed that the brains of elderly people with longer education periods would show
traces of repeated activation as increased synaptic connectivity and capillary in brain areas involved in
learning, memory, and language. Utilizing positron emission topography (PET), this study examined the
effect of education in the human brain utilizing the regional cerebral glucose metabolism rates (rCMRglcs). 26 elderly women with high-level education (HEG) and 26 with low-level education (LEG) were
compared with regard to their regional brain activation and association between the regions. Further,
graphical theoretical analysis using rCMRglcs was applied to examine differences in the functional network properties of the brain. The results showed that the HEG had higher rCMRglc in the ventral cerebral
regions that are mainly involved in memory, language, and neurogenesis, while the LEG had higher rCMRglc in apical areas of the cerebrum mainly involved in motor and somatosensory functions. Functional
connectivity investigated with graph theoretical analysis illustrated that the brain of the HEG compared
to those of the LEG were overall more efficient, more resilient, and characterized by small-worldness.
This may be one of the brain’s mechanisms mediating the reserve effects found in people with higher
education.
© 2014 Elsevier Ireland Ltd and the Japan Neuroscience Society. All rights reserved.

1. Introduction
Education in schools involves systematic learning in an environment conducive to learning new information and academic skills.
It involves repetitive learning of various cognitive functions and
skills such as memory, language, mathematics, and logic. This process usually takes place from childhood to early adulthood when
brain plasticity is the greatest. On the other hand, training involves
specific physical or mental repetitive learning aimed at improving
one’s capability, productivity and performance (Neubauer and Fink,

∗ Corresponding author at: Department of Psychology, Seoul National University,
Gwanak-ro 599, Gwanak-gu, Seoul 151-742, South Korea. Tel.: +82 2 880 6432;
fax: +82 2 877 6428.
E-mail address: [email protected] (J. Chey).

2009; Ericsson and Ward, 2007). Physical or job training is usually narrower in extent and shorter in duration compared to school
education. Learning, whether by training or school education, is a
process that modifies the brain through experiences. Therefore, it
can be postulated that school education will result in changes in the
broader brain regions involved in memory, language, mathematics,
and logic. However, training will likely be followed by change in a
limited brain region involved in the specific skill or job that was
trained.
Numerous studies have investigated how experiences an animal
has modify its brain. Diamond and her colleagues (1988) were one
of the first teams to find experimental evidence where experience
changes the brain at various levels. They analyzed the brains of rats
bred in enriched environments (EE) and those bred in impoverished environments (IE) over many years. The results showed that
there were differences in the thickness of the cortex, the size of the

http://dx.doi.org/10.1016/j.neures.2014.12.009
0168-0102/© 2014 Elsevier Ireland Ltd and the Japan Neuroscience Society. All rights reserved.

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

pyramidal cell body, the length and number of dendritic branches,
the number and contact areas of the synapses, and the number of
glial cells between the two conditions in many regions of the rat
brain. Although the trend decreased with age, it nonetheless continued. Similarly, Sirevaag and Greenough (1987) found that rats
in complex environments had a greater number of synapses per
neuron compared to those in isolated environments. They also had
increased metabolism such as larger capillary volume per neuron
and larger mitochondria volume per neuron in the occipital cortex. More recent studies examined the effects of motor training on
the brain and found that it increased the number of postsynaptic
dendritic spines over time in the mouse cortex and left minute, but
more permanent marks like lifelong surviving spines and synapses
(Yang et al., 2009; Xu et al., 2009).
Although sparse, studies on humans had similar findings
regarding training or EE on the brain. A selective increase in brain
gray matter was reported in elderly people after motor training
(Boyke et al., 2008; Draganski et al., 2004). Other studies examined the effect of training on expert performance. For example, a
well-known London Taxi driver study showed how extensive training in the spatial memory function was associated with increased
hippocampus size in the drivers (Woollett and Maguire, 2011).
Additionally, a comparison of professional musicians with amateur
musicians and non-musicians showed differences in gray matter
volume in motor, auditory, and visual–spatial brain regions (Gaser
and Schlaug, 2003).
Furthermore, studies (Haier et al., 1988, 1992a, 1992b; Parks
et al., 1988) examining the relationship between training and
absolute cerebral glucose metabolic rates quantified with PET and
18-fluoro-2-deoxyglucose (FDG) have shown a significant decrease
in whole brain glucose metabolism after learning. This supported
the efficiency hypothesis (Haier et al., 1988), which states that the
brains of those who are better at a task are more efficient (i.e.
use less energy performing the task). However, studies investigating the effects of education on the brain have been sparse. One
experimental study found that during auditory sustained attention tasks, people with longer period of education had relatively
higher glucose metabolism in the left posterior cingulate gyrus, the
left precuneus, and bilateral lingual gyri in a small sample of adult
subjects (Eisenberg et al., 2005).
An indirect support for the effect of education on the human
brain comes from clinical studies that report longer delays in
dementia manifestation in patients with higher education (Bennett
et al., 2003; Roe et al., 2007; Garibotto et al., 2008; Stern, 2002). This
hypothesis is supported by neuropathologic studies that show variability in the relationship between the severity of brain pathology
and clinical manifestations (Garibotto et al., 2008). This is consistent with animal experiments where enhanced learning (multiple
learning sessions or relatively strong stimulation) had protective effects against amnestic treatments regardless of the task,
reinforcement type, animal type, and the agent causing amnesia
(Rodriguez-Ortiz and Bermudez-Rattoni, 2007; Prado-Alcalá et al.,
1972, 1978, 1980; Prado-Alcalá and Cobos-Zapiaín, 1977, 1979;
Quiroz et al., 2003). Enhanced learning by humans is likely to take
the form of longer education or training in broader sense that
may involve job or expertise. Consequently, it might be similar to
the results found in animal studies. The cognitive reserve effect
observed in the highly educated elderly population (Coffey et al.,
1999; Stern, 2002; Hanyu et al., 2008) is consistent with the results
of enhanced learning. Even though, cognitive reserve has been
studied extensively to elucidate the role of education in dementia development, no study, to our knowledge, has examined the
direct effects of extensive education on the senescent brain.
Education involves repetitive learning of a variety of cognitive skills and knowledge. These acquired skills and knowledge,
developed from extensive education, are likely to construct

