Increased brain activity in frontal and parietal cortex underlies the development of visuospatial working memory capacity during childhood.

Abstract

The aim of this study was to identify changes in brain activity associated with the increase in working memory (WM) capacity that occurs during childhood and early adulthood. Functional MRI (fMRI) was used to measure brain activity in subjects between 9 and 18 years of age while they performed a visuospatial WM task and a baseline task. During performance of the WM task, the older children showed higher activation of cortex in the superior frontal and intraparietal cortex than the younger children did. A second analysis found that WM capacity was significantly correlated with brain activity in the same regions. These frontal and parietal areas are known to be involved in the control of attention and spatial WM. The development of the functionality in these areas may play an important role in cognitive development during childhood.

Full text

Page 1

Increased Brain Activity in Frontal and Parietal Cortex
Underlies the Development of Visuospatial Working
Memory Capacity during Childhood
Torkel Klingberg, Hans Forssberg, and Helena Westerberg
Abstract
& The aim of this study was to identify changes in brain
activity associated with the increase in working memory (WM)
capacity that occurs during childhood and early adulthood.
Functional MRI (fMRI) was used to measure brain activity in
subjects between 9 and 18 years of age while they performed a
visuospatial WM task and a baseline task. During performance
of the WM task, the older children showed higher activation of
cortex in the superior frontal and intraparietal cortex than the
younger children did. A second analysis found that WM
capacity was significantly correlated with brain activity in the
same regions. These frontal and parietal areas are known to be
involved in the control of attention and spatial WM. The
development of the functionality in these areas may play an
important role in cognitive development during childhood. &
INTRODUCTION
The amount of information one can keep in working
memory (WM) increases throughout childhood and
early adulthood (Gathercole, 1999; Luciana & Nelson,
1998; Hale, Bronik, & Fry, 1997). This increase in WM
capacity is thought to be important for the development
of a wide range of cognitive skills, including reading and
logical reasoning (Engle, Kane, & Tuholski, 1999; Fry &
Hale, 1996; Hulme & Roodenrys, 1995). The changes in
WM capacity coincide in time with several neuronal
developmental processes, including a decrease in syn-
aptic density (Bourgeois & Rakic, 1993; Huttenlocher,
1979), axonal elimination (LaMantia & Rakic, 1990),
changes in global cerebral metabolism (Chugani &
Phelps, 1986; Kennedy & Sokoloff, 1957), myelination
(Klingberg, Vaidya, Gabrieli, Moseley, & Hedehus, 1999;
Paus et al., 1999; Yakovlev & Lecours, 1967; Flechsig,
1920), and changes in catecholamine receptor structure
and density (Lambe, Krimer, & Goldman-Rakic, 2000).
However, there have been no studies measuring any of
these maturational processes and correlating them with
WM capacity in the same individuals.
Functional neuroimaging studies show that children
activate similar areas as adults do during the perform-
ance of WM tasks (Nelson et al., 2000; Thomas et al.,
1999; Casey et al., 1995). Data from WM studies includ-
ing both adults and children is very limited (Thomas
et al., 1999), and none have made any direct voxel-by-
voxel comparisons of regional changes in brain activity
with age. Moreover, it has proven difficult to design the
tasks so that the different age groups perform at the
same level during scanning. When performance during
scanning varies between two groups, differences in
brain activity between the same groups are difficult to
interpret.
In the present study, we used a visuospatial WM task
(Figure 1) that is sensitive for measuring developmental
changes in WM capacity (Hale et al., 1997; Fry & Hale,
1996). Children between 9 and 18 years of age per-
formed the task while their brain activity was measured
with functional MRI (fMRI). By keeping the WM load low
in the scanner version of the task, we intentionally
created a ceiling effect in order to minimize interindi-
vidual differences in behavior during scanning. Differ-
ences in WM capacity was estimated outside the scanner.
The WM-related brain activity was then analyzed to find
correlations between brain activity and age, as well as
between brain activity and WM capacity.
RESULTS
Accuracy during scanning was generally high, except in
one subject where d0= 0.03, which was >3 SD from
the mean of the other subject. This subject was there-
fore excluded from the analysis. After exclusion of this
subject, the mean d0was 2.71 (SD = 0.29), with no
correlation between accuracy during scanning and age
(r = .31, p = .30). The mean reaction time during the
WM tasks was 718 (SD = 118) msec, with a non-
significant trend towards shorter reaction times in old-
er children (r = ¡.50; p = .08).
Karolinska Institute
© 2002 Massachusetts Institute of TechnologyJournal of Cognitive Neuroscience 14:1, pp. 1- 10

