c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

Available online at www.sciencedirect.com

Journal homepage: www.elsevier.com/locate/cortex

Research report

Effects of working memory training on functional connectivity

and cerebral blood flow during rest

Hikaru Takeuchi a,*, Yasuyuki Taki b, Rui Nouchi a, Hiroshi Hashizume b,

Atsushi Sekiguchi c, Yuka Kotozaki a, Seishu Nakagawa c,
Calros M. Miyauchi c, Yuko Sassa b and Ryuta Kawashima a,b,c
a
Smart Ageing International Research Center, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan
b
Division of Developmental Cognitive Neuroscience, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan
c
Department of Functional Brain Imaging, Institute of Development, Aging and Cancer, Tohoku University, Sendai, Japan

article info abstract

Article history: Working memory (WM) training (WMT) alters the task-related brain activity and structure
Received 7 March 2012 of the external attention system (EAS). We investigated whether WMT also alters resting-
Reviewed 20 June 2012 state brain mechanisms, which are assumed to reflect intrinsic brain activity and
Revised 29 July 2012 connectivity. Our study subjects were subjected to a 4-week WMT program and brain scans
Accepted 13 September 2012 before and after the intervention for determining changes of functional connectivity and
Action editor Bradley Postle regional cerebral blood flow during rest (resting-FC/resting-rCBF). Compared with no-
Published online xxx intervention, WMT (a) increased resting-FC between the medial prefrontal cortex (mPFC)
and precuneus, which are key nodes of the default mode network (DMN), (b) decreased
Keywords: resting-FC between mPFC and the right posterior parietal cortex/right lateral prefrontal
Working memory training cortex (LPFC), which are key nodes of the EAS, and (c) increased resting-rCBF in the right
Functional connectivity LPFC. However, the training-related decreases in resting-FC between the key DMN node
Cerebral blood flow and the nodes of EAS were only observed when the whole brain signal was regressed out in
Rest individual analyses, and these changes were not observed when the whole brain signal
Plasticity was not regressed out in individual analyses. Further analyses indicated that these
differences may be mediated by a weak but a widespread increase in resting-FC between
the nodes of EAS and activity of multiple bilateral areas across the brain. These results
showed that WMT induces plasticity in neural mechanisms involving DMN and the EAS
during rest and indicated that intrinsic brain activity and connectivity can be affected by
cognitive training.
ª 2012 Elsevier Srl. All rights reserved.

1. Introduction over a short time period (Baddeley, 2003). WM is a functionally

important system that underlies a wide range of higher-order
Working memory (WM) is the limited capacity storage system cognitive activities (Baddeley, 2003; Osaka and Nishizaki,
involved in the maintenance and manipulation of information 2000). Reduced working memory capacity (WMC) is associated

* Corresponding author. Smart Ageing International Research Center, IDAC, Tohoku University, 4-1 Seiryo-cho, Aoba-ku, Sendai
980-8575, Japan.
E-mail address: takehi@idac.tohoku.ac.jp (H. Takeuchi).
0010-9452/$ e see front matter ª 2012 Elsevier Srl. All rights reserved.
http://dx.doi.org/10.1016/j.cortex.2012.09.007

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
2 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

with neurological and psychiatric disorders (Baddeley, 2003; based on the observation of many studies. We further suggested
Goldman-Rakic, 1994; Rose and Ebmeier, 2006) as well as normal usage dependent synaptic formation and reduction may
aging (Wingfield et al., 1988). underlie these rGMV changes (Takeuchi et al., 2011b, 2011d). It is
Some aspects of functional connectivity (or correlation of unclear whether this relatively mild form of WMT leads to
activities between different brain regions) and cerebral blood increased rGMV.
flow (CBF) during rest [resting-FC/resting-regional CBF (rCBF)] Here young adult subjects underwent a 4-week WMT (up to
are associated with conditions with reduced WMC. Regions in 1 h per day). Before and after the intervention, they underwent
the external attention system (EAS), which is dedicated to scanning sessions in which resting-FC associated with DMN
external attention (Buckner et al., 2008; Corbetta and Shulman, and the EAS, resting-rCBF and rGMV were measured. Subjects
2002; Laird et al., 2005), such as the lateral prefrontal cortex in the WMT group completed a 4-week intensive adaptive
(LPFC) and posterior parietal cortex (e.g., the inferior/superior WMT program, and subjects in the control group did not
parietal lobule), are involved with WM and are activated during participate in any such training during the study period. The
WM performance (Baddeley, 2003). The EAS is divided into two lack of an active control group (placebo training) is common to
networks; the dorsal attention network and the ventral almost all imaging studies of cognitive training (Dahlin et al.,
attention network (Fox et al., 2006). In contrast, regions in the 2008; McNab et al., 2009; Olesen et al., 2004; Takeuchi et al.,
default mode network (DMN), such as the medial prefrontal 2010a, 2011b; Westerberg et al., 2004), while an active control
cortex (mPFC), posterior cingulate cortex (PCC), and pre- group has been widely used in psychological studies of
cuneus, are deactivated during WM performance. Activities in cognitive training (Klingberg et al., 2005, 2002). It is an appro-
brain regions that belong to the same network correlate with priate approach for this type of study as the effects of placebo
one another during rest, and a strong anticorrelation is training in a cognitive training study of this type, which
observed between the abovementioned two networks during involves normal adults, are not known and cannot be detected
rest (Fox et al., 2005). In other words, when one network is statistically (Takeuchi et al., 2011d). In addition, the real
activated, the other is deactivated. The strength of resting-FC effects of some cognitive intervention that can be used as an
among regions in DMN is correlated with individual WMC active control intervention cannot be ruled out (Takeuchi
(Hampson et al., 2006). The strength of anticorrelations et al., 2011b). Our subjects also participated in psychological
between the two networks is generally reduced under condi- experiments in which they completed a number of cognitive
tions with reduced WMC (Broyd et al., 2009; Sambataro et al., tests. We hypothesized that WMT would increase resting-FC
2010). It was shown resting-rCBF in PFC or anterior regions within DMN, increase anticorrelations between DMN and
was decreased under conditions with reduced WMC (Martin EAS, increase resting-CBF in PFC, and increase rGMV in EAS.
et al., 1991; Weinberger et al., 1986) described above. These The hypotheses relating to resting-FC and resting-CBF are
brain activities during rest are assumed to reflect brain’s based on the abovementioned previous studies that showed
intrinsic activity and connectivities (Fox and Raichle, 2007). that conditions with reduced WMC are generally character-
Previous studies have shown the effects of WM training ized by a decrease in resting-FC within DMN, a decrease in
(WMT) on psychological measures and neural systems anticorrelations between DMN and EAS, and a decrease in
(Klingberg, 2010; Takeuchi et al., 2010b; Uchida and Kawashima, resting-CBF in PFC. The hypothesis relating to rGMV is based
2008). Moreover, changes in brain activity, gray/white matter on our previous proposition described above.
structures, and dopamine D1 receptor density in LPFC and
posterior parietal regions after WMT have been demonstrated
(Klingberg, 2010; Takeuchi et al., 2010b). Nevertheless, to our 2. Methods
knowledge, no previous study has investigated the effect of
WMT on resting-FC and resting-rCBF. Using these analyses of 2.1. Subjects
rest-related neural mechanisms, we were able (a) to determine
how brain regions interact not only with other brain regions to We enrolled 81 healthy, right-handed university students [59
which they are directly connected structurally but also with men and 22 women; mean age 21.1 years, standard deviation
regions to which they are not structurally connected and (b) to (SD) 1.9]. They had normal vision and no history of neuro-
investigate the state of DMN and the WMN during the cognitive logical or psychiatric illness, which was assessed using
processes involved during rest. Given that altered resting-FC a routine questionnaire. Handedness was evaluated using the
underlies conditions with reduced WMC and that resting-FC Edinburgh Handedness Inventory (Oldfield, 1971). Written
underlies individual WMC, both of which are of scientific and informed consent was obtained from each individual for the
clinical interest, it is important to investigate the extent of projects in which they participated. The Ethics Committee of
plasticity in resting-FC and resting-rCBF. Our previous study Tohoku University approved all procedures. This study was
demonstrated that regional gray matter volume (rGMV) in EAS performed together with another intervention study involving
was decreased after a 5-day intensive (4-h training per day) multitask (MT) training. The different subjects underwent the
WMT program involving mental calculations (Takeuchi et al., different training protocols (WMT and MT training). We aimed
2011d). We previously proposed that this type of intensive and to determine the following: (1) effects of WMT on resting-FC,
concentrated training leads to decreased rGMV, possibly after resting-rCBF, rGMV and performance on the belowmen-
an initial increase in rGMV, whereas training that is not so tioned cognitive tests, (2) effects of WMT on social/emotional/
extensive (e.g., up to 1 h per day for 4 weeks) leads to increased self/behavioral variables, (3) cognitive and neural factors
rGMV, which may be followed by a decrease in rGMV after affecting the WMT outcome, and (4) whether polymorphism
further training (Takeuchi et al., 2011b, 2011d). This proposal is or genetic factors affects the WMT outcome. Given these