51

new, intensive, efficient, and stable (resilient and robust) neural
networks that mediate cognitive functions like memory, language,
and higher-cognitive processes. Such skills and knowledge appear
to inhibit inefficient networks while enhancing efficient ones in
order to save resources during tasks (Haier et al., 1992a, 1992b).
Since more intensive and efficient networks would consist of connections with larger diameters and more synapses with more
vessels in the vicinity, significantly more energy will be required
by the network in the resting state. Therefore, we hypothesized
that the elderly with high-level education would have increased
glucose metabolism in brain regions associated with memory, language, and higher-cognitive processes than those with low-level
education. In addition, functional connectivity among the regions
was hypothesized to be more efficient and resilient in the highly
educated elderly population.
Utilizing graph theoretical analysis methods, important topological parameters of the brain network like small-worldness,
global efficiency, network resilience, and the hubs or highly connected regions as well as the local network parameters were
examined. Small-worldness is a feature characterized by high clustering with short path lengths between nodes (Watts and Strogatz,
1998). Global efficiency is an indicator of parallel information transfer in the network (Achard and Bullmore, 2007) while network
resilience is an indicator of responses to random failure and targeted attack. The hub is an important regulator of information that
has high centrality, interacts with many other regions, facilitates
functional integration, and plays a key role in network resilience to
insults.
The purpose of the current study was to examine the effect of
education on regional glucose metabolism and its functional connectivity in the human brain, both based on the principle of synaptic
plasticity, which would predict that long-term education would
change the brain regions involved in cognitive skills and knowledge
even in senescence. Using PET, we first compared the regional glucose metabolic rates (rCMRglcs) of elderly women with high-level
education (henceforth HEG) and those with low-level education
(henceforth LEG) in the resting state by using the region of interest
(ROI) analysis to compare the rCMRglc between the two groups. A
graph theoretical analysis using the rCMRglcs computed from ROI
analysis examined differences in the functional network properties
of the two education groups.

2. Participants and methods
2.1. Participants
Participants consisted of women between the ages of 61 and
85 who registered for a recreation and culture course at a senior
program operated by a church in Gangnamgu, Seoul from 2004 to
2008. These women participated in the Seoul Aging Study, which
involves dementia evaluations using the Korean-Dementia Rating
Scale (K-DRS; Chey, 2006) and the Elderly Memory Disorder Scale
(EMS; Chey, 2007) that assesses overall cognitive functioning and
memory, respectively. Additionally, semi-structured interviews
and questionnaires were conducted to measure daily cognitive
and behavioral functioning. Subjects were deemed normal and
not demented based on the two neuropsychological tests as well
as interviews and questionnaires. Individuals with head trauma
history and any clinical neuropathology as well as left-handed
or ambidextrous individuals were excluded from the analysis.
Those who had physical conditions known to compromise cognitive capacity (such as hypertension or uncontrollable diabetes)
were excluded as well. Only 85 out of the 105 subjects passed
the aforementioned criteria. Within this sample, we were able to
define two groups of elderly women with respect to their education

52

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

Table 1
The demographics and mean scores of Korean Dementia Rating Scales for the two education groups.
HEG (n = 26)

Age
Years of formal education
Total scores of K-DRS

LEG (n = 26)

Mean

SD

Mean

SD

69.88
14
135.92

4.95
2
4.53

71
4.62
128.15

5.17
1.72
7.16

t

p

−0.79
18.13
4.67

0.430
<0.001
<0.001

Note: K-DRS: Korean Dementia Rating Scale; n: number of subjects.

status. Twenty-six members were age-matched into each education group (two sample independence t-test of the mean age,
t = −0.79, p > 0.43). The high-education group (HEG) members had
more than 12 years of education, while those in the low education
group (LEG) had less than 6 years of schooling (Table 1).
Occupational history did not differ significantly between the
two education groups. The majority of subjects were full-time
housewives both in the HEG (18/26) and the LEG (11/26). Those
who had jobs in the HEG were teachers, pharmacists, and secretaries, while those in the LEG were hair dressers, vendors, and retail
store clerks. All women were the primary individuals responsible
for the household duties. The limited social activity of women in
the sample was very similar to those of their peers within the time
period the subjects graduated from school (1940s–1960s). In addition, as women with lower socioeconomic status (SES), heavily
represented in the LEG, tended to work for a living compared to
those with higher SES who were mostly full-time housewives, the
analysis of previous occupation were deemed inappropriate in this
sample, and were excluded from the main analyses.
This study was approved by the Seoul National University institutional Review Board (SNUIRB). All subjects were informed of the
study procedure and the PET scan. Informed consent forms were
signed by all participants of the study.
2.2. PET imaging
PET scans were taken by a Philips Allegro PET scanner
(Philips Electronics, New York) with an intrinsic full-widthat-half-maximum (FWHM) resolution of 5.2 mm. Before F18fluoro-2-deoxyglucose (FDG) administration, a transmission scan
was performed for 5 min with a Ge-68 rod source to correct for attenuation. PET Images were reconstructed using a 3D
RAMLA algorithm and displayed in a 128 × 128 matrix (pixel size
2 mm × 2 mm × 2 mm). Images were simultaneously collected for
90 continuous planes at 2 mm thickness. PET scanning continued
for 10 min in three-dimensional mode 40 min after an intravenous
injection of 5 MBq/kg FDG in the resting state with open-eyes,
non-pegged ears, and no intended thought in a dark and quiet environment. The environment was such that the elderly subjects did
not suffer any stress from closed space.
2.3. Image analysis
Processing and statistical analyses of the PET images were performed using Statistical Parametric Mapping 2 (SPM2; Members &
Collaborators of the Wellcome Trust Centre for Neuroimaging, UCL,
UK) implanted in Matlab (Mathworks Inc., Sherborn, Natick, MA).
Prior to the statistical analysis, all images were spatially normalized
in the MNI standard template to remove inter-subject anatomical
variations. The spatially normalized images were smoothed by convolution using an isotropic Gaussian kernel with a 16 mm FWHM.
The statistical analysis data were the relative glucose metabolic
rates, which are the regional glucose metabolic values of voxels
divided by the whole brain glucose metabolic value and multiplied
by a scale factor of 50 (Brett, 1999).