Page 2

Areas activated as a main effect of WM, across all ages
and across both Load 3 and Load 5, are shown in Figure
2a- b and listed in Table 1. These areas covered parts of
the prefrontal cortex in the superior and middle frontal
gyri and inferior and superior frontal sulci, cingulate
cortex, and large parts of the parietal and occipital
cortices. The following analyses were restricted to those
areas in which a main effect of WM was seen.
Subtraction images first were created by subtracting
the brain activity during the baseline task from the mean
activity during the WM tasks. One subtraction image was
created for each subject. The subtraction images were
then used in a second-level analysis where age was used
as a covariate to find values in these subtraction images
that correlated with age. A positive correlation with age
was found bilaterally in the superior frontal sulcus, in the
intraparietal and superior parietal cortex, and in the left
occipital cortex (Table 1; Figure 2c- d and Figure 3). In
statistical terms, this correlation is an interaction be-
tween the effect of age and the effect of WM. All of these
areas were also significantly activated when accuracy
during scanning (d0from each subject) was used as a
confound, that is, when variance in brain activity that
could be attributed to differences in performance during
scanning was removed prior to correlation with age. A
tendency towards a negative interaction was found in
the right inferior frontal sulcus (Table 1). As described
in the Introduction, such interaction could possibly
be explained by the trend of negative correlation
between reaction time and age. We therefore included
reaction times as a confound, that is, removed variance
due to differences in reaction times before analyzing the
effect of age. However, including reaction times as a
confound only marginally influenced this effect, decreas-
ing the overall significance of the cluster from p = .08 to
p = .11. In a different analysis, we investigated the
interaction between WM load and age (difference be-
tween Load 5 and Load 3 was correlated with age). No
such significant interaction was found.
WM capacity was measured outside the scanner and
used as a covariate in the same way as the effect of age
was analyzed. Correlation between WM capacity and WM
activity was found in the left superior frontal sulcus and
in the left intraparietal cortex (Figure 2e- f; Table 1), thus
confirming that the correlation between age and WM
related activity was due to the age trends in WM
capacity.
To exclude that there was a lower signal-to-noise ratio
in the images from younger subjects, a ‘‘control-sub-
traction’’ was done. Because there were more visual
stimuli in the baseline task than in the Load 3 WM task
(five circles vs. three circles), subtraction of Load 3 from
baseline would presumably show an effect of this differ-
ence in visual stimulation. When brain activity during
Load 3 was subtracted from baseline task, a large,
significant activation of primary visual cortex and sur-
rounding early visual areas showed up. However, no
interaction was found between activity in this area and
age ( p = .98, using the same thresholds as in the
analysis of interaction between WM and age). This lack
of interaction between visually induced activity and age
confirmed that the interactions between WM-induced
activity and age were not due to a generally lower signal-
to-noise ratio in younger children.
To investigate if there was an age-related trend in the
amount of cortex activated, we performed a separate
statistical analysis of the main effect of WM in each single
subject. Significant clusters were identified (z > 3.09;
cluster size > 45), and the total number of voxels
in these clusters were counted for each subject and
correlated with age. However, there were no age-related
trends in the amount of cortex activated (r = .09,
p = .76).
DISCUSSION
A positive correlation between age and WM-related brain
activity was found in several areas, including the cortex
Figure 1. Working memory
task performed during scan-
ning. The subjects were asked
to remember the location of red
circles that were presented se-
quentially in a 4 £ 4 grid. After a
1500-msec delay, a circle ap-
peared and the subjects pressed
a button if the circle was in the
same location as any of the
dots. Two versions of this task
was given during scanning: one
with three circles to remember
and one with five. In the base-
line condition, subjects watched
as green circles were presented
sequentially in either of the four corner boxes (top-left, top-right, bottom-left, or bottom-right). After a 1500-msec delay, a green circle appeared in
the middle of the grid and the subjects pressed the button.
2Journal of Cognitive NeuroscienceVolume 14, Number 1

Page 3

Table 1. Areas Activated during WM Tasks and Correlation between Age and Activity
Coordinates
Areaxyzt Value
Areas activated during performance of the WM tasks
Superior frontal sulcus
Right 22
¡8
10
60 8.67
20646.29
Left
¡26
¡38
0
6
56
60
10.93
5.11
Middle frontal sulcus
Right42
44
2 28
40
7.99
7.62 26
Left
¡50
¡52
4 32
32
9.97
6.42 30
Frontal operculum
Right 3820
¡8
¡8
48
36
7.64
Left
¡46 187.13
Cingulate cortex4
0
12
26
8.97
8.50
0 38367.87
Caudate nucleus
Right648 4.01
Inferior and intraparietal cortex
Right22
42
¡72
¡42
¡68
¡46
56
44
11.32
8.85
Left
¡24
¡44
60
48
14.50
10.29
Correlation between WM-related activity and age
Positive correlations
Superior frontal sulcus
Right30052 4.72
Left
¡24
¡28
¡452
60
4.48
2.836
Superior and intraparietal cortex
Right22
¡64
¡60
484.75
Left
¡2660 4.65
Middle occipital gyrus
Left
¡32
¡72163.55
Negative correlations
Inferior frontal sulcus
Right42440 2.90*
Klingberg, Forssberg, and Westerberg3