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 3

diverse aims and the possible infinite sources of variation, we compliance verification. The experimenter provided training
focused on the first aim. This and the other study involving feedback to the subjects as necessary. MRI scanning and
MT training shared the subjects of the control group, psychological tests were performed immediately before and
psychological and neuroimaging outcome measures, training after the 4-week training. In other words, pre-training MRI
period, and training frequency. Groups of participants scans and psychological tests were performed on day 1,
completed the pre- and post-magnetic resonance imaging training was provided from day 2 to day 28, and post-training
(MRI) studies and psychological experiments during different MRI scans and psychological tests were performed on day 29.
predetermined experimental periods (for example, one group
participated in the project starting 4 weeks from November 2.3. Training tasks
4th, another group participated in the project starting 4 weeks
from November 10th, etc.). Participants in one predetermined Four WMT tasks were presented during each training session.
experimental period were randomly assigned to one of the In all training tasks, difficulties (number of items to be
two groups (e.g., the MT training group or the no-intervention remembered) were modulated based on subjects’ perfor-
control group). The two groups assigned were different for mance. (a) A visuospatial WM task in which circles are pre-
each predetermined experimental period. This means, for sented one at a time at a 1/sec rate in an interface where 10
example, that participants in the 4-week period starting from squares are distributed irregularly (circles are presented in one
November 4th were assigned to the WMT group or the no- of these squares). After stimuli presentation, the subjects
intervention control group, but participants in the 4-week indicate the location and order of the presented stimuli by
period starting from November 10th were assigned to the clicking on a computer screen with a mouse. (b) An auditory
MT training group or the no-intervention control group. The backward operation span task in which pairs of single digits
participants selected the period they wished to participate in (0e9) are presented verbally at a 1 pair/3-sec rate. Within this
and were not notified that there were two intervention groups 3-sec period, one digit is presented per second but in the final
before the experiment. The characteristics of each group are second, no stimuli are presented (e.g., four pairs of digits would
shown in Table 1. The WMT and control groups did not differ be presented as follows: 1, 3, , 4, 9, , 3, 7, , 2, 5, ). The subjects
significantly ( p > .1, two-tailed t tests) in age, sex, and score on have to remember the single digit of the sum of the presented
Raven’s Advanced Progressive Matrices (RAPM) or other WM pairs of digits and the order in which they are presented (in the
measures (visuospatial WM and digit span tasks). Subjects above example, they would need to remember 4, 3, 0, 7) and
who misunderstood the rules for the psychological measures, then repeat the sequence by pressing numbered buttons on
tended to fall asleep during the psychological tests, or could the screen in the reverse order (7, 0, 3, 4 in this order for the
not participate as planned were excluded from the relevant above example). (c) In the dual WM task, which is similar to the
analyses. task implemented in the previous study (Dash et al., 2010),
the subjects have to concurrently perform a visuospatial WM
2.2. Procedure task and an auditory digit span task. Here circles are presented
one at a time at a 1/3-sec rate in the same interface as that of
The WMT program consisted of computerized, in-house the task (a). After stimuli presentation, the subjects indicate
developed Borland Cþþ programs comprising four comput- the location and order of the presented stimuli by clicking on
erized tasks. Subjects undertook approximately 4 weeks a computer screen with a mouse. Simultaneous with the
(27 days) of training (each day, 20e60 min in most cases). presentation of the circle, one single digit (0e9) is presented
However, the total training time depended on the level and verbally. After stimuli presentation, the subjects indicate the
time between trials. The subjects used the program provided digits and the order of the stimuli by pressing numbered
to them on their personal computers. They were recom- buttons on the screen in the presented order. The subjects can
mended to perform the WMT tasks every day; two training perform either of the tasks first. In tasks [a] to [c], when the
sessions for a week were conducted in the laboratory. When subjects solved the problems correctly, the task level increased
they could not perform the tasks because of computer prob- by one, and when they failed to do so three times in a row, the
lems, illness and other reasons, the subjects were allowed to level decreased by one. The difficulty level was varied by
miss a session. They could undergo WMT more than once changing the number of presented stimuli in these tasks. In
a day. Performances in each block (a period when stimuli were each session of each task [a] to [c], when subjects provided
presented sequentially) were logged in a computer file, and correct responses to a task with level N, this was logged in the
occasionally, the subjects were asked to mail the logs for computer and accumulated. When the sum of these Ns sur-
passed a predetermined number, each session of each task [a]
Table 1 e The characteristics of subjects included in each to [c] ended. [d] In a dual WM task, which is similar to the task
of the three groups in this experiment. implemented in a previous study of WMT (Jaeggi et al., 2008),
WMT Control (N ) MT training squares at eight different locations are presented sequentially
(N ¼ 41) (N ¼ 20) (N ¼ 20) on a computer screen at a 3-sec rate (stimulus length,
500 msec; interstimulus interval, 2500 msec). Simultaneous
Age, mean (SD) 20.9 (1.6) 21.4 (2.2) 21.6 (2.1)
Female, number 14 5 3
with the presentation of the squares, one of eight consonants
is presented sequentially using headphones. A response is
Please note that the MT training intervention had the same control required whenever one of the presented stimuli matched the
group as the WMT group. However, this study had nothing else to
one presented n positions back in the sequence. The n value
do with the study of MT training.
was the same for both streams of stimuli. Six auditory and six

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
4 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

visual targets existed per block (four appearing in only one versus sameedifferent judgments (Japanese kana characters;
modality and two appearing in both modalities simulta- judge whether a pair of meaningless Japanese strings are the
neously), and their positions were determined randomly. Here same), filling in a sequence of numbers (fill in the blanks of
subjects responded by pressing “F” on a standard keyboard for a number sequence with suitable numbers according to the
visual targets and “J” for auditory targets. No responses were rules of number arrangement), marking figures [select forms
provided for non-targets. The difficulty level was varied by identical to three samples from a series (sequence) of eight
changing the n level. After each block, the subjects’ individual different forms], and filling in figures (complete incomplete
performance was analyzed, and in the following block, the figures so that they are the same as the sample figures when
n level was adapted accordingly. For fewer than three mistakes rotated). [F] The Stroop task (Hakoda’s version) (Hakoda and
per modality, the n level was increased by 1. For more than five Sasaki, 1990), which measures response inhibition and
mistakes, it was decreased by 1, and in all other cases, impulsivity. Hakoda’s version is a matching-type Stroop task
n remained unchanged. Here after each block, when requiring subjects to check whether their chosen answers are
n remained unchanged, it was logged in the computer and correct, unlike the traditional oral naming Stroop task. The
accumulated. When the n level increased by 1, the n  3 was test consists of two control tasks (WordeColor and Colore
similarly accumulated. A session of dual N-back tasks ended Word tasks), a Stroop task, and a reverse Stroop task. In the
when the sum of this accumulation surpassed a pre- WordeColor task, a color name (e.g., “red”) is presented in the
determined number. Training was completed for the day after leftmost of six columns. The other five columns are painted
subjects completed all tasks. with five colors, and subjects have to check the column whose
color corresponds to the color name in the leftmost column. In
2.4. Psychological outcome measures the ColoreWord task, the leftmost of six columns is painted
with a color and the other five columns contain color names.
Neuropsychological tests and questionnaires were adminis- Subjects have to check the column with the name corre-
tered for pre- and post-training evaluation. These included sponding to the color painted in the leftmost column. In the
the following: [A] RAPM (Raven, 1998), a non-verbal reasoning reverse Stroop task, in the leftmost of six columns, a color
task. [B] Bochumer Matrizen-Test (BOMAT) (Hossiep et al., name is printed in another color (e.g., “red” is printed in blue
1999) (performed groupwise) and as described in Jaeggi et al. letters) and the other five columns are painted in five different
(2008). [C] A (computerized) digit span task, a verbal WM colors from which subjects have to check the column whose
task (Takeuchi et al., 2011c). [D] A (computerized) visuospatial color corresponds to the color name in the leftmost column. In
WM task in which circles are presented one at a time at a 1/sec the Stroop task, in the leftmost of six columns, a color name is
rate in a four-by-four grid-like interface. After stimuli printed in another color (e.g., “red” is printed in blue letters)
presentation, the subjects indicate the location and order of and the other five columns contain color names. Subjects
the presented stimuli by clicking the grid-like interface on have to check the column with the name of the color in which
a computer screen with a mouse in the presented (forward the word in the leftmost column is printed. In each task,
visuospatial WM task) or reverse order (backward visuospatial subjects have to complete as many of these exercises as
WM task). The number of items to be remembered was possible in 1 min. Reverse Stroop and Stroop interference
initially two and progressively increased. Three sequences rates are calculated as follows:
were provided at each level, until the subjects responded

Reverse Stroop interference ¼ ðcorrect answers of Word  color test  correct answers of reverse Stroop testÞ=
ðcorrect answers of Word  color testÞ  100

Stroop interference ¼ ðcorrect answers of Color  word test  correct answers of Stroop testÞ=ðcorrect answers of Color
 word testÞ  100:

incorrectly to all three sequences, at which point the task was [G] Arithmetic tasks. These tests measure multiplication
ended. Each test score was equal to the sum of the number of performance consisting of two forms of one-digit times one-
items correctly repeated in the forward and backward visuo- digit multiplication problems (a simple arithmetic task with
spatial WM tasks. [E] Tanaka B-type intelligence test (Tanaka numbers between 2 and 9) and two forms of two-digit times
et al., 2003) type 3B. This non-verbal mass intelligence test, two-digit multiplication problems (a complex arithmetic task
used for 3rd-year junior high school and older examinees, with numbers between 11 and 19). The two forms of each
does not include story problems but uses figures, single task are the same, but the numbers used in the problems are
numbers, and letters as stimuli. In all subtests, the subjects ordered differently. Each form of the simple and complex
have to solve as many problems as possible before a certain arithmetic tasks has to be completed in 30 and 60 sec,
time (a few minutes). This test consists of the maze test (trace respectively. [H] Kyodai SX test’s (Umemoto et al., 1963)
a maze with a pencil from start to finish), counting cubes subtests for numerical factors. The Kyodai SX test is a stan-
(count cubes piled up in a three-dimensional manner), dardized paper-and-pencil type test used in Japan to assess
a displacement task [substitute a figure (nine figures) with psychometric IQ for subjects with higher cognitive abilities. It
a number (1e9) according to a model chart], identification comprises several subtests. Here two subtests that involve

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 5

complex mathematical problems and that are used for et al., 2011). During the ASL scan, the subjects had to remain
calculating numerical factors (numerical ability score) were still with their eyes closed and could not sleep or think about
used. In one subtest, in each question, one or two equations anything. Three images with no diffusion weighting (b
are presented (such as 15  3 ¼ 18) and another equation in value ¼ 0 sec/mm2) (b ¼ 0 images) and one b ¼ 0 image were
which one of the numbers is replaced by a square is pre- acquired from 52 and 9 subjects, respectively, using a spin-echo
sented (54  6 ¼ ,). In this subtest, the former equations echo-planar imaging (EPI) sequence (TR ¼ 10,293 msec,
become correct when symbols (except “ ¼ ”) are replaced by TE ¼ 55 msec, FOV ¼ 22.4 cm, 2  2  2 mm3 voxels, 60 slices).
another one (in case of the example above, 15 þ 3 ¼ 18 is The mean image of three b ¼ 0 images (for 52 subjects) or one
a correct equation). The subjects have to identify the number b ¼ 0 image (for seven subjects) was used in the preprocessing of
that should be in the square [54  (þ)6 ¼ 60, so the answer is imaging data. Our study subjects also participated in other
60]. In another subtest, in each question, arithmetic word studies or projects. Subjects completed the fMRI paradigms of
problems are presented and subjects have to identify the the N-back WM task (2-back WM condition and 0-back control
combinations of symbols required to calculate the answer condition) before and after training. Details relating to these
(e.g., þ, ). In each question, five answer options are pre- N-back fMRI paradigms can be found in our previous report
sented to the subjects. [I] The SA creativity test (Takeuchi et al., 2011c). We used the fMRI data from the 2-back
(Society_For_Creative_Minds, 1969; Takeuchi et al., 2010d, task of 80 subjects who participated in the fMRI study to show
2010e), which measures creativity through divergent the brain regions activated during the 2-back WM task before
thinking, involves three types of tasks (generate unique ways training (Fig. 2b). Only some of the MRI scans performed in this
of using typical objects, imagine desirable functions for study were described here. Other scans included scans for
ordinary objects, and imagine the consequences of unimag- diffusion tensor imaging and fMRI scans for the emotion task.
inable things happening). The SA test scores the four
dimensions of the creative process (fluency, originality, 2.6. Preprocessing and individual-level statistical
elaboration, and flexibility). Here the sum of the graded analysis of resting-FC data
scores of the four dimensions was used for analysis.
Several questionnaires designed to assess the traits or Preprocessing and analysis of FC data were performed using
states of the subjects were collected but are not described SPM5 implemented in Matlab. Before analysis, blood oxygen
here. These choices were not arbitrary, and consistent with level-dependent (BOLD) images from the pre- and post-training
those reported in our previous studies (Takeuchi et al., 2010c, scans were corrected for slice timing, re-aligned, and re-sliced to
2011d), results from all performance-type cognitive measures the mean image of the BOLD images from the pre-training scan.
were reported. A tester blinded to the groups performed all As has been done in our previous study (Takeuchi et al., 2011c),
neuropsychological assessments. the skull and skin parts of all BOLD images of each subject were
then stripped by masking the raw BOLD images of each subject
2.5. Image acquisition and analysis with a threshold of given signal intensity in the spatially
smoothed [8-mm full width at half maximum (FWHM)] mean
MRI data acquisition was conducted using a 3 T Philips Intera image of all BOLD images of each subject. By doing this, we were
Achieva scanner. Using an magnetization-prepared rapid able to delete the skin and skull parts of the images but not the
gradient echo (MPRAGE) sequence, high-resolution T1- brain parenchyma parts. This is because the skin and skull parts
weighted structural images (240  240 matrix, TR ¼ of the BOLD images have little signal intensity immediately over
6.5 msec, TE ¼ 3 msec, FOV ¼ 24 cm, 162 slices, 1.0-mm slice or under themselves. Thus, by spatial smoothing, the signal
thickness) were acquired. For the resting-state functional MRI intensity of the skin parts is decreased compared with the brain
(fMRI), 34 transaxial gradient-echo images (64  64 matrix, parenchyma parts (signals of small portions of voxels that have
TR ¼ 2000 msec, TE ¼ 30 msec, flip angle ¼ 70 , FOV ¼ 24 cm, an unusually high signal intensity compared with the voxels
3.75-mm slice thickness) covering the entire brain were around them also decrease). The threshold for performing skull
acquired using an echo planar sequence. For this scan, 160 stripping was the same for all subjects, and it was determined
functional volumes were obtained while subjects were resting. by visual inspection ensuring that the skulls of the subjects
During the resting-state scanning, the subjects had to close were stripped but their brain parenchyma parts were not
their eyes but not move, sleep, or think about anything (Greicius deleted. The first smoothing for skull stripping was performed
et al., 2003). Arterial spin labeling (ASL), an MR imaging method, to make mask images for skull stripping the unsmoothed BOLD
was used to measure resting-rCBF non-invasively. ASL was images. Thus, the processed BOLD images that were used in the
performed with quantitative signal-intensity targeting by next processing step were unsmoothed images. Furthermore,
alternating radiofrequency pulse labeling of arterial regions, using the movement-related parameters obtained in the
a pulsed ASL method (Petersen et al., 2006a, 2006b). Details of realignment procedure described above, we calculated an esti-
the sequence and the calculation method for the perfusion mate of the total amount of movement in the resting-state
parameters have been described elsewhere (Petersen et al., fMRI session. This was done by calculating the sum of
2006a, 2006b). The actual imaging parameters were as follows: O[(xvol_n  xvol_n1)2 þ ( yvol_n  yvol_n1)2 þ (zvol_n  zvol_n1)2]
64  64 matrix, TR ¼ 300 msec, TE ¼ 22 msec, FOV ¼ 24 cm, seven (n ¼ 2, 3, 4, . 160) for each session.
slices, 7.0-mm slice thickness (2.0-mm gap), SENSE ¼ 2.5, 84 The masked BOLD images were coregistered to a skull-
averages, scan duration, 5 min 52 sec. We determined the stripped b ¼ 0 image, which was created by a method
position of the slice by putting the fourth of seven slices on the similar to the one used for making skull-stripped BOLD
body of the corpus callosum in the coronal scout view (Taki images, and they were spatially normalized onto a skull-