2.4. Selection of ROIs
ROIs were selected from regions furnished with their masks
via the WFU PickAtlas 2.4 implanted in the SPM2 toolbox
(http://www.fil.ion.ucl.ac.uk/spm/software/spm2/). A total of 120
ROIs were selected: 116 from Anatomical Automatic Labeling (AAL)
(Tzourio-Mazoyer et al., 2002), the hypothalamus from TD Brodmann areas+, and the midbrain, the pons, and the medulla from TD
lobes (ANSIR Laboratory, Department of Radiologic Sciences, WFU
School of Medicine) in order to include as much brain gray matter as
possible while avoiding overlap. The ROIs were then condensed into
97 ROIs by merging 26 cerebellar regions of AAL into three regions
to decrease the number of parameters for the statistics (Table 2).
2.5. PET data analysis
All statistical models and tests for rCMRglc were analyzed
using SPM2. We compared the rCMRglcs of the HEG with those
of the LEG to find brain regions that showed significantly different
metabolisms in the two education groups by utilizing the ROI-based
analysis.
The rCMRglc of each ROI for the HEG members was computed
while controlling for the age as the computation for the LEG members. Subsequently, while controlling for the mean rCMRglcs of 97
ROIs, the rCMRglcs for the ROIs were compared in the two education groups (permutation test; p < 0.05, one-tailed).
To visually compare the whole brain, the voxels where the mean
rCMRglcs of the two groups significantly differed were plotted
(two-sample t-test, p < 0.001, uncorrected).
2.6. Analysis of the functional connectivity
Hagman and colleagues reported substantial correspondence
between the structural connectivity and resting-state functional
connectivity measured with diffusion spectrum imaging and fMRI,
respectively, in the human cortex (Hagmann et al., 2008). This
study used correlational analysis across subjects in the resting
state PET, in contrast to recent trend using a correlational analysis within subject and across time on data from fMRI, to compare
the brain functional connectivity of the two education groups. Di
et al. (2012), however, reported that the homotopic inter-subject
metabolic covariances from rest-state PET were comparable to the
corresponding fMRI resting-state time-series correlations.
A graph theoretical analysis was applied to compare the functional network properties among brain regions of the HEG and
LEG members. The graph analysis toolbox (GAT), a Matlab-based
Package, was performed on the binary connection matrix (Hosseini
et al., 2012a), while the brain connectivity toolbox (BCT), also
Matlab-based, on the weighted connection matrix (Rubinov and
Sporns, 2010). The raw data for graph theoretical analysis were
rCMRglcs of the two group members for each of the 97 ROIs.
A linear regression analysis was conducted at every ROI to control for the rCMRglcs means of the 97 ROIs. The residuals of this
regression were then substituted for the raw data (Hosseini et al.,
2012a; Bernhardt et al., 2011). In order to test the statistical significance between-group differences in all network measures, a

J. Kim et al. / Neuroscience Research 94 (2015) 50–61
Table 2
Full names and order of 97 ROIs.
Frontal lobe
1 Precentral gyrus-L
2 Precentral gyrus-R
3 Superior frontal
gyrus-L
4 Superior frontal
gyrus-R
5 Superior frontal
orbital gyrus-L
6 Superior frontal
orbital gyrus-R
7 Middle frontal
gyrus-L
8 Middle frontal
gyrus-R
9 Middle frontal orbital
gyrus-L
10 Middle frontal
orbital gyrus-R
11 Inferior frontal
operculum-L
12 Inferior frontal
operculum-R
13 Inferior frontal
triangular gyrus-L
14 Inferior frontal
triangular gyrus-R
15 Inferior frontal
orbital gyrus-L
16 Inferior frontal
orbital gyrus-R
17 Rolandic
operculum-L
18 Rolandic
operculum-R
19 Supplementary
motor area-L
20 Supplementary
motor area-R
21 Olfactory cortex-L
22 Olfactory cortex-R
23 Superior frontal
medial gyrus-L
24 Superior frontal
medial gyrus-R
25 Medial frontal
orbital gyrus-L
26 Medial frontal
orbital gyrus-R
27 Rectus gyrus-L
28 Rectus gyrus-R
Para-limbic organ
29 Insula-L
30 Insula-R
31 Anterior cingulate
gyrus-L
32 Anterior cingulate
gyrus-R
33 Middle cingulate
gyrus-L
34 Middle cingulate
gyrus-R
35 Posterior cingulate
gyrus-L
36 Posterior cingulate
gyrus-R
37 Hippocampus-L
38 Hippocampus-R
39 Parahippocampal
gyrus-L
40 Parahippocampal
gyrus-R
41 Amygdala-L
42 Amygdala-R

Occipital lobe
43 Calcarine cortex-L
44 Calcarine cortex-R
45 Cuneus-L

49 Superior occipital
gyrus-L
50 Superior occipital
gyrus-R
51 Middle occipital
gyrus-L
52 Middle occipital
gyrus-R
53 Inferior occipital
gyrus-L
54 Inferior occipital
gyrus-R
55 Fusiform gyrus-L

Temporal lobe
80 Heschl gyrus-L
81 Heschl gyrus-R
82 Superior temporal
gyrus-L
83 Superior temporal
gyrus-R
84 Superior temporal
pole-L
85 Superior temporal
pole-R
86 Middle temporal
gyrus-L
87 Middle temporal
gyrus-R
88 Middle temporal
pole-L
89 Middle temporal
pole-R
90 Inferior temporal
gyrus-L
91 Inferior temporal
gyrus-R
Cerebellum

56 Fusiform gyrus-R

92 Cerebellum-L

Parietal lobe

93 Cerebellum-R

57 Postcentral gyrus-L

94 Vermis

58 Postcentral gyrus-R

Brain stem

59 Superior parietal
gyrus-L
60 Superior parietal
gyrus-R
61 Inferior parietal
gyrus-L
62 Inferior parietal
gyrus-R
63 Supramarginal gyrus-L
64 Supramarginal gyrus-R

95 Midbrain

46 Cuneus-R
47 Lingual gyrus-L
48 Lingual gyrus-R

65 Angular gyrus-L
66 Angular gyrus-R

96 Pons
97 Medulla

nonparametric permutation test with 1000 repetitions was used.
The 95th percentile score of the difference distribution was set up as
the critical value. To compare over a wide range, one-tailed p-value
was calculated based on the percentile position.
Recent studies prefer the binary undirected matrix, since binary
networks are usually simpler to characterize and have a more easily defined null model (Rubinov and Sporns, 2010). On the other
hand, weighted characterization can filter out the influence of weak
and potentially non-significant connections (Rubinov and Sporns,
2010; Saramaki et al., 2007). Despite the fact that the methodology
for the weighted networks has not been firmly established, especially on the definition of weight (Nakamura et al., 2009; Saramaki
et al., 2007), we analyzed the connectivity utilizing both methods
to examine consistency.
2.6.1. Analysis based on binary adjacency matrices
The correlation matrix based on the corrected rCMRglcs was
generated for the two groups. Binary adjacency matrices were
derived by thresholding the correlation matrix at a range of densities (Dmin :0.02:0.40). The lower bound of the range is determined
as the minimum density (Dmin ), which is a density where the largest
component size is 97, viz. the networks of both groups are not fragmented and paths exist between each node and every other node.
The minimum density of the study was at 0.10 and all further network analyses on binary adjacency matrices are performed over
this density range.
2.6.1.1. Global network measures. In this study, six essential properties were analyzed to compare global connectivity: global
efficiency, normalized clustering coefficient (Gamma), normalized path length (Lambda), small-worldness (Sigma), network
resilience, and network hub. To evaluate the brain network topology, the parameters for small-worldness were compared to the
corresponding mean values of a random graph with the same number of nodes (97 ROIs), total edges (links between ROIs), and degree
(number of links connected to a node) distribution as the network of
interest. Thus, we obtained the global efficiency index of a network
by
1
1
E=
Ei =
n
n
i∈N

67 Precuneus-L
68 Precuneus-R
69 Paracentral lobule-L
70 Paracentral lobule-R
Basal ganglia and
diencephalon
71 Caudate-L
72 Caudate-R
73 Putaman-L
74 Putaman-R
75 Pallidum-L
76 Pallidum-R
77 Thalamus-L
78 Thalamus-R
79 Hypothalamus

Note: L: left hemisphere; R: right hemisphere.