Page 4

in the superior frontal sulcus and intraparietal sulcus,
bilaterally. In a second analysis, a positive correlation
between WM capacity and WM activity was found in the
left superior frontal and left intraparietal areas. This
second analysis suggests that the interaction between
brain activity and age in these areas was due to the
increases in WM capacity with age.
The positive correlation between WM capacity and
brain activity is contrary to the results of some prior
imaging studies that have associated improved perform-
ance as a result of training with lower activity in pre-
frontal areas (Klingberg & Roland, 1998; Jenkins, Brooks,
Nixon, Frackowiak, & Passingham, 1994; Raichle et al.,
1994). The tasks where frontal activity decreased, how-
ever, can presumably become automatic with training,
which is not the case with WM tasks. In general, lower
prefrontal activity associated with superior performance
(Haier et al., 1988) could be due to shorter reaction
times, and thus less time on task (Poldrack, 2000;
D’Esposito et al., 1997). Higher accuracy can also be
associated with less prefrontal activity related to error
detection and error correction. In the present study,
however, interindividual differences in behavior during
scanning were minimized. The increased activity with
higher age therefore represents a developmental trend
in the relationship between behavior and brain activity.
The anatomy and functionality of the frontal and
parietal areas will first be commented on, before discus-
sing the possible mechanism underlying the interaction
between age and WM activity.
Functions of the Superior Frontal Sulcus
Correlation between age and WM activity was found in
the cortex lining the superior frontal sulcus, extending
anteriorly from the intersection of the superior frontal
sulcus and precentral sulcus. Several previous studies of
visuospatial WM have also reported activation at this
location (Rowe, Toni, Josephs, Frackowiak, & Passing-
ham, 2000; Postle & D’Esposito, 1999; Courtney, Petit,
Maisog, Ungerleider, & Haxby, 1998; Sweeney et al.,
1996; Smith et al., 1995; Jonides et al., 1993). In addition
to spatial WM tasks, visual but nonspatial WM tasks
activate this region (Postle & D’Esposito, 1999; Kling-
berg, 1998; Cohen et al., 1997). Several studies have
further characterized the temporal dynamics of activity
in this area and found that it exhibits sustained activity
during the delay period when information is held in WM
(Rowe et al., 2000; Cohen et al., 1997; Courtney, Un-
gerleider, Keil, & Haxby, 1997). The activity thus resem-
bles the delay-specific activity, which is generally
considered to be the hallmark of WM activity in electro-
physiological studies of WM in macaques (Funahashi,
Bruce, & Goldman-Rakic, 1989; Fuster & Alexander,
1971).
The region where we observed a significant correla-
tion was located close, although slightly superior and
anterior, to previously reported activations associated
with voluntary eye movements. Eye-movement-related
activity is generally found along the anterior wall of the
precentral sulcus (Corbetta et al., 1998; Luna et al., 1998;
Paus, 1996; Fox, Fox, Raichle, & Burde, 1985). Further-
more, in the present study, the number of stimuli per
trial, and thus presumably the number of saccades, was,
on average, higher during the baseline condition (five
dots) than in the WM conditions (three and five dots,
respectively); therefore, the activation in the WM tasks
in this study was presumably not due to saccadic eye
movements. Courtney et al. (1998) suggested that activ-
ity underlying visuospatial WM is topographically sepa-
rate from that produced by voluntary saccades. In their
study, eye-movement-related and WM-related activities
overlapped in the intersection of the superior frontal
sulcus and the precentral sulcus. However, the eye
movement- related activity extended inferiorly along
the precentral sulcus, while the WM-related activity
extended anteriorly along the superior frontal sulcus,
consistent with the location of the WM-related activity in
the present study and with eye-movement-related activ-
ity in other studies (Corbetta et al., 1998; Paus, 1996; Fox
Table 1. (continued )
Coordinates
Areaxyzt Value
Correlation between WM-related activity and WM capacity
Positive correlations
Superior frontal sulcus
Left
¡26 4 603.59
Intraparietal cortex
Left
¡36
¡50
¡76
¡46
445.83
Inferior parietal cortex 56 2.90**
For all clusters p < .05 corrected for multiple comparison, except *p = .08; **p = .13.
4Journal of Cognitive Neuroscience Volume 14, Number 1

Page 5

Figure 2. BOLD responses during the WM tasks. (a, b) Main effect of WM in the right (a) and left (b) hemisphere. (c, d) Areas showing a correlation
between WM activity and age: (c) horizontal section showing activity bilaterally in the superior frontal sulcus and intraparietal sulcus; (d) coronal
section showing activity in the left superior frontal sulcus. (e, f ) Areas showing a correlation between WM capacity and WM-related activity: (e)
horizontal section showing the left superior frontal sulcus and left inferior- and intraparietal cortex; (f ) coronal section showing activity in the left
superior frontal sulcus.
Klingberg, Forssberg, and Westerberg5