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
6 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

Fig. 1 e (a, b) WMT-related changes in resting-FC with the mPFC. Increase in resting-FC with mPFC in the WMT group
compared with the control group. Results are shown with a threshold of p < .005, uncorrected. Compared with no-
intervention, WMT resulted in an increase in resting-FC between mPFC and the precuneus. The red sphere is the schematic
seed region in mPFC. (a) The resting-FC results when the time course of the whole brain signal was regressed out in
individual analyses. (b) The resting-FC results when the time course of the whole brain signal was not regressed out in
individual analyses.

stripped b ¼ 0 image template, which was created from data passes visual inspection of all images. This might be because
obtained using our scanner (Takeuchi et al., 2010e). We did not both the images are taken using an EPI sequence and have
use T1-weighted structural images for coregistering because similar characteristics including distortion (Reber et al., 1998).
coregisteration of the BOLD images taken in our studies to the Second, b ¼ 0 images have clearer anatomical characteristics
T1-weighted structural images used in our laboratory often that allow precise normalization. Third, the structure of the
fails when visually inspected. This failure might possibly be orbitofrontal cortex (OFC) is typically lost in BOLD images
caused by differences in the two images due to distortion of taken using the 3T scanner (Stenger, 2006). In a few cases,
the BOLD images. We used b ¼ 0 images for the normalization direct normalization of the BOLD images to the EPI template of
procedures for the following reasons. First, coregisteration of SPM5 distorted the structures around OFC to compensate for
the BOLD images taken in our studies to the b ¼ 0 images the loss of OFC in the BOLD images taken in our study. On the

Fig. 2 e (a) WMT-related changes in resting-FC. Decrease in resting-FC with mPFC in the WMT group compared with the
control group ( p < .005, uncorrected). Compared with no-intervention, WMT resulted in a decrease in resting-FC between
mPFC and the region in the right posterior parietal cortex (Please note this is the only significant result). Tendency of the
results are seen in the bilateral LPFCs. These are the resting-FC results when the time course of the whole brain signal was
regressed out in individual analyses. (b) These are the resting-FC results when the time course of the whole brain signal
was not regressed out in individual analyses. Everything else is presented in the same manner as for (a). The significant
results and tendencies observed in (a) were not observed in this analysis. (c) Brain regions that are activated during the
2-back task before the intervention period. Results are shown with a threshold of p < .005, uncorrected.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 7

other hand, b ¼ 0 images appear to have relatively more of the would show correlation of activity due to global brain activity
structure around OFC on visual inspection, and this prevents as well as region- or network-specific activity synchronization.
this type of distortion and allows for better normalization. Our underlying assumption is that we would observe correla-
However, because b ¼ 0 and BOLD images from the same tions due to network- or region-specific activity synchroniza-
subjects have similar characteristics, these images also tion and not those due to region- or network-irrelevant global
probably match well in the coregisteration process, despite brain activity. However, how the omission of the procedure of
the difference in the structures around OFC. The normaliza- regressing out the whole brain signal changes the results is
tion step resulted in BOLD images with 3  3  3 mm voxels. interesting. Therefore, in addition to the results of analyses of
They were then smoothed (8-mm FWHM). the abovementioned procedures, we included the results of the
Individual-level statistical analyses were performed using analyses where the whole brain signal was not regressed out in
a general linear model. We removed low-frequency fluctua- individual analyses. Correlation maps were produced by
tions with a high-pass filter cut-off value of 128 sec (1/128 Hz). extracting the BOLD time course from seed regions, then
Slow signal drifts with a period longer than this, probably not computing the correlation coefficient between that time course
based on brain activities, are removed by this value. On the and the time course from all other brain voxels. For the present
other hand, we did not use a low-pass filter. Serial correlations study, we examined correlations associated with mPFC and the
in the BOLD signal were accounted for by a first-degree autor- right dorsolateral prefrontal cortex (DLPFC), key DMN nodes,
egressive correction (Della-Maggiore et al., 2002) instead of and the EAS, the latter of which also showed a WMT-related
being removed by a low-pass filter. The use of an autore- increase in resting-rCBF (see Results). The mPFC seed region
gressive model (instead of low-pass filter) is recommended was a 6-mm-radius sphere centered on the focus. This peak
when the signals of the model frequently change over time, voxel of mPFC (x, y, z ¼ 1, 47, 4) was defined as previously
such as in rapid presentation event-related designs (Della- (Fox et al., 2005). The right DLPFC seed region was defined using
Maggiore et al., 2002). When the signals of the model do not WFU_PickAtlas (http://fmri.wfubmc.edu/cms/software) and in
change frequently neither an autoregressive model nor a low- a similar manner as in the previous study (Song et al., 2008). We
pass filter is necessary to control for false positives (Della- defined the right DLPFC seed region by intersecting BA46, the
Maggiore et al., 2002). In a study by Della-Maggiore et al. right middle frontal gyrus and GM in WFU_PickAtlas, and then
(2002), there were no circumstances in which a low-pass re-sliced the generated regions into the same spatial resolution
filter or the combination of an autoregressive model and as the preprocessed fMRI images (3  3  3 mm3).
a low-pass filter was better than an autoregressive model With individual-level analysis, contrast images represent-
without a low-pass filter. This was so both in terms of the ing changes in resting-FC with the seed regions following the
power of the analysis and in the control of false positives. 27-day intervention and those before intervention were esti-
Probably because there are no situations where a low-pass mated for each subject after preprocessing. These images
filter has been shown to be better than an autoregressive were then subjected to the belowmentioned group analysis.
model, the low-pass filter is no longer available in the version
of SPM that we used. Several sources of spurious variances 2.7. Preprocessing of resting-rCBF data
were then regressed by putting these variances into the
following regressors: (i) six parameters obtained by a rigid body Maps of raw resting-rCBF and longitudinal relaxivity (R1) of
correction of head motion, (ii) the whole brain signal averaged each subject were obtained using a dedicated software running
over a whole-brain mask, and (iii) time courses of brain signal on IDL (Research Systems, Boulder, Colo) (2006a; National
changes in unsmoothed normalized images averaged over the Neuroscience Institute, Singapore). The following constants
brain masks of white matter and cerebrospinal fluid (CSF) were used in the CBF calculation: T1 of arterial blood, 1.65 sec;
areas. The brain masks of white matter and CSF consisted of inefficiency, 95%; bloodebrain partition coefficients for gray
areas of tissue probability >.95 in the SPM5 tissue probability and white matters, .98 and .82, respectively. For an explanation
maps for each tissue. This procedure removes fluctuations of the details and technicalities of rCBF calculation, see Petersen
unlikely to be involved in specific regional correlations. et al. (2006a).
Regressing out the global signal shifts the distribution of Preprocessing and data analysis were performed using
correlation coefficients, ensuring that they are approximately SPM5 implemented in Matlab, except in the belowmentioned
centered around zero in such a manner that their sum is less segmentation procedure using SPM2. This measure was per-
than or equal to zero. This method of data cleaning has been formed because the T1-weighted images taken using the
criticized as “artifactually” creating anticorrelations (Murphy abovementioned MRPAGE sequence were incompatible with
et al., 2009), though, a recent study suggested that anti- VBM5 and SPM5 preprocessing and resulted in many apparent
correlations observed in resting-state connectivity are not an segmentation errors.
artifact introduced by global signal regression and might have Our original template of the T1-weighted structural image,
biological origins (Chai et al., 2012). However, more impor- created from the 63 subjects of our previous study (Takeuchi
tantly, regardless of the recent controversy about regressing et al., 2011c), was used in resting-rCBF map preprocessing as
out the whole brain signal, we believe that for the purposes of follows: (a) The T1-weighted structural image of each subject
this study, it was important to regress out the global signal was spatially normalized to the T1 template in SPM5. (b) The
because of the following reasons. First, global signal correlation normalized T1-weighted structural image of each subject was
is observed partly because the brain is globally activated then smoothed using a Gaussian kernel of 8-mm FWHM and
(Schölvinck et al., 2010). In this case, when these global brain finally averaged across subjects to make an original template
activities are not regressed out, FC between two brain regions of the T1-weighted structural image.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
8 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