53



d−1
j∈N j =
/ i ij
n−1

i∈N

where Ei is the efficiency of node i, n is the number of nodes, N is the
set of all nodes in the network, dij is shortest path length between
nodes i and j, and the small-worldness index of a network from
sigma = [C/Crand ]/[L/Lrand ]. The C denotes the clustering coefficient
(measures of functional segregation) as
C=

1
1  2ti
Ci =
,
n
n
ki (ki−1 )
i∈N

i∈N

where Ci is the clustering coefficient of node i, ti is the number
of triangles around node i and the L denotes the characteristic path
length (measures of functional integration), or the average shortest
path length. This is given as
1
1
L=
Li =
n
n
i∈N

i∈N



d
j∈N j =
/ i ij
n−1

where Li is the average distance between node i and all other
nodes. The Crand and the Lrand were the mean clustering coefficient
and the mean characteristic path length of the random network
(20 random networks), respectively. In order to test the statistical
significance of the between-group difference in network topology, a nonparametric permutation test with 1000 repetitions was
used. Between-group differences in global network measures were

54

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

investigated over a range of densities (0.1:0.02:0.40). In order to
see the network resilience, we calculated the largest remaining
component size in response to successive random node removals
(thresholded at minimum density, Dmin (0.10)). The same procedure analyzed the network behaviors in response to targeted
attacks, but removing the nodes in order of decreasing nodal
betweenness centrality (a measure of functional centrality). Hubs
are essential components for efficient communication. If the degree
of a node is 2*SD (standard deviation) greater than average network
degree, the node is considered a hub (Hosseini et al., 2012a, 2012b;
Rubinov and Sporns, 2010).

The algorithm based on weighted matrix is not considerably different from that based on the binary matrix, except for matrix entries
and the definition of path length. The weighted characteristic path
length is defined as
1
Lw =
n
i∈N

2.6.2. Analysis based on weighted matrices
In order to assess the weight’s effect, the connection strength
between ROIs, the four essential properties, viz. global efficiency, normalized clustering coefficient (Gamma), normalized
path length (Lambda), and small-worldness (Sigma) were analyzed
based on weighted matrices of the two groups, and then compared.

dw
j∈N,j =
/ i ij
n−1

,

where dijw is the shortest weighted path length between i and j
defined as
dijw =



u,v∈g w

2.6.1.2. Regional network measures. In order to examine the network of the two education groups on local scale, local clustering
– a fraction of triangles around individual nodes – was calculated
as the measure of segregation, while local betweenness centrality
as a measure of centrality of the local networks. Since comparing every network density for the regional measures results in
a large number of comparisons (number of densities × number
of ROIs), the areas under a curve (AUC) of the regional network
measures were compared. For this purpose, the curves extracted
from thresholding across the density ranges (0.10:0.02:0.40)
were used. Each of these curves depicts the changes in a specific network measure for each group as a function of network
density. The areas were normalized for between-group comparison since the coefficients of the two groups were significantly
different.



1
wuv

i↔j

w is the shortest weighted distance between nodes i and j
where gi↔j
in the graph crossing on nodes u and v, and wuv is the weight of the
connection between nodes u and v (Rubinov and Sporns, 2010).
In constructing weighted networks, each weighted matrix component (correlation coefficient rij ) was labeled as rij , if rij was
positive. In cases that involve self-correlation and negative correlation, it was considered 0. In contrast to the unweighted binary
matrix, which showed whole curves across threshold ranges, the
weighted matrix analysis was performed with a fixed density. As
with the binary analysis, a nonparametric permutation test with
1000 repetitions was used to test the difference between two
groups on all measures.

3. Results
3.1. Mean comparison of the rCMRglc between the two education
groups in 97 ROIs
The ROIs where the mean rCMRglc of the HEG was significantly higher than that of the LEG (nonparametric permutation test
(Mann–Whitney), p < 0.05, one-tailed) are shown in Table 3 and the

Table 3
The ROIs where a rCMRglc of the HEG is significantly higher than that of the LEG (permutation test, p < 0.05, one-tailed).
Region

HEG
rCMRglc (mean) ± SD

Middle temporal gyrus-L
Fusiform gyrus-L
Superior temporal gyrus-L
Olfactory cortex-R
Insula-L
Olfactory cortex-L
Parahippocampal gyrus-L
Hypothalamus
Middle temporal gyrus-R
Inferior frontal orbital gyrus-L
Insula-R
Heschl gyrus-R
Inferior temporal gyrus-L
Hippocampus-L
Parahippocampal gyrus-R
Amygdala-R
Middle temporal pole-R
Heschl gyrus-L
Amygdala-L
Rolandic operculum-L
Inferior temporal gyrus-R
Superior temporal gyrus-R
Caudate-L
Superior temporal pole-L
Rectus-L
Middle temporal pole-L
Superior frontal orbital gyrus-L
Rectus-R

75.92
76.10
80.09
71.64
78.80
72.75
62.83
59.95
77.83
73.57
79.80
83.46
70.14
65.33
64.04
70.00
60.57
83.29
69.11
78.52
73.04
80.43
61.69
60.58
78.10
57.42
74.22
79.22

Note: L: left hemisphere; R: right hemisphere.
*
Survival after multiple comparisons (Bonferroni correction).

±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±

2.08
2.22
2.17
2.40
2.34
2.71
2.42
3.23
2.73
2.25
2.60
3.04
2.24
2.77
2.36
3.33
2.63
2.76
3.69
2.07
2.68
2.44
3.76
2.39
2.69
2.75
2.90
2.61

LEG
rCMRglc (mean) ± SD
74.29
74.20
78.08
69.06
77.08
70.52
60.75
56.93
76.24
72.14
78.31
81.18
68.79
63.47
62.08
68.05
58.90
80.90
67.19
77.03
71.82
78.60
59.31
58.92
77.23
56.17
73.32
78.44

±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±
±

2.68
1.96
3.26
2.82
2.20
3.11
2.87
3.88
2.38
2.96
2.53
4.62
2.15
2.87
3.12
4.32
3.18
4.30
3.89
2.64
2.31
3.31
4.14
3.42
2.74
2.59
3.39
2.69

p
<0.0005*
<0.0005*
<0.0005*
0.001
0.001
0.002
0.002
0.002
0.002
0.003
0.003
0.004
0.004
0.006
0.006
0.006
0.006
0.008
0.010
0.015
0.016
0.018
0.019
0.021
0.023
0.025
0.028
0.044

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

55

Table 4
The ROIs where a rCMRglc of the LEG is significantly higher than that of the HEG (permutation test, p < 0.05, one-tailed).
Region

HEG
rCMRglc (mean) ± SD

Supplementary motor area-R
Postcentral gyrus-R
Precuneus-R
Paracentral lobule-R
Paracentral lobule-L
Precentral gyrus-R
Supplementary motor area-L
Precuneus-L
Postcentral gyrus-L

82.02
77.42
87.80
82.10
80.57
79.17
81.49
87.05
76.62

±
±
±
±
±
±
±
±
±

3.43
2.78
2.35
5.06
3.78
2.30
3.33
2.69
2.12

LEG
rCMRglc (mean) ± SD
85.66
78.82
89.71
85.78
83.54
80.59
83.84
88.86
77.49

±
±
±
±
±
±
±
±
±

4.12
2.37
2.60
4.93
4.78
2.64
3.97
3.88
2.32

p
<0.0005*
0.004
0.005
0.005
0.006
0.013
0.013
0.023
0.044

Note: L: left hemisphere; R: right hemisphere.
*
Survival after multiple comparisons (Bonferroni correction).

ROIs where the mean rCMRglc of the LEG was significantly higher
than that of the HEG are shown in Table 4 (see Table 2 for ROIs).
For reference, the voxels where the mean rCMRglc of the HEG
and that of the LEG were significantly different were examined
at each lobe over the whole-brain. The statistical parametric map
(two-sample t-test) is shown in Fig. 1. The mean rCMRglc of the HEG
located in the ventral cerebral regions and the subcortex were significantly higher than that of the LEG. On the other hand, the mean
rCMRglc of the LEG located in the apical regions of the cerebral
cortex was significantly higher than that of the HEG.
3.2. A graph theoretical analysis
3.2.1. Measures based on binary matrix
The correlation (association) matrices, the binary (adjacency)
matrices, and the brain connectivity maps (excluding cerebellum,
brain stem, and hypothalamus) generated at a minimum density
(0.10) for each group are shown in Fig. 2. The brain connectivity
was visualized with the BrainNet Viewer (Xia et al., 2013). The correlation coefficients of the LEG were generally higher than those
of the HEG (Fig. 2I). The brains of the HEG members had longer
connections that are mediated by hubs-nodes located at temporal

areas and render information flow among remote areas. Its connectivity was comparatively uniform over the whole brain, while that
of the LEG were characterized by locally clustered network with
scanty long connections (Fig. 2III).
Between-group differences in global network measures on
networks thresholded at a range of densities (0.10:0.02:0.40) were
investigated. Compared with the LEG, the HEG network showed
larger global efficiency at every density and showed significant difference in the 0.14–0.34 density range. Compared with the LEG,
the HEG network showed larger normalized clustering coefficient
(Gamma) and smaller normalized path length (Lambda) at every
density. Lambda showed significant differences in the 0.20–0.32
density ranges between the two education groups. This pattern led
to larger small-world indices (Sigma) in the HEG network at every
density and showed significantly larger small-world indices in the
0.10–0.20 and 0.24–0.28 density ranges (permutation test, p < 0.05)
(Fig. 3).
In order to analyze the network behavior in response to random
attacks, the size of the largest remaining component in response
to successive node removal, in random order, was calculated at
the minimum density (0.1). The remaining largest component size
change of the network as a function of randomly removed node

Fig. 1. The voxels that showed significant difference in the mean rCMRglc of the HEG and that of the LEG are illustrated in colors (overlaid on a single T1 image template of
SPM2). Arabic numbers above the images indicate MNI template z-coordinates (p < 0.001, uncorrected).

56

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

Fig. 2. Correlation matrices, binary matrices, and brain connectivity maps thresholded at the minimum density (0.10). (I) Correlation (association) matrix for the HEG and
the LEG; the color-bar shows the strength of the connection. (II) Binary (adjacency) matrix for the HEG and the LEG; red color represents connection existence. (III) The brain
connectivity for the HEG and LEG. Each number in the horizontal and vertical axes indicates the ROIs. See Table 2 for the name of ROI corresponding to each number in this
figure.

fractions is shown in Fig. 4 (left). In most of the removed node fractions, the remaining largest component size of the HEG network
was larger compared to LEG, and in some the differences were significant. The same procedure analyzed the network behaviors in
response to targeted attacks, but removed the nodes in decreasing
nodal betweenness centrality order. The HEG network was more
robust to targeted attacks compared to the LEG network and the
difference reached significance at a few attacked node fractions
(permutation test, p < 0.05) (Fig. 4, right).
The identified HEG hubs were the bilateral olfactory cortices,
the left hippocampus, the left parahippocampus, the left amygdala,
and the midbrain. The LEG hubs were the left pallidum, the right
superior temporal pole, and the pons.

Between-group differences in local network measures were
analyzed, and are shown in Fig. 5. The areas under a curve (AUC)
for each network measure thresholded at a range of densities
(0.10:0.02:0.40) were used for comparison. The local clustering
coefficients and local betweenness centralities were compared
between HEG and LEG to investigate the segregation and centrality
of the network at local scales. The ROIs that HEG had significantly
larger clustering coefficient than LEG were the left posterior cingulate gyrus, the bilateral postcentral gyrus, the bilateral paracentral
lobules, the bilateral putamens, and the bilateral pallidums. The
reverse was the right Inferior frontal operculum, the left hippocampus, the right angular gyrusi, the left caudate, and the right middle
temporal pole (Fig. 5I). The ROIs that HEG had significantly larger

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

57

Fig. 3. Global network measures as a function of network density. Global efficiency, Gamma (=C/Crand : normalized clustering), Lambda (=L/Lrand : normalized path length),
and Sigma (=Gamma/Lamda: small-world index) based on binary adjacent matrix. The + sign indicates that between-group difference is significant at the density (p < 0.05,
permutation test). Note the significantly higher global efficiency and small-worldness of the HEG.

betweenness centrality than the LEG were the right insula, the left
hippocampus, the right angular gyrus, and the left caudate. The
inverse was the right Inferior frontal orbital gyrus, the left putamen,
and the left pallidum (Fig. 5II).

3.2.2. Measures based on weighted connection matrix
The weighted matrix was thresholded at a density of 0.232,
at which the pair ROI connection weight of the LEG was significant, viz. the p-value of correlation between pair ROIs was under

Fig. 4. Between-group differences in network resilience to random attack and targeted attack. Changes in the size of the largest component of the networks after random
attack (left) and targeted attack (right). The + sign indicates between-group difference is significant at the removed node fraction (p < 0.05, permutation test).

58

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

Fig. 5. Between-group difference in AUC (area under a curve) of normalized local network measures. (I) Between-group difference in AUC of normalized local clustering.
(II) Between-group difference in AUC of normalized local betweenness centrality. Vertical coordinates larger than 0 indicate LEG > HEG. Each number in the horizontal axes
indicates the ROIs. The + sign indicates that between-group difference was significant at the region (permutation test, p < 0.05). See Table 2 for the ROI corresponding to each
number in this figure.