We segmented the T1-weighted images into the GM, white change correction (modulation) by modulating each voxel with
matter, and CSF. Then, using segmented gray and white the Jacobian determinants derived from spatial normalization,
matter images as masks, we removed parts that did not allowing the determination of regional differences in the
belong to the gray and white matter images from the absolute amount of GM (Ashburner and Friston, 2000). Subse-
T1-weighted image and created images consisting of gray and quently, all images were smoothed by convolving them with an
white matter parts (called “gray matter þ white matter isotropic Gaussian kernel of 12 mm FWHM. Finally, the signal
T1-weighted image”). change in rGMV between pre- and post-intervention images
R1 maps from the pre- and post-scans of each subject was computed at each voxel for each participant. In this
(these do not show the skull and skin of the head and retain computation, we included only voxels that showed GMV
their alignment with the rCBF maps of each subject) were then values >.10 in both pre- and post-scans to avoid possible partial
coregistered to the gray matter þ white matter T1-weighted volume effects around the borders between GM and white
image from the pre-scan of each subject using the within- matter, as well as between GM and CSF. The resulting maps
subject registration method. representing the rGMV change between the pre- and post-MRI
Next, a GM mask image consisting of voxels with values experiments (rGMV post e rGMV pre) were then forwarded to
higher than .2 in the GM image was generated. This image was the group-level analysis described below.
applied to the rCBF maps from the pre- and post-scans to limit
the analysis to resting-rCBF in GM areas and to generate a GM 2.9. Statistical group-level analysis of imaging and
rCBF image. behavioral data
Next, a raw T1-weighted structural image from the pre-scan,
which maintained its alignment with the gray matter þ white Behavioral data were analyzed using SPSS 16.0 (SPSS Inc.,
matter T1-weighted image from the pre-scan and GM rCBF Chicago, IL). Because the superiority of training was our
maps from the pre- and post-scans, of each subject was primary interest, in our behavioral analysis, testeretest
normalized to our original template of the T1-weighted struc- changes in the WMT group were compared to those in the
tural image. control group using one-tailed one-way analyses of covari-
Using the parameters for this normalizing procedure, GM ance (ANCOVAs) with the difference between pre- and post-
rCBF maps from the pre- and post-scans of each participant test measures as dependent variables and pretest scores as
were spatially normalized to create images with 2  2  2 mm3 independent variables ( p < .05). The use of one-tailed test is
voxels. The processed normalized GM rCBF maps from the consistent with our previous study as well as those of other
pre- and post-scans were then spatially smoothed using laboratories (Klingberg et al., 2005, 2002; Takeuchi et al., 2011b,
a Gaussian kernel of 9-mm FWHM. Finally, the signal change 2011d). However, two-tailed ANCOVAs were used in behav-
in (resting-)rCBF between the pre- and post-scan images was ioral measures in which the definition of “superiority” was
computed at each voxel by subtracting the former image from unclear, namely for the reverse Stroop interference rate
the latter for each participant. The maps representing the GM showing age-related decline (Sasaki and Hakoda, 1985) and for
(resting-)rCBF from the pre-scan and (resting-)rCBF changes an increase in the Stroop interference rate in schizophrenics
from the pre- to post-scan were subjected to the belowmen- (Sasaki et al., 1993). The two-tailed ANCOVA was also used to
tioned group-level analysis. compare group differences in changes in creativity test scores,
which are associated with impaired selective attention
2.8. Preprocessing and analysis of structural data systems, psychosis, and cognitive disinhibition (Beech and
Claridge, 1987; Necka, 1999; Stavridou and Furnham, 1996),
Voxel-based morphometry (VBM), which is a method for the as was the case with our previous study of WMT involving
in vivo study of human brain structures that can detect changes mental calculation (Takeuchi et al., 2011d). Creativity seems to
in regional GM caused by training (Driemeyer et al., 2008; Ilg be an obvious positive trait; however, a vast amount of liter-
et al., 2008), was used to investigate the effect of WMT on ature shows that creativity is associated with psychopathol-
brain structures. Preprocessing of the morphological data was ogies and impaired selective attention systems [(for the full
performed using the VBM2 software (Gaser, 2007), an extension discussion of this matter, see Takeuchi et al., 2011c)]. These
of SPM2. To reduce the scanner-specific bias, we used choices were not arbitrary and were consistent with those
a customized GM anatomical template and prior probability reported in our previous study of WMT (Takeuchi et al.,
maps of GM and white matter images created from T1-weighted 2011d). Two-tailed Student’s t-tests were applied to investi-
structural images obtained using this scanner in our previous gate group differences in cognitive performance before the
study (Takeuchi et al., 2010d, 2011). Next, the T1-weighted intervention, and the results are shown in Table 2.
structural images from each subject were segmented into GM In group-level imaging analysis, we tested for groupwise
and white matter partitions using the abovementioned custom differences in changes in resting-FC with each seed region,
GM and white matter prior probability maps. The resulting resting-rCBF and rGMV across the entire brain.
images included extracted GM and white matter partitions in In the analyses of resting-FC, we performed voxel-wise
the native space. The GM partition was then normalized to the ANCOVAs with the difference of each measure between pre-
abovementioned custom GM probability map. The normaliza- and post-scans at each voxel as dependent variables, and the
tion parameters determined from this initial step were applied value for the pre-scan at each voxel, the total amount of
to the native T1-weighted structural image. These normalized movement during the scan (which was calculated above), and
T1-weighted structural data were then segmented into GM and change of the total amount of movement during the scan from
white matter partitions. In addition, we performed a volume pre-scan to post-scan as independent variables. In the analysis

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 9

(Hayasaka et al., 2004) with an underlying voxel-level of p < .05

Table 2 e The average of all subjects’ highest
performances in trained WM tasks among the first and corrected for false discovery rate (Genovese et al., 2002). Non-
last three training sessions. isotropic adjusted cluster size tests can and should be applied
when cluster size tests are applied to data known to be non-
First three Last three
stationary (i.e., not uniformly smooth), such as VBM data
sessions (N ) sessions (N )
(Hayasaka et al., 2004). The abovementioned analysis using
Visuospatial WM task 8.85  .78 11.44  1.88 biological parametric mapping (BPM) was not applied to the
Auditory backward 8.95  1.44 15.33  3.78 rGMV analysis, as to our knowledge, BPM does not use the
operation span task
non-isotropic adjusted cluster size.
Dual WM task 7.64  .81 10.10  1.55
Finally, using simple regression analyses, correlations
Dual N-back task 2.77  .73 4.92  1.19
between improvements in performance on WMT tasks and
Note that the number of training sessions differed for each subject, the amount of neural change in significant areas identified in
and therefore, it is difficult to show the average progressive
the abovementioned ANCOVAs were investigated. For this
improvement of performance as the number of sessions increased.
analysis, performance on WMT tasks for each participant in
the first and last three sessions was calculated as follows:
(highest level of the visuospatial WM task achieved in the first
of resting-rCBF, we performed voxel-wise ANCOVAs with the or last three completed sessions) þ (highest level of the
difference of each measure between pre- and post-scans at auditory backward operation span task achieved in the first or
each voxel as dependent variables and the value for the pre- last three completed sessions) þ 2  (highest level of the dual
scan at each voxel as independent variables. This analysis WM task achieved in the first or last three completed
was performed using Biological Parametrical Mapping sessions) þ 2  (highest level of the dual N-back task achieved
(Casanova et al., 2007) implemented in SPM5 and images rep- in the first or last three completed sessions). Performance on
resenting changes in resting-FC/resting-rCBF and resting-FC/ the dual WM task and dual N-back task were multiplied by two
resting-rCBF in the pre-scan. In the analyses of resting-FC because when the level of the dual N-back task was increased
and resting-rCBF, regions with significance were inferred by one the number of stimuli to be remembered was increased
using cluster-level statistics (Friston et al., 1996). In this by two. An increase in performance on WMT tasks for each
procedure, the null hypothesis was rejected for clusters with participant from the first three sessions to the last three
a large spatial extent. Cluster size distribution can be deter- sessions was regarded as an improvement in performance on
mined by parametric methods based on the Gaussian random WMT tasks. The rationale for this calculation using the first/
field theory, which accounts for image volume, smoothness, last three sessions (see also the Results of training data
and the cluster-defining threshold. In cluster level, inference is subsection of Results) was to obtain stable results since
drawn from the cluster size, i.e., the probability that any subjects have to perform four tasks in one session and there
cluster larger than the critical cluster size is controlled. Only simply is not sufficient time to ensure stable performance in
clusters with p < .05, after correction for multiple comparisons one session. We did not perform correlation analysis using the
at cluster size with a voxel-level cluster-determining threshold amount of WMT as a covariate because in our study, the
of p < .005 uncorrected, were considered statistically signifi- amount of WMT varied little among subjects (see Results for
cant in this analysis. However, for regions with the a priori details). The mean changes in the clusters identified as
hypothesis, the threshold of p < .1 (instead of .05, so note this is significant in the abovementioned ANCOVAs were calculated
not lenient at all), after correction for multiple comparisons at for all the significant clusters in the analyses of rGMV, resting-
cluster size with a voxel-level cluster- determining threshold rCBF, and resting-FC for each subject. For all clusters identified
of p < .005 uncorrected, was applied. Using liberal thresholds as significant by ANCOVA simple regression analyses were
on areas with a strong a priori hypothesis has been performed performed to determine the improvement in performance on
elsewhere (e.g., Pochon et al., 2002) The precuneus and WMT tasks and the neural changes calculated as described
posterior cingulate gyrus were set as regions with the a priori above. In addition, we performed the same procedure using
hypothesis in the analysis of resting-FC with mPFC because improvements in RAPM instead of improvements in perfor-
resting-FC between mPFC and the region around the posterior mance on WMT tasks.
cingulate gyrus and precuneus is correlated with the perfor-
mance of the WM task (Hampson et al., 2006).
In the analyses of rGMV, we used the factorial design option
in SPM5. In the imaging analysis, the effects of the interven- 3. Results
tions, estimated by comparing changes in pre- to post-
measures as described above, were compared between the 3.1. Results of training data
groups at each voxel with total GMV in the pre-measurement,
pretest scores of measures of general intelligence (RAPM), and Subjects in the WMT group completed 25.87 sessions (SD, 2.18)
two measures of WM (digit span task and visuospatial WM on an average and at least 17 sessions during the 27-day
task), as covariates. The differences between the WMT group intervention period. In all four WMT tasks, the highest
and the control group were investigated. This analysis corre- performance achieved during the last three training sessions
sponds to the ANCOVA. In the analysis of rGMV, applying by each participant was significantly increased compared
VBM5 (Gaser, 2007), the level of statistical significance was set with the highest performance achieved during the first three
at p < .05, corrected at the non-isotropic adjusted cluster level training sessions (paired t-test, p < .001, Table 2).

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
10 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

3.2. The effect of WMT on behavioral measures 3.3. The effect of WMT on resting-FC with mPFC

The WMT group showed significantly greater pre- to post-test We next compared changes in resting-FC with mPFC (Fig. 1) in
increases in the performance of tasks used to determine the WMT and control groups. mPFC was chosen as a seed
behavioral measures, such as a digit span task ( p < .001, one- region for DMN because (a) this is one of the key nodes of DMN
tailed) and a visuospatial WM task ( p < .001, one-tailed), than and (b) a previous study showed that resting-FC between this
the no-intervention (control) group. Improvements in the region and PCC was correlated with WM performance (this
performance of non-trained WM tasks, including ones with region was also used as the seed region in that study),
stimuli modalities different from those of the trained tasks, (Hampson et al., 2006).
following WMT were consistent with previous studies When the images of resting-FC that regressed out the
(Klingberg, 2010; Takeuchi et al., 2010b). In WMT tasks, whole brain signal were used, the analysis revealed a statis-
auditory, and not visual, verbal stimuli were used, whereas in tically significant WMT-related increase from pre- to post-
the WM test of the outcome measure, the opposite was the measures in resting-FC between mPFC and the precuneus
case. (x, y, z ¼ 0, 60, 33, t ¼ 3.80, p ¼ .010, corrected for multiple
Furthermore, compared with the control group, the WMT comparisons at the cluster level with a cluster-determining
group showed significantly greater pre- to post-test increases threshold of p < .005, uncorrected; Fig. 1a). These results
in performance on (RAPM; p ¼ .019, one-tailed) and signifi- indicate that WMT increases the resting-FC between the two
cantly greater pre- to post-test decreases in Stroop interfer- central DMN nodes, namely mPFC and the precuneus.
ence ( p ¼ .007, one-tailed) (Table 3 presents results for all A statistically significant WMT-related decrease from pre- to
psychological measures). post-measures was also observed in resting-FC between
Finally, compared with the control group, the WMT group mPFC and a region in the right posterior parietal cortex (the
did not show significantly greater pre- to post-measure anatomical cluster that includes the inferior and superior
changes in the estimate of the total amount of movement parietal lobules) (x, y, z ¼ 51, 42, 54, t ¼ 4.10, p ¼ .001, cor-
during the resting-state fMRI sessions (Table 3). rected for multiple comparisons at the cluster level with

Table 3 e Pre- and post-test scores for psychological measures (mean ± S.E.M).