0.05. At the density, values of global efficiency were HEG; 0.26,
LEG; 0.31 (p = 0.005*), values of Gamma were HEG; 2.52, LEG; 2.16
(p = 0.095), values of Lambda were HEG; 1.45, LEG; 1.58 (p = 0.039*),
and values of Sigma were HEG; 1.76, LEG; 1.38 (p = 0.043*). Data at
four other densities are used to compare with the data based on
the binary matrix (Fig. 3). The results were very similar to those
based on binary matrices with the exception of global efficiency
(see Figs. 3 and 6).
4. Discussion
The major findings of the study were that regional glucose
metabolism and the functional connectivity of these regions differed significantly between the high and the low education groups
of elderly Koreans. It was found that regional glucose metabolism of
the ventral regions of the cerebral cortex, such as the lateral and the
medial temporal gyri, the inferior frontal cortex, and the caudate,
were significantly higher in elderly women with higher education

(HEG) with some bias to the left (Fig. 1 and Table 3). Those of the
apical part of the cerebral cortex that involve the pre- and the postcentral areas as well as the adjacent precuneus were significantly
higher in elderly women with lower education (LEG) (Fig. 1 and
Table 4). Consistent with the cerebral asymmetry theories (Hines,
1987), the HEG members tended to show higher metabolism in the
left side of the brain than the LEG members, although this was not as
great as the disparity between the inferior versus superior regions
(Tables 3 and 4 and Fig. 1).
The two education groups also differed significantly in the functional connectivity of the brain regions. We found that the brain
network of the HEG members had better balance between integration (Lambda and Global efficiency) and segregation (Gamma)
resulting in greater small-worldness (Sigman; Watts and Strogatz,
1998.) This means there were more functionally segregated ROIs
with a robust number of integrating links in the HEG. The HEG also
demonstrated greater resilience both to random attacks and targeted attacks (Fig. 4) and larger global efficiency (Fig. 3) based on

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

59

Fig. 6. Global efficiency, Gamma, Lambda, and Sigma values of the HEG and the LEG based on weighted connection matrices (permutation test, p < 0.05). The * sign indicates
the significance of between-group difference.

binary matrices. To our knowledge, this is the first study to investigate the effects of education on functional connectivity of the brain
regions.
Consistent with the principle of brain plasticity, the regions that
showed higher activation in the HEG were found mostly in association cortices that are involved in memory and language processing.
The medial temporal areas, such as the hippocampus complex, are
well known for their roles in declarative memory while the caudate is one of the major sites of procedural memory (Squire, 2004).
The key language and auditory sites in the left cerebral cortex also
showed significantly higher activation in the HEG. While the left
inferior frontal regions of the cerebrum has long been recognize
as the center for spoken language production, the left fusiform
gyrus has been found to be associated with orthographic processing
in written language (Tsapkini and Rapp, 2010). These results are
consistent with the fact that people with higher education are
more likely to utilize language skills both in spoken and written
forms. Supporting this notion is the significantly higher activation in insula, and temporal pole by the HEG. These areas along
with the inferior frontal regions have been associated with complex sentence processing and semantics such as metaphors (Rapp
et al., 2004; Diaz and Hogstrom, 2011), figurativeness (Schmidt and
Seger, 2009; Diaz and Hogstrom, 2011), and abstraction (Jefferies
et al., 2009). In addition, insular is not only an integral component
of the central auditory nervous system, but also a vital relay station
in auditory perception. Bilateral damage to this region results in
total auditory agnosia (Bamiou et al., 2003).
In this context, a higher activation of the apical regions of
the cerebral cortex like the supplementary motor area, the precentral gyrus, the paracentral lobule, and the postcentral gyrus,
suggests that motor and somatosensory functions have been utilized heavily in elderly subjects with low education who could have
been more frequently involved in activities that require physical
labor. It is possible that the LEG members would have developed
better motor skills and physical awareness than those of the HEG
members.

It is noteworthy that brain regions involved in neurogenesis in
mammalian brains revealed significant metabolic difference in the
two education groups as well. Compared to the LEG, HEG showed
significantly higher glucose metabolism not only in the hippocampus and the olfactory cortex but also in, the hypothalamus, and the
caudate (Bernier et al., 2000). On the other hand, areas involved
in general intelligence and reasoning, namely, the prefrontal cortex and the inferior parietal cortex, failed to show any difference
between the HEG and the LEG with the exception of the left inferior
frontal cortex.
The graph analysis results supported the ROI analysis in that
the hub brain areas identified in the HEG, based on binary connection matrices, have been associated with declarative and emotional
memories. The left hippocampus, the left parahippocampus, the
left amygdala, and the bilateral olfactory cortices were important
hubs found in the brain of the HEG members. This is particularly
interesting because the olfactory cortex, directly connected with
the amygdala and hippocampus, has been recently found to play
a key role in evoking emotional memories (Menini, 2010). Major
hubs in these brain regions appear to reflect the everyday lives of
elderly women with higher education. Like all women of this generation, they were responsible for taking care of the needs of their
family members while managing the household. It is likely, however, that family responsibilities of the women with HEG would
increase, while the manual labor portion of the household duties
would decrease possibly with the availability of the household help.
Since taking care of family members requires delicate psychological processing that is likely to involve emotional memories, it is
quite possible that this may be reflected in the higher metabolism
of these regions.
It is noteworthy that the observed features from the network
analysis were consistent with and supportive of the findings from
the ROI-based analysis. For instance, the regions identified as the
hub of the brain network in the HEG, such as the bilateral olfactory cortices, the left hippocampus, the left parahippocampus, the
left amygdala and the midbrain, overlapped with regions that had

60

J. Kim et al. / Neuroscience Research 94 (2015) 50–61

the greatest metabolism in the HEG compared to the LEG. The
local analysis further supported the key role the left hippocampus had in the brain networks of the HEG, showing significantly
greater centrality, i.e., the local betweenness centrality. Therefore,
the left hippocampus appears to be not only well-connected to
adjacent cortices but to remote brain regions making it the hub of
information processed in the brain of the HEG members. On the
other hand, the hubs in the brain network of the LEG were the
left pallidum, the right superior temporal pole, and the pons. The
left pallidum was identified as the local hub as well in the LEG
along with the left putamen and the right inferior frontal orbital
gyrus.
Overall, the network analysis suggests that the HEG have more
efficient and more robust information transfer paths and more
shortcuts among ROIs, which are likely to be the result of increased
synaptic plasticity due to longer and more intensive education.
The HEG also had more number of significant local clustering
coefficients, which indicate that they have more developed local
modules as well. Both in the brains of the HEG and the LEG, however, the ROIs with relatively lower metabolism tended to have
greater local clustering, which could indicate that these regions
were developed modularly but was not frequently activated.
Major topology parameter analysis of the brain network based
on weighted matrices showed similar results to binary connection matrix analysis with the exception of global efficiency
(Figs. 3 and 6). Differences in global efficiency between two analyses seem to be due to the arbitrary definition of distance based
on weighted matrix (Rubinov and Sporns, 2010). Since smallworldness is the best index of network development balancing both
integration and segregation, it appears safe to conclude that the
brain of the HEG would transfer the information more efficiently.
Limitations of our study should be mentioned. Although the
two education groups were matched in terms of age, individual
variations in the characteristics of their education, talents, and
occupations may have influenced the results. Further, the results
may be different in men. We, however, think the finding that
education changes the regional metabolism and its connectivity
would be valid, and hope future studies will elaborate and support our findings in other populations. We acknowledge that the
experience after higher education could be very different, such as
occupation, and it is quite likely that the difference in regional
brain metabolism would reflect this as well as the education
itself.

5. Conclusion
Years of formal education affected the regional metabolism and
the connectivity of the regions in elderly women. To our knowledge, this is the first study to investigate the effects of education on
resting-state regional brain metabolism and its functional connectivity. For women with higher education, the increased metabolism
was observed in the ventral parts of the cerebral cortex that are
associated with memory, language, and brain plasticity. On the
other hand, for women who received minimal education, regional
brain metabolism was increased in the apical parts of the cortex,
which are mainly involved in motor action and somatic senses.
These findings are consistent with the notion that higher education
promotes learning, memory, and language skills. They also suggest
that people with more education would have greater brain plasticity, which would be critically important for the reserve of the
brain. Findings from the graph theoretical analyses provide further
evidence that longer education enables greater resilience and efficiency in the brain. Our findings have implications on the neural
reserve in the development of Alzheimer’s dementia; i.e., how education might moderate the effect of neuropathology in Alzheimer’s

disease, since the brain regions found to be well-developed in the
women with higher education were key areas affected early in
Alzheimer’s disease, such as the hippocampus and the surrounding
medial temporal lobe (Kim et al., 2005).
Conflict of interest
The authors have no financial motivations or other objectives
that could be interpreted as a conflict of interest with regard to this
article.
Ethical approval
We have read and abided by the statement of ethical standards
for manuscripts submitted to Neuroscience Research. All authors
have agreed to the final manuscript submission.
Acknowledgements
This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF)
funded by the Ministry of Education, Science, and Technology
(2012R1A1A3013137).
References
Achard, S., Bullmore, E., 2007. Efficiency and cost of economical brain functional
networks. PLoS Comput. Biol. 3 (2), e17.
ANSIR Laboratory, Department of Radiologic Sciences, WFU School of Medicine.
http://fmri.wfubmc.edu/software/PickAtlas
Bamiou, D.E., Musiek, F.E., Luxon, L.M., 2003. The insula (Island of Reil) and its role
in auditory processing. Brain Res. Rev. 42, 143–154.
Bennett, D.A., Wilson, R.S., Schneider, J.A., Evans, D.A., Mendes de Leon, C.F., Arnold,
S.E., Barnes, L.L., Bienias, J.L., 2003. Education modifies the relation of AD pathology to level of cognitive function in older persons. Neurology 60, 1909–1915.
Bernhardt, B.C., Chen, Z., He, Y., Evans, A.C., Bernasconi, N., 2011. Graph-theoretical
analysis reveals disrupted small-world organization of cortical thickness correlation networks in temporal lobe epilepsy. Cereb. Cortex 21, 2147–2157.
Bernier, P.J., Vinet, J., Cossette, M., Parent, A., 2000. Characterization of the subventricular zone of the adult human brain: evidence for the involvement of Bcl-2.
Neurosci. Res. 37, 67–78.
¨
Boyke, J., Driemeyer, J., Gaser, C., Buchel,
C., May, A., 2008. Training-induced brain
structure changes in the elderly. J. Neurosci. 28, 7031–7035.
Brett, M., 1999. Introduction to SPM Statistics. MRC Cognition and Brain Sciences
Unit http://imaging.mrc-cbu.cam.ac.uk/imaging/PrinciplesStatistics
Chey, J.Y., 2007. Elderly Memory Disorder Scale. Hakjisa, Seoul, Korea.
Chey, J.Y., 2006. Korean Dementia Rating Scale. Hakjisa, Seoul, Korea.
Coffey, C.E., Saxton, J.A., Ratcliff, G., Bryan, R.N., Lucke, J.F., 1999. Relation of education
to brain size in normal aging. Neurology 53, 189–196.
Di, X., Biswal, B.B., Alzheimer’s Disease Neuroimaging Initiative, 2012. Metabolic
brain covariant networks as revealed by FDG-PET with reference to resting-state
fMRI networks. Brain Connect. 2 (5), 275–283.
Diamond, M., 1988. Enriching Heredity: The Impact of Environment on the Anatomy
of the Brain. Free Press, New York.
Diaz, M.T., Hogstrom, L.J., 2011. The influence of context on hemispheric recruitment
during metaphor processing. J. Cogn. Neurosci. 23 (11), 3586–3597.
Draganski, B., Gaser, C., Busch, V., Schuierer, G., Bogdahn, U., May, A., 2004. Neuroplasticity: changes in grey matter induced by training. Nature 427, 311–312.
Eisenberg, D.P., London, E.D., Matochik, J.A., Derbyshire, S., Cohen, L.J., Steinfeld, M.,
Prosser, J., Prosser, J., Galynker, I.I., 2005. Education-associated cortical glucose
metabolism during sustained attention. Neuroreport 16, 1473–1476.
Ericsson, K.A., Ward, P., 2007. Capturing the naturally occurring superior performance of experts in the laboratory. Curr. Dir. Psychol. Sci. 16, 346–350.
Garibotto, V., Borroni, B., Kalbe, E., Herholz, K., Salmon, E., Holtoff, V., Sorbi, S., Cappa,
S.F., Padovani, A., Fazio, F., Perani, D., 2008. Education and occupation as proxies for reserve in aMCI converters and AD: FDG-PET evidence. Neurology 71,
1342–1349.
Gaser, C., Schlaug, G., 2003. Brain structures differ between musicians and nonmusicians. J. Neurosci. 23 (27), 9240–9245.
Hagmann, P., Cammoun, L., Gigandet, X., Meuli, R., Honey, C.J., Wedeen, V.J., Sporns,
O., 2008. Mapping the structural core of human cerebral cortex. PLoS Biol. 6 (7),
e159.
Haier, R.J., Siegel, B.V., Nuechterlein, K.H., Hazlett, E., Wu, J.C., Paek, J., Browning, H.L.,
Buchsbaum, M.S., 1988. Cortical glucose metabolic rate correlates of abstract
reasoning and attention studied with positron emission tomography. Intelligence 12, 199–217.