WMT Control (no-intervention) Planned p valuea p valueb

contrast
Pre Post Pre Post

Non-verbal reasoning
RAPM (score) 28.7  .5 31.9  .4 29.1  .9 31.2  .9 WMT > control .019 .654
BOMAT (score) 8.03  .36 9.26  .32 7.72  .57 9.72  .54 WMT > control .848 .646

WM
Digit span (score) 36.4  1.3 47.1  1.7 35.6  1.4 36.7  1.6 WMT > control <.001 .702
Visuospatial WM (score) 28.7  1.3 34.4  .7 27.9  1.0 30.2  .9 WMT > control <.001 .466

Intelligence test with speeded tasks

Tanaka B-type intelligence test 115.1  1.5 123.9  1.6 112.4  2.1 120.3  2.8 WMT > control .258 .274

Stroop
Reverse Stroop interference (%) 14.4  1.6 17.3  1.5 20.7  2.3 17.6  2.1 Two-tailed .387 .031
Stroop interference (%) 8.7  1.3 6.1  1.1 8.7  1.9 11.6  2.4 WMT < control .007 .995

Arithmetic
Simple arithmetic (items) 33.3  .9 34.7  .9 31.8  1.2 33.0  1.4 WMT > control .362 .334
Complex arithmetic (items) 7.10  .53 7.77  .55 6.72  .55 7.25  .72 WMT > control .693 .663

Complex mathematic
Numerical factor in Kyodai SX test 11.0  .3 12.2  .3 11.3  .5 12.3  .5 WMT > control .480 .576

Creativity
SA creativity test (total grade) 25.7  .9 26.8  .8 28.8  1.3 27.3  1.2 Two-tailed .313 .054

Movement
Movement during resting-state 12.4  .3 12.2  .3 12.4  .7 13.3  .9 WMT < control .066 .967
fMRI scans (mm)

All tasks were untrained tasks. The visuospatial WM task used as an outcome measure is similar to one of the training tasks because it is also
a computerized visuospatial WM task. The digit span task is a visual verbal WM task, and visual verbal WM components were not included in
the training tasks.
a One-way ANCOVAs with testeretest differences in psychological measures as dependent variables and pretest scores of the psychological
measures as covariates.
b Two-tailed Student’s t-tests were used to compare group differences in cognitive performance before training.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 11

a cluster-determining threshold of p < .005, uncorrected; a statistically significant WMT-related decrease from pre- to
Fig. 2a). This parietal region belongs to the areas activated by post-measures in resting-FC between the right DLPFC and the
WM task as shown in Fig. 2c. Substantial WMT-related anatomical cluster that included mPFC and the ventral ante-
decreases from pre- to post-measures were also observed rior cingulate gyrus (x, y, z ¼ 3, 36, 3, t ¼ 4.54, p ¼ .004,
in resting-FC between mPFC and regions in the bilateral corrected for multiple comparisons at the cluster level with
LPFCs (including the inferior and middle frontal gyri) (>15 a cluster-determining threshold of p < .005, uncorrected,
voxels, with a cluster-determining threshold of p < .005, Fig. 3a). Together with the findings from the analysis of
uncorrected). For consistency of the results when the right resting-FC with mPFC, these results showed that WMT
DLPFC was taken as a seed region of interest (ROI), see the decreases resting-FC between mPFC and nodes of the EAS. No
subsection of “The effect of WMT on resting-FC with the right region showed statistically significant WMT-related increases
DLPFC” below. in resting-FC with the right DLPFC. However, there was
When the images of resting-FC in which the whole brain a tendency toward a WMT-related increase in resting-FC
signal was not regressed out were used, a statistically signif- between the right DLPFC and the left posterior parietal
icant WMT-related increase from pre- to post-measures in cortex (15 voxels, with a cluster-determining threshold of
resting-FC between mPFC and the precuneus (x, y, z ¼ 0, 60, p < .005, uncorrected).
33, t ¼ 3.92, p ¼ .015, corrected for multiple comparisons at the When the images of resting-FC in which the whole brain
cluster level with a cluster-determining threshold of p < .005, signal was not regressed out were used, the same analysis did
uncorrected; Fig. 1b) was observed. However, no statistically not reveal a statistically significant or substantial tendency
significant or substantial tendency toward a WMT-related toward any WMT-related decrease from pre- to post-
decrease from pre- to post-measures in resting-FC between measures in resting-FC between the right DLPFC and an
mPFC and a region in the right posterior parietal cortex anatomical cluster that included mPFC and the ventral ante-
(0 voxels, with a cluster-determining threshold of p < .005, rior cingulate gyrus (0 voxels, with a cluster-determining
uncorrected) was observed. There were no other significant threshold of p < .005, uncorrected; Fig. 3b). There were no
results. other significant results. However, there was a tendency
toward a WMT-related increase in resting-FC between the
3.4. The effect of WMT on resting-FC with the right right DLPFC and the left posterior parietal cortex (62 voxels,
DLPFC with a cluster-determining threshold of p < .005, uncorrected).
To sum up the results of the effect of WMT on resting-FC
Next, we compared changes in resting-FC with the right with mPFC and those of the effect of WMT on resting-FC with
DLPFC. The right DLPFC is chosen as the seed ROI for EAS mPFC, regardless of the removal of the whole brain signal in
because the DLPFC is the key node of the EAS and as described individual analyses, WMT increased resting-FC between mPFC
below the right DLPFC showed resting-rCBF changes following and the precuneus (within network resting-FC). However,
WMT. when the whole brain signal was not regressed out in indi-
When the images of resting-FC in which the whole brain vidual analyses, a WMT-related decrease in resting-FC between
signal was regressed out were used, this analysis revealed mPFC and the nodes of EAS (WMT-related increase in

Fig. 3 e (a) WMT-related changes in resting-FC. Decrease in resting-FC with right DLPFC in the WMT group compared with
the control group. Results are shown with a threshold of p < .005, uncorrected. Compared with no-intervention, WMT
resulted in a decrease in resting-FC between right DLPFC and mPFC consistent with the tendency of the results seen in
Fig. 2. These are the resting-FC results when the time course of the whole brain signal was regressed out in individual
analyses. (b) These are the resting-FC results when the time course of the whole brain signal was not regressed out in
individual analyses. Everything else is presented in the same manner as for (a). The significant results and tendencies seen
in (a) were not seen in this analysis.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
12 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

anticorrelations between the nodes of EAS and the node of toward a positive correlation between improvements in
DMN), which was observed when the whole brain signal was performance on WM tasks and the amount of rGMV changes
regressed out in individual analyses, was not detected. in significant clusters in the frontal area (r ¼ .18, p ¼ .28), left
temporal area (r ¼ .20, p ¼ .23), right parietal area (r ¼ .22,
3.5. The effect of WMT on resting-rCBF p ¼ .18), left parietal area (r ¼ .25, p ¼ .11), and left caudate
(r ¼ .26, p ¼ .10). No significant correlation was observed
Next, we compared changes in resting-rCBF in the WMT and between the amount of resting-CBF changes in the significant
control groups. This analysis revealed a statistically signifi- clusters identified in resting-CBF analysis and the amount of
cant WMT-related increase from pre- to post-measures in resting-FC changes in any of the three significant clusters
resting-rCBF in the right LPFC (including the right inferior and identified in resting-FC analyses ( p > .05). This may be due to
middle frontal gyri) (x, y, z ¼ 34, 26, 26, t ¼ 3.51, p ¼ .031, cor- a lack of effective training-related variance (see Results of
rected for multiple comparisons at the cluster level with training data) since the amount of training was controlled.
a cluster-determining threshold of p < .005, uncorrected; Next, simple regression analyses that tested for correlations
Fig. 4). No region showed statistically significant WMT-related between improvements in RAPM performance and the amount
decreases in resting-rCBF. of neural change in significant clusters identified in the above-
mentioned ANCOVAs were performed. No significant correla-
3.6. The effect of WMT on rGMV tions were detected for any of the clusters. Regarding resting-
state measures and RAPM, among all the subjects analyzed in
Next, we compared changes in rGMV in the WMT and control this study, resting-FC between mPFC and the right posterior
groups. This analysis revealed a statistically significant WMT- parietal cortex before the intervention was significantly and
related increase from pre- to post-measures in rGMV in an negatively correlated with performance on RAPM before the
extensive anatomical cluster that included the bilateral intervention (r ¼ .508, t ¼ 4.53, p ¼ 2.89  105). This pattern
DLPFC, the left ventral LPFC, the bilateral lateral rostral PFCs, (a negative correlation between a psychometric intelligence
mPFC, the anterior cingulate cortex, the left perisylvian area, measure and resting-FC between nodes of DMN and EAS) is
the superior parts of the bilateral parietal cortices, an congruent with that observed in a previous report (Song et al.,
anatomical cluster around the left caudate, the right cere- 2008). However, there were no other significant results, and
bellum, and an anatomical cluster that included regions in the given the countless number of regression tests which were not
left middle and inferior temporal gyri ( p < .05, corrected for planned or designed before the experiment, the result may have
multiple comparisons at the non-isotropic adjusted cluster to be interpreted somewhat cautiously.
level with an underlying voxel-level of p < .05 corrected for There could be a number of reasons for the observed lack of
a false discovery rate; Fig. 5a and b, Table 4). No region showed correlation between training-related RAPM changes and
statistically significant WMT-related decreases in rGMV. neural changes. One possible reason is a lack of statistical
power because we could only analyze data from subjects in
3.7. Regression analysis the WMT group. On the other hand, large individual differ-
ences in performance changes were observed in the control
Simple regression analyses that tested correlations between group, indicating that much of the variance in performance
improvements in performance on WMT tasks and the amount changes in the WMT group may not have been related to
of neural changes in the significant clusters identified in the WMT, either. In addition, this experiment was not designed to
abovementioned ANCOVAs were also performed. The results perform this type of correlation analysis. While the amount of
revealed no significant correlations in any of the clusters. training is correlated with training-related neural changes
There was, however, a somewhat insignificant tendency (Takeuchi et al., 2010a), in this study, we strictly controlled the

Fig. 4 e Increase in resting-rCBF in the WMT group compared with the control group ( p < .05, corrected for multiple
comparisons at cluster size with an underlying voxel-level of p < .005, uncorrected). Compared with no-intervention, WMT
resulted in an increase in resting-rCBF in the right LPFC. The scale cannot be presented due to the limitation of the software.
The density of the red color represents both T values and depth of the voxels.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 13

Fig. 5 e Increase in rGMV in the WMT group compared with the control group ( p < .05, corrected for multiple comparisons
at the non-isotropic adjusted cluster level with an underlying voxel-level of p < .05, corrected for false discovery rate). (a)
Compared with no-intervention, WMT resulted in an increase in rGMV in the bilateral extensive lateral prefrontal areas, left
perisylvian area, bilateral parietal regions, left middle/temporal gyrus, and right cerebellum. (b) Compared with no-
intervention, WMT also resulted in an increase in rGMV in the bilateral anterior cingulate cortex and left caudate. The
intensity of the color represents the T values in Figures (a) and (b). However, the scaling is different in the two figures, and
the same intensity does not represent the same T value in the two figures. This cannot be altered because of a limitation of
the software.

amount of training in the WMT group. Therefore, little vari-

ance was observed among subjects in this respect. To perform 4. Discussion
these types of correlation analyses successfully, we should
not have controlled the amount of training so strictly. The present study revealed the effect of WMT on resting-FC,
Furthermore, all subjects in the WMT group underwent pro- resting-rCBF and rGMV. WMT (a) increases resting-FC
longed and intense training for 1 month, whereas the control between mPFC and the precuneus, (b) decreases resting-FC
group received no such training. However, the significance of between mPFC and the right posterior parietal cortex/right
group differences in RAPM was only marginal. This means LPFC, and (c) increases resting-rCBF in the right LPFC.
that the ratio of meaningful variance in the WMT group to However, training-related changes in resting-FC between the
meaningless variance in the control group was low. nodes of the different networks (b) were only observed when

Table 4 e WMT-related rGMV changes when compared with the control group.

Area MNI Coordinates T score Corrected p value

(cluster)
x y z

Middle frontal gyrus(B)/Superior frontal gyrus(B)/Inferior B 37 19 35 5.06 <.001

frontal rygus(L)/Anterior cingulate cortex (B)/Medial
frontal gyrus/Superior temporal gyrus(L)/Rolandic operculum(L)
Superior parietal lobule/Postcentral gyrus L 21 49 75 4.46 <.001
Middle temporal gyrus/Inferior temporal gyrus/Middle temporal pole L 59 0 32 4.31 <.001
Extra-nuclear/Caudate L 9 22 10 4.17 <.001
Precentral gyrus/Postcentral gyrus/Paracentral lobule R 16 30 73 4.05 <.001
Cerebellum R 18 78 37 3.97 <.001

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
14 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

the whole brain signal was regressed out in individual anal- because many cognitive and neural mechanisms may
yses. They were not observed when the whole brain signal underlie differences in resting-rCBF, the precise mechanisms
was not regressed out in individual analyses. On the other by which resting-rCBF is associated with conditions with
hand, training-related changes in resting-FC within the reduced WMC cannot be revealed from this study. However,
network (a) were not affected by whether the whole brain one possible mechanism is that increases in blood vessels in
signal was regressed out in individual analyses. Furthermore, the relevant region may lead to increased resting-rCBF in the
in this study WMT led to increase of rGMV in the extensive region and improved performance in demanding tasks by
regions in the bilateral LPFCs, the mPFC and the anterior allowing a rich supply of rCBF. The other possible mechanism
cingulate cortex as well as the bilateral parietal cortices, the is that increased neuronal structures in the relevant regions
left perisylvian area, the right cerebellum and left middle and facilitate cognitive performance and increase metabolic
inferior temporal gyrus. demand, which may lead to increased resting-rCBF. By
The lower resting-FC among the DMN nodes, higher utilizing these mechanisms, resting-FC among the DMN
resting-FC between the DMN nodes and the node of the EAS nodes, anticorrelations between DMN and the EAS, and lower
(reduced anticorrelations between the two networks), and resting-rCBF in LPFC may be associated with WMC or condi-
lower resting-rCBF in PFC are all associated with conditions tions with reduced WMC.
with reduced or lowered WMC. Thus, WMT-related changes in We speculate that WMT-related increases in resting-FC
these resting-FC and resting-rCBF may underlie or reflect an between the key DMN nodes and a decrease in resting-FC
increase in WMC following WMT and may provide insights between mPFC and nodes of the EAS (LPFC and the posterior
into the application of this training for many conditions with parietal cortex) may be mediated by a primary change in the
reduced WMC. A lower resting-FC among the DMN nodes is function and structure of LPFC induced by WMT. WMT leads
associated with lower individual WMC (Hampson et al., 2006) to changes in the function, structure, and neurochemistry of
and generally [for comments on discrepancies among studies, the EAS (Takeuchi et al., 2010b). However, in this study,
see (Broyd et al., 2009)] with many conditions with reduced changes were observed in resting-FC within the network and
WMC, which include Alzheimer’s disease, schizophrenia, between networks involving DMN. These changes might be
autism spectrum disorder, attention deficit/hyperactivity caused by the primary change in LPFC because LPFC of the EAS
disorder, and aging (Baddeley, 2003; Broyd et al., 2009; has been suggested to be the region that suppresses the
Goldman-Rakic, 1994; Sambataro et al., 2010; Steele et al., activity of DMN when it is activated (Greicius et al., 2003).
2007; Westerberg et al., 2004; Wingfield et al., 1988). Albeit Following WMT, LPFC whose functions are augmented
less frequently, reduced anticorrelations between the two through training may increasingly regulate DMN activation.
networks are observed in schizophrenia, autism spectrum These changes in the regulation of DMN activation through
disorder, attention deficit/hyperactivity disorder, and aging LPFC may lead to increases in anticorrelations between the
(Broyd et al., 2009; Sambataro et al., 2010; Whitfield-Gabrieli EAS and DMN and increases in resting-FC within DMN.
et al., 2009). Finally, reduced rCBF in PFC has been reported The present results complement a series of recent findings
in schizophrenia (Weinberger et al., 1986), Alzheimer’s disease that have shown experience-dependent change in resting-FC.
(Johnson et al., 2005), and normal aging (Martin et al., 1991). In previous studies, motor learning (Albert et al., 2009), motor
Thus, these observed WMT-related changes in resting-FC and training (Taubert et al., 2011), and visual perceptual learning
resting-rCBF may underlie the increase in WMC and amelio- (Lewis et al., 2009) were shown to induce changes in resting-FC
ration of certain disorders with reduced WMC, such as within networks thought to be involved with tasks or resting-FC
attention deficit/hyperactivity disorder, following WMT between those networks and other networks. Our results are
(Klingberg, 2010) and may provide insights into the application congruent with these findings because EAS plays a key role in
of this training for many other conditions with reduced WMC. WM and DMN may also be actively involved in WM (Hampson
Resting-FC among the DMN nodes, anticorrelations et al., 2006; Owen et al., 2005). We showed here for the first
between DMN and the EAS, and lower resting-rCBF in LPFC time that WMT, which is associated with improvements in
may be specifically associated with WMC or conditions with general cognitive abilities (Jaeggi et al., 2008; Klingberg et al.,
reduced WMC in their respective ways. Resting-FC between 2005), can induce a change in resting-FC between two intrinsic
mPFC and the precuneus and FC between the two regions networks, DMN and EAS, both of which are associated with
during a WM task are correlated with individual WM perfor- general cognitive abilities (Hampson et al., 2006; Song et al.,
mance (Hampson et al., 2006). Thus, although the precise 2008).
mechanism remains unknown, the integrated DMN has been WMT-related decreases in resting-FC between the nodes of
suggested to facilitate the performance of demanding cogni- the different networks (increased anticorrelations between
tive tasks (Hampson et al., 2006). Regarding anticorrelations nodes of EAS and DMN) were not observed when the whole
between DMN and the EAS, deactivation of DMN during brain signal was not regressed out in individual analyses of
cognitive tasks is considered to reflect reallocation of cogni- resting-FC; this could be due to the effects of changes in
tive resources from the network active during rest (DMN) to aspects of global brain activity. As described in the Methods,
the network actively involved in the task (EAS) (McKiernan regressing out the whole brain signal in individual resting-FC
et al., 2003). Thus, increased anticorrelations between the has been criticized as “artifactually” creating anticorrelations
two networks may reflect conditions in which cognitive (Murphy et al., 2009) although a recent study suggested that
resources are easily reallocated from one network to the anticorrelations observed in resting-state connectivity are not
other, and these conditions may help one successfully an artifact introduced by global signal regression and might
perform cognitive tasks using rich cognitive resources. Finally, have biological origins (Chai et al., 2012). The controversy is

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 15

ongoing. In this study, we cannot enter into in-depth discus- Experience-dependent neural plasticity involves increases in
sions related to this controversy because we did not measure the width and density of the capillary (Borowsky and Collins,
respiratory- or physiologically related signals and therefore 1989; Sirevaag and Greenough, 1987). It also involves increases
cannot regress out these signals. Moreover, since this in the number and volume of glia (Diamond et al., 1966) and
controversy is new, we do not know much about the meaning mitochondria, respectively (Sirevaag and Greenough, 1987),
of the different results obtained when the whole brain signal which leads to increases in metabolic demand. WMT may also
was regressed out in individual analyses and when it was not. increase resting-rCBF through these changes in the capillary
However, a global signal correlation was observed partly and increases in metabolic demand.
because the brain was globally activated (Schölvinck et al., In our previous study, relatively mild WMT led to an
2010). Thus, when these global brain activities are not increase in rGMV in regions of EAS in which rGMV decreased
regressed out, FC between two brain regions shows correlated after a 1-week intensive program of concentrated WMT, sug-
activity because of global brain activity as well as because of gesting that training protocols affect whether rGMV increases
region- or network-specific activity synchronization, but or decreases in EAS following WMT. In this study, relatively
when these global brain activities are regressed out, FC mild WMT (up to 1 h per day for 4 weeks) led to increased
between the two brain regions shows correlated activity rGMV in bilateral DLPFCs, superior parts of the bilateral pari-
because of network-specific activity synchronization. In this etal cortices, and the left perisylvian area. These regions
case, the significant WMT-related results in resting-FC anal- showed a decrease in rGMV after the 1-week intensive
yses in which the whole brain signal was regressed out can be concentrated WMT (4-h training per day) in our previous study
explained by WMT-related changes in region- or network- (Takeuchi et al., 2010b). These findings are congruent with our
specific activity synchronization. The difference in results hypothesis that cognitive training may lead to nonlinear
relating to WMT-related changes in resting-FC when the changes in rGMV (an initial increase followed by a decrease)
whole brain signal is regressed out and when it is not would be that are affected by training length and intensity (greater
due to some aspect of WMT-related change in global brain intensity leads to more rapid nonlinear change) (Takeuchi
activity. On the other hand, if WMT simply increased the et al., 2011b). The findings are also consistent with those of
effects of global brain activity on resting-FC uniformly, then previous studies in which relatively mild training led to
the WMT-related results of resting-FC when the whole brain increased rGMV (Draganski et al., 2004; Driemeyer et al., 2008)
signal was not regressed out would show larger WMT-related as well as our previous cognitive training study in which a 1-
increases in resting-FC between mPFC and the precuneus; week intensive concentrated WMT (4-h training per day) led
however, no such difference was observed. One possible mainly to a decrease in rGMV (Takeuchi et al., 2011b).
reason for the differences in the abovementioned results is WMT also led to changes in performance on certain
that after WMT, the whole brain signal conforms to EAS cognitive tests. Our previous review (Takeuchi et al., 2010b)
activity (but not DMN activity). In other words, the contribu- showed that the effects of WMT on some cognitive functions
tion of EAS activity to the whole brain signal might increase vary between studies, the differences between this study and
after WMT. If this happens, the removal of the whole brain other studies (Dash et al., 2010; Jaeggi et al., 2008; Takeuchi
signal would not affect the training-related results regarding et al., 2011d) may be explained by difference of subjects’
resting-FC between mPFC and the precuneus (within DMN characteristics, training tasks and statistical deviations. Of
resting-FC). On the other hand, in this case, since DMN activity note, in this study, WMT led to average increase in perfor-
must always contribute to (correlate with) global brain mance on the creativity test. This is in clear contrast to the
activity, when the whole brain signal was not regressed out, it result of our previous study of WMT involving mental calcu-
seems after WMT, the relationship between EAS and DMN lations, which showed a clear training-related decrease in
activities must shift toward an increase in resting-FC between creativity. One possible cause of this discrepancy in WMT
these networks although this shift can be explained by an tasks could be division of attention. In this study, we included
increase in the correlation between EAS activity and global dual WM and dual N-back tasks, both of which apparently
brain activity alone. Consistent with this notion, when the require divided attention. In our previous study, the main
whole brain signal was not regressed out, the tendency training task was mental multiplication, which may require an
toward a WMT-related increase in resting-FC with the right extended period of focused attention with no distractions to
DLPFC was observed in a substantially larger area and across avoid losing retained information. On the other hand, crea-
a multiple bilateral brain areas compared with when the tivity has been suggested to be associated with widened
whole brain signal was regressed out (Fig. 6a). In fact, in attention, which allows one to process a lot of information
the former analysis, the largest cluster spread across multiple concurrently (Mendelsohn, 1976), but also with impaired
bilateral areas was significant when the cluster size test using selective attention (Necka, 1999). Thus, cognitive training that
a cluster-determining threshold of p < .05 uncorrected was requires divided attention may improve creativity, while
used ( p ¼ 2.50  106, corrected at the cluster level under the cognitive training that requires focused attention may
cluster-determining threshold of p < .05, uncorrected). suppress creativity. WMT has previously been shown to
However, no such difference was observed in the case of the increase performance on non-verbal reasoning tasks
analyses of WMT-related increases in resting-FC with mPFC (Takeuchi et al., 2010b), but this finding was recently contested
(Fig. 6b). Thus, WMT might lead to a weak increase in the (Redick et al., in press). Our study results relating to this aspect
resting-FC between the nodes of EAS and multiple areas. are weak and inconclusive, but given the recent controversy,
WMT may increase resting-rCBF in the right LPFC through they may be worth noting. First, the Tanaka B-type intelligence
changes in the capillary and increases in metabolic demand. test consists of several speeded tasks (although these include