J. Kim et al. / Neuroscience Research 94 (2015) 50–61
Haier, R.J., Siegel, B.V., MacLachlan, A., Soderling, E., Lottenberg, S., Buchsbaum, M.S.,
1992a. Regional glucose metabolic changes after learning a complex visuospatial/motor task: a positron emission tomographic study. Brain Res. 570, 134–143.
Haier, R.J., Siegel, B., Tang, C., Abel, L., Buchsbaum, M.S., 1992b. Intelligence and
changes in regional cerebral glucose metabolic rate following learning. Intelligence 16, 415–426.
Hanyu, H., Sato, T., Shimizu, S., Kanetaka, H., Iwamoto, T., Koizum, K., 2008. The effect
of education on rCBF changes in Alzheimer’s disease: a longitudinal SPECT study.
Eur. J. Nucl. Med. Mol. Imaging 35, 2182–2190.
Hines, T., 1987. Left brain/right brain mythology and implications for management
and training. Acad. Manage. Rev. 12 (4), 600–606.
Hosseini, S.M.H., Hoeft, F., Kesler, S.R., 2012a. GAT: a graph-theoretical analysis
toolbox for analyzing between-group differences in large-scale structural and
functional brain networks. PLoS ONE 7 (7), e40709.
Hosseini, S.M.H., Koovakkattu, D., Kesler, S.R., 2012b. Altered small-world properties
of gray matter networks in breast cancer. BMC Neurol. 12, 28.
Jefferies, E., Patterson, K., Joes, R.W., Rambon Ralph, M.A., 2009. Comprehension of
concrete and abstract words in semantic dementia. Neuropsychology 23 (4),
402–499.
Kim, E.J., Cho, S.S., Jeong, Y., Park, K.C., Kang, S.J., Kang, E., Kim, S.E., Lee, K.H., Na, D.L.,
2005. Glucose metabolism in early onset versus late onset Alzheimer’s disease:
an SPM analysis of 120 patients. Brain 128, 1790–1801.
Menini, A. (Ed.), 2010. The Neurobiology of Olfaction. CRC Press, Boca Raton, FL, pp.
367–394.
Nakamura, T., Hillary, F.G., Biswal, B.B., 2009. Resting network plasticity following
brain injury. PLoS ONE 4 (12), 1–9.
Neubauer, A.C., Fink, A., 2009. Intelligence and neural efficiency. Neurosci. Biobehav.
Rev. 33 (7), 1004–1023.
Parks, R.W., Loewenstein, D.A., Doill, K.L., Barker, W.W., Yoshii, F., Chang, J.Y., Emran,
A., Apicella, A., Sheramata, W., Duara, R., 1988. Cerebral metabolic effects of a
verbal fluency test: a PET scan study. J. Clin. Exp. Neuropsychol. 10, 565–575.
Prado-Alcalá, R.A., Grinberg-Zylberbaun, J., Alvarez-Leefmans, J., Gómez, A., Singer,
S., Brust-Carmona, H., 1972. A possible caudate-cholinergic mechanism in two
instrumental conditioned responses. Psychopharmacologia 25, 339–346.
Prado-Alcalá, R.A., Cobos-Zapiaín, G.G., 1977. Learning deficits induced by cholinergic blockade of the caudate nucleus as a function of experience. Brain Res. 138,
190–196.
Prado-Alcalá, R.A., Bermúdez-Rattoni, F., Velasquez-Martínez, D.N., Gustavo, B.M.,
1978. Cholinergic blockade of the caudate nucleus and spatial alternation performance in rats: overtraining induced protection against behavioral deficits.
Life Sci. 23 (9), 889–896.
Prado-Alcalá, R.A., Cobos-Zapiaín, G.G., 1979. Interference with caudate nucleus
activity by potassium chloride. Evidence for a “moving” engram. Brain Res. 172
(3), 577–583.

61

Prado-Alcalá, R.A., Kaufmann, P., Moscona, R., 1980. Scopolamine and KCl injections
into the caudate nucleus: overtraining-induced protection against deficits of
learning. Pharmacol. Biochem. Behav. 12 (2), 249–253.
Quiroz, C., Martínez, I., Quirarte, G.L., Morales, T., Díaz-Cintra, S., Prado-Alcalá, R.A.,
2003. Enhanced inhibitory avoidance learning prevents the memory-impairing
effects of post-training hippocampal inactivation. Exp. Brain Res. 153, 400–402.
Rapp, A.M., Leube, D.T., Erb, M., Grodd, W., Kircher, T.T.J., 2004. Neural correlates of
metaphor processing. Cogn. Brain Res. 20, 395–402.
Rodriguez-Ortiz, C.J., Bermudez-Rattoni, F., 2007. Memory reconsolidation or updating consolidation? In: Bermudez-Rattoni, F. (Ed.), Neural Plasticity and Memory.
CRC Press, Boca Raton, FL, pp. 209–224.
Roe, C.M., Xiong, C., Miller, J.P., Morris, J.C., 2007. Education and Alzheimer disease without dementia: support for the cognitive reserve hypothesis. Neurology
68, 223–228.
Rubinov, M., Sporns, O., 2010. Complex network measures of brain connectivity:
uses and interpretations. Neuroimage 52, 1059–1069.
Saramaki, J., Kivela, M., Onnela, J.P., Kaski, K., Kertesz, J., 2007. Generalizations of the
clustering coefficient to weighted complex networks. Phys. Rev. 75, 027105.
Schmidt, G.L., Seger, C.A., 2009. Neural correlates of metaphor processing: the roles
of figurativeness, familiarity and difficulty. Brain Cogn. 71, 375–386.
Sirevaag, A.M., Greenough, W.T., 1987. Differential rearing effects on rat visual cortex
synapses. III. Synaptic morphometry. Brain Res. 424 (2), 320–332.
Squire, L.R., 2004. Memory systems of the brain: a brief history and current perspective. Neurobiol. Learn. Mem. 82, 171–177.
Stern, Y., 2002. What is cognitive reserve? Theory and research application of the
reserve concept. J. Int. Neuropsychol. Soc. 8, 448–460.
Tsapkini, K., Rapp, B., 2010. The orthography-specific functions of the left fusiform
gyrus: evidence of modality and category specificity. Cortex 46 (2), 185–205.
Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix,
N., Mazoer, B., Joliot, M., 2002. Automated anatomical labeling of activations in
SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject
brain. Neuroimage 15 (1), 273–289.
Watts, D.J., Strogatz, S.H., 1998. Collective dynamics of ‘small-world’ networks.
Nature 393, 440–442.
Woollett, K., Maguire, E.A., 2011. Acquiring “the knowledge” of London’s layout
drives structural brain changes. Curr. Biol. 21, 2109–2114.
Xia, M., Wang, J., He, Y., 2013. BrainNet Viewer: a network visualization tool
for human brain connectomics. PLOS ONE 8, e68910 http://www.nitrc.org/
projects/bnv/
Xu, T., Yu, X., Perlik, A.J., Tobin, W.F., Zweig, J.A., Tennant, K., et al., 2009. Rapid
formation and selective stabilization of synapses for enduring motor memories.
Nature 462, 915–919.
Yang, G., Pan, F., Gan, W., 2009. Stably maintained dendritic spines are associated
with lifelong memories. Nature 462, 920–925.