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
16 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

Fig. 6 e Tendencies of WMT-related increases in resting-FC. (a) Increase in resting-FC with right DLPFC in the WMT group
compared with the control group. Results are shown with a threshold of p < .05, uncorrected. Red areas show resting-FC
when the time course of the whole brain signal was regressed out in individual analyses. Green areas show resting-FC
when the time course of the whole brain signal was not regressed out in individual analyses. As can be seen, when the
whole brain signal was not regressed out, the tendency toward a WMT-related increase in resting-FC with the right DLPFC
covered a substantially larger total area and included more brain areas than when the whole brain signal was regressed out.
(b) Increase in resting-FC with the mPFC in the WMT group compared with the control group. Results are shown with
a threshold of p < .05, uncorrected. Red areas show resting-FC when the time course of the whole brain signal was
regressed out in individual analyses. Green areas show resting-FC when the time course of the whole brain signal was not
regressed out in individual analyses. As can be seen, no substantial difference between red and green areas was observed.

complex speeded tasks) (Takeuchi et al., 2011b), and the nature fluid intelligence tasks have recently been contested and there
of this test content differs from that in RAPM and BOMAT. The may be a number of reasons for the lack of significant effects
insignificant results relating to the Tanaka intelligence test on BOMAT or other non-verbal reasoning tasks (Redick et al., in
observed in this study may be somewhat congruent with the press; Takeuchi et al., 2010b). Our previous study of WMT
results of our previous review (Takeuchi et al., 2010b), which involving mental calculations (Takeuchi et al., 2011d) involved
concluded that WMT may not affect performance on pro- an active control group, an intervention group, and testers
cessing speed tasks in normal subjects. WMT led to improved blinded to the group assignments, and this study failed to
performance on RAPM but not on BOMAT. This finding show any significant effects of training on RAPM although the
contrasts with the results of previous studies that reported training led to a marked increase in performance on the letter
significant effects of WMT involving dual N-back or dual WM span task, which has barely any components in common with
tasks on BOMAT (Jaeggi et al., 2008; Takeuchi et al., 2010b). the training tasks. In the present study, we performed two
Although the exact cause of this discrepancy is unclear, the non-verbal reasoning fluid intelligence tests and only one was
performance of subjects on BOMAT in this study was lower weakly significant; the other showed no hint of any training
than that in the previous study, and this might have made the effects. These discrepancies could be due to subject charac-
test unsuitable for assessing performance (Jaeggi et al., 2008). teristics such as a tendency toward attention deficits as WMT
The difference may also be due to lower testeretest reliability may affect attention (Takeuchi et al., 2010b). In that case, the
due to the unusual manner in which the test was adminis- results may differ based on country because of great variation
tered, but the low number of subjects in the no-intervention in the prevalence of attention deficit hyperactivity disorder
group made it difficult to reliably assess testeretest reli- across countries (Skounti et al., 2007), or even within the same
ability. On the other hand, although several studies have country based on how the subjects were recruited. Thus, the
showed the effects of WMT on Raven matrix tests (Takeuchi effects of WMT on non-verbal reasoning fluid intelligence
et al., 2010b), the effects of WMT on non-verbal reasoning tasks remain elusive.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 17

One limitation of this study was its complex training questionnaire measures of cognition as it would be much
protocols (Jaeggi et al., 2008; Tang et al., 2007). These chal- easier to report that they became smarter than to actually
lenges are common in this type of study in which the training show that they became smarter. We are not trying to say that
is related to WMT (Jaeggi et al., 2008), video games (Green and active control training has been proven to be ineffective in all
Bavelier, 2003) or meditation (Tang et al., 2007). These types of existing behavioral and imaging measures because this is
studies typically have no strict control groups or any other impossible. Instead, we are saying that there is no scientific
conditions as those in normal fMRI studies. Although it would evidence indicating that there are effects in what the active
be statistically challenging, it would be interesting to disen- control group is trying to control for (fatigue, expectancy
tangle the multiple complex cognitive training protocols (for effects, computer use, contact with experimenters, etc.) in
example, do auditory and visual WMT have different effects?) experiments such as this one. On the other hand, what the
and investigate the effect of each component of training in the active control group is “not” trying to control for does have
future. Finally, with regard to the control group, unlike our effects on behavioral and imaging measures in some cases.
previous study (Takeuchi et al., 2011d), we did not include an Video games, learning of substantially new knowledge or
active control group because all the available evidence indi- skills, and low-level cognitive training in subjects with
cate that an active control group does not have any impact in compromised cognitive abilities have also been used as active
this type of study. It is true that under certain conditions, the control training (Takeuchi et al., 2010b). However, there is
components that an active control group tries to control can evidence indicating that these interventions affect cognitive
affect behavioral performance. For example, taking placebo performance and brain structures (Takeuchi et al., 2010b).
drugs can affect cognitive performance under certain condi- Because of such limitations, we cautioned against the too easy
tions (Oken, 2008; Oken et al., 2008), whereas active control use of active control groups in imaging studies in our previous
training does not affect cognitive performance, especially in review of WMT (Takeuchi et al., 2010b). The lack of an active
young adults (Takeuchi et al., 2011d). Training in computer control group may be a limitation of this study, but if so, then
use may also involve many learning processes for individuals all intervention studies of cognitive training published to date
who are not used to using a computer, which may affect will have bigger or equivalent limitations, and if we use an
neural mechanisms. However, the use of computerized active control group in imaging intervention studies, we
training tasks in young adults does not have such effects usually end up having to face other possible limitations
(Takeuchi et al., 2011d). Contact with the experimenter or depending on the purpose of the study. The inclusion of two
another individual may be meaningful for older adults who (active and passive) control groups may be ideal, but practi-
have little contact with others (Bassuk et al., 1999). However, cally, the design would have a lower statistical power
we included young, healthy university students who were (Takeuchi et al., 2010b).
willing to take part in the experiment alone, and for these We discussed the likelihood of the effects of factors that
individuals, contact with the experimenter does not have any active control groups are usually trying to control for above.
effect on cognitive performance (Takeuchi et al., 2011d). Nonetheless, it is still a limitation of our study that the
Besides, the only prolonged social interaction with the experimental group had a much different experience than the
experimenters occurred during the training period when the control group. For resting-FC, for example, resting-FC within
training tasks were explained on the first day, and this period DMN and EAS is associated with a wide range of diseases and
ended after less than 20 min. Furthermore, the additional cognitive functions (Broyd et al., 2009; Hampson et al., 2006;
analysis of the profile of mood states (Yokoyama, 2005) per- Song et al., 2008; Takeuchi et al., in press). Thus, although the
formed on the first and last days of the experiment (Takeuchi effects of some motor and other perceptual training appear to
et al., 2011a) revealed that subjects in the WMT group did not be related to resting-FC involving networks that are thought to
show significant WMT-related improvements in vigor be involved with such tasks some other interventions may
( p ¼ .235, one-tailed ANCOVA), which may well be related to also affect resting-FC of EAS and DMN, as shown in this study.
motivation or any WMT-related increase in fatigue ( p ¼ .919, Thus, future studies need to clarify this area.
one-tailed ANCOVA, indicating that, if anything, WMT was
associated with reduced fatigue). These results indicate that
chronic fatigue induced by cognitive training and increased 5. Summary and conclusion
motivation caused by training or expectancy effects can be
ruled out (although subjects may have been specifically In summary, this is the first study to investigate the effects of
motivated during certain measures). We can also say that the WMT on resting-FC and resting-rCBF. The present results
subjects were probably not trying to show that they were show that WMT increases resting-FC between the key DMN
smarter, determined as per our analysis of the questionnaire nodes, decreases resting-FC between the key DMN node and
responses. For one related measure, we administered nodes of the EAS, and increases resting-rCBF in the right LPFC.
a psychological questionnaire relating to critical thinking Lower resting-FC among the DMN nodes, higher resting-FC
(Hirayama and Kusumi, 2004). When the score for this ques- between DMN and the EAS, and lower resting-rCBF in PFC
tionnaire was analyzed like the other psychological measures, have been observed in many conditions with reduced WMC.
no significant effects of WMT were detected, and the average Our results indicated resting-state neural mechanisms which
score after the training period was actually lower in the WMT are assumed to reflect brain’s intrinsic activities and connec-
group. If the subjects were attempting to show that their tivities are affected by cognitive training. However, the
cognitive performance had improved because of the training, training-related decreases in resting-FC between the key DMN
the training would have had a stronger effect on node and the nodes of EAS was only observed when the whole

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
18 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

brain signal was regressed out in individual analyses, and Corbetta M and Shulman GL. Control of goal-directed and
these changes were not observed when the whole brain signal stimulus-driven attention in the brain. Nature Reviews
was not regressed out in individual analyses. Further analyses Neuroscience, 3(3): 201e215, 2002.
Dahlin E, Neely AS, Larsson A, Backman L, and Nyberg L. Transfer
indicated that these differences may be mediated by a weak
of learning after updating training mediated by the striatum.
but a widespread increase in resting-FC between the nodes of Science, 320(5882): 1510e1512, 2008.
EAS and activity of multiple bilateral areas across the brain. Dash D, McNab F, and Klingberg T. Effect of dual working memory
And, the present results may provide insight into the appli- training on a reasoning task e An fMRI study. 2010.
cation of WMT for many conditions with reduced WMC. Della-Maggiore V, Chau W, Peres-Neto PR, and McIntosh AR. An
Moreover, the WMT program used in this study was relatively empirical comparison of SPM preprocessing parameters to the
mild and led to an increase in rGMV in regions of EAS, whereas analysis of fMRI data. NeuroImage, 17(1): 19e28, 2002.
Diamond MC, Law F, Rhodes H, Lindner B, Rosenzweig MR,
a decrease in rGMV after a 5-day intensive (4-h training per
Krech D, et al. Increases in cortical depth and glia numbers in
day) WMT program involving mental calculations was rats subjected to enriched environment. The Journal of
observed in similar regions in our previous study. This result is Comparative Neurology, 128(1): 117e125, 1966.
congruent with our hypothesis that training intensity deter- Draganski B, Gaser C, Busch V, Schuierer G, Bogdahn U, and
mines whether rGMV increases or decreases after training. May A. Neuroplasticity: Changes in grey matter induced by
training. Nature, 427(6972): 311e312, 2004.
Driemeyer J, Boyke J, Gaser C, Büchel C, and May A. Changes in
gray matter induced by learning e Revisited. PLoS One, 3(7):
e2669, 2008.
Acknowledgments Fox MD, Corbetta M, Snyder AZ, Vincent JL, and Raichle ME.
Spontaneous neuronal activity distinguishes human dorsal
We thank Yuki Yamada for operating the MRI scanner, Sarah and ventral attention systems. Proceedings of the National
Michael for checking the English of this manuscript, Haruka Academy of Sciences, 103(26): 10046e10051, 2006.
Nouchi for conducting the psychological tests, the subjects, Fox MD and Raichle ME. Spontaneous fluctuations in brain
activity observed with functional magnetic resonance
and all our other colleagues in IDAC, Tohoku University for
imaging. Nature Reviews Neuroscience, 8(9): 700e711, 2007.
their support. This study was supported by JST/RISTEX, JST/
Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, and
CREST and a Grant-in-Aid for Young Scientists (B) (KAKENHI Raichle ME. The human brain is intrinsically organized into
23700306) from the Ministry of Education, Culture, Sports, dynamic, anticorrelated functional networks. Proceedings of the
Science, and Technology (MEXT). National Academy of Sciences, 102(27): 9673e9678, 2005.
Friston KJ, Holmes A, Poline JB, Price CJ, and Frith CD. Detecting
activations in PET and fMRI: Levels of inference and power.
NeuroImage, 4(3): 223e235, 1996.
references Gaser C. VBM Toolbox for SPM2, VBM Toolbox for SPM5, http://
dbmneurouni-jenade/vbm/; 2007.
Genovese CR, Lazar NA, and Nichols T. Thresholding of statistical
Albert NB, Robertson EM, and Miall RC. The resting human brain maps in functional neuroimaging using the false discovery
and motor learning. Current Biology, 19(12): 1023e1027, 2009. rate. NeuroImage, 15(4): 870e878, 2002.
Ashburner J and Friston KJ. Voxel-based morphometry e The Goldman-Rakic PS. Working memory dysfunction in
methods. NeuroImage, 11(6): 805e821, 2000. schizophrenia. The Journal of Neuropsychiatry and Clinical
Baddeley A. Working memory: Looking back and looking forward. Neurosciences, 6(4): 348e357, 1994.
Nature Reviews Neuroscience, 4(10): 829e839, 2003. Green CS and Bavelier D. Action video game modifies visual
Bassuk SS, Glass TA, and Berkman LF. Social disengagement and selective attention. Nature, 423(6939): 534e537, 2003.
incident cognitive decline in community-dwelling elderly Greicius MD, Krasnow B, Reiss AL, and Menon V. Functional
persons. Annals of Internal Medicine, 131(3): 165e173, 1999. connectivity in the resting brain: A network analysis of the
Beech A and Claridge G. Individual differences in negative default mode hypothesis. Proceedings of the National Academy of
priming: Relations with schizotypal personality traits. British Sciences, 100(1): 253e258, 2003.
Journal of Psychology, 78(3): 349e356, 1987. Hakoda Y and Sasaki M. Group version of the Stroop and reverse-
Borowsky IW and Collins RC. Metabolic anatomy of brain: A Stroop test: The effects of reaction mode, order and practice.
comparison of regional capillary density, glucose metabolism, Kyoikushinrigakukenkyu (Educational Psychology Research), 38(4):
and enzyme activities. The Journal of Comparative Neurology, 389e394, 1990.
288(3): 401e413, 1989. Hampson M, Driesen NR, Skudlarski P, Gore JC, and Constable RT.
Broyd SJ, Demanuele C, Debener S, Helps SK, James CJ, and Brain connectivity related to working memory performance.
Sonuga-Barke EJS. Default-mode brain dysfunction in mental Journal of Neuroscience, 26(51): 13338, 2006.
disorders: A systematic review. Neuroscience & Biobehavioral Hayasaka S, Phan KL, Liberzon I, Worsley KJ, and Nichols TE.
Reviews, 33(3): 279e296, 2009. Nonstationary cluster-size inference with random field and
Buckner RL, Andrews-Hanna JR, and Schacter DL. The brain’s permutation methods. NeuroImage, 22(2): 676e687, 2004.
default network (The Year in Cognitive Neuroscience 2008). Hirayama R and Kusumi T. Effect of critical thinking disposition
Annals of the New York Academy of Sciences, 1124: 1e38, 2008. on interpretation of controversial issues: Evaluating evidences
Casanova R, Srikanth R, Baer A, Laurienti PJ, Burdette JH, and drawing conclusions. Japanese Journal of Educational
Hayasaka S, et al. Biological parametric mapping: A statistical Psychology, 52(2): 186e198, 2004.
toolbox for multimodality brain image analysis. NeuroImage, Hossiep R, Turck D, and Hasella M. Bochumer Matrizentest: BOMAT-
34(1): 137e143, 2007. Advanced-Short Version. Gottingen: Hogrefe, 1999.
Chai XJ, Castañón AN, Öngür D, and Whitfield-Gabrieli S. Ilg R, Wohlschlager AM, Gaser C, Liebau Y, Dauner R, Woller A,
Anticorrelations in resting state networks without global et al. Gray matter increase induced by practice correlates with
signal regression. NeuroImage, 59(2): 1420e1428, 2012. task-specific activation: A combined functional and

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0 19

morphometric magnetic resonance imaging study. Journal of Petersen ET, Zimine I, Ho YC, and Golay X. Non-invasive
Neuroscience, 28(16): 4210e4215, 2008. measurement of perfusion: A critical review of arterial spin
Jaeggi SM, Buschkuehl M, Jonides J, and Perrig WJ. Improving labelling techniques. British Journal of Radiology, 79(944):
fluid intelligence with training on working memory. 688e701, 2006b.
Proceedings of the National Academy of Sciences, 105(19): Pochon JB, Levy R, Fossati P, Lehericy S, Poline JB, Pillon B, et al.
6829e6833, 2008. The neural system that bridges reward and cognition in
Johnson NA, Jahng GH, Weiner MW, Miller BL, Chui HC, Jagust WJ, humans: An fMRI study. Proceedings of the National Academy of
et al. Pattern of cerebral hypoperfusion in Alzheimer disease Sciences of the United States of America, 99(8): 5669e5674, 2002.
and mild cognitive impairment measured with arterial spin- Raven J. Manual for Raven’s Progressive Matrices and Vocabulary
labeling MR imaging: Initial experience. Radiology, 234(3): Scales. Oxford: Oxford Psychologists Press, 1998.
851e859, 2005. Reber PJ, Wong EC, Buxton RB, and Frank LR. Correction of off
Klingberg T. Training and plasticity of working memory. Trends in resonance-related distortion in echo-planar imaging using
Cognitive Sciences, 14(7): 317e324, 2010. EPI-based field maps. Magnetic Resonance in Medicine, 39(2):
Klingberg T, Fernell E, Olesen PJ, Johnson M, Gustafsson P, 328e330, 1998.
Dahlstrm K, et al. Computerized training of working memory Redick TS, Shipstead Z, Harrison TL, Hicks KL, Fried DE, Hambrick
in children with ADHD e A randomized, controlled trial. DZ, et al. No evidence of intelligence improvement after
Journal of the American Academy of Child and Adolescent working memory training: A randomized, placebo-controlled
Psychiatry, 44(2): 177e186, 2005. study. Journal of Experimental Psychology: General, in press.
Klingberg T, Forssberg H, and Westerberg H. Training of working Rose EJ and Ebmeier KP. Pattern of impaired working memory
memory in children with ADHD. Journal of Clinical and during major depression. Journal of Affective Disorders, 90(2e3):
Experimental Neuropsychology, 24(6): 781e791, 2002. 149e161, 2006.
Laird AR, McMillan KM, Lancaster JL, Kochunov P, Turkeltaub PE, Sambataro F, Murty VP, Callicott JH, Tan HY, Das S,
Pardo JV, et al. A comparison of label-based review and ALE Weinberger DR, et al. Age-related alterations in default mode
meta-analysis in the Stroop task. Human Brain Mapping, 25(1): network: Impact on working memory performance.
6e21, 2005. Neurobiology of Aging, 31(5): 839e852, 2010.
Lewis CM, Baldassarre A, Committeri G, Romani GL, and Sasaki M and Hakoda Y. The group version of the Stroop and
Corbetta M. Learning sculpts the spontaneous activity of the reverse-Stroop test (1). In Proceedings of Japan Educational
resting human brain. Proceedings of the National Academy of Psyhological 27th Annual Meeting, 1985: 208.
Sciences, 106(41): 17558e17563, 2009. Sasaki M, Hakoda Y, and Yamagami R. Schizophrenia and
Martin AJ, Friston KJ, Colebatch JG, and Frackowiak RSJ. Decreases reverse-Stroop interference in the group version of the Stroop
in regional cerebral blood flow with normal aging. Journal of and reverse-Stroop test. Shinrigaku kenkyu: The Japanese Journal
Cerebral Blood Flow and Metabolism, 11(4): 684e689, 1991. of Psychology, 64(1): 43e50, 1993.
McKiernan KA, Kaufman JN, Kucera-Thompson J, and Binder JR. A Schölvinck ML, Maier A, Ye FQ, Duyn JH, and Leopold DA. Neural
parametric manipulation of factors affecting task-induced basis of global resting-state fMRI activity. Proceedings of the
deactivation in functional neuroimaging. Journal of Cognitive National Academy of Sciences, 107(22): 10238e10243, 2010.
Neuroscience, 15(3): 394e408, 2003. Sirevaag AM and Greenough WT. Differential rearing effects on
McNab F, Varrone A, Farde L, Jucaite A, Bystritsky P, Forssberg H, rat visual cortex synapses. III. Neuronal and glial nuclei,
et al. Changes in cortical dopamine D1 receptor binding boutons, dendrites, and capillaries. Brain Research, 424(2):
associated with cognitive training. Science, 323(5915): 800e802, 320e322, 1987.
2009. Skounti M, Philalithis A, and Galanakis E. Variations in
Mendelsohn GA. Associative and attentional processes in creative prevalence of attention deficit hyperactivity disorder
performance. Journal of Personality, 44(2): 341e369, 1976. worldwide. European Journal of Pediatrics, 166(2): 117e123, 2007.
Murphy K, Birn RM, Handwerker DA, Jones TB, and Bandettini PA. Society_For_Creative_Minds Manual of S-A Creativity Test. Tokyo:
The impact of global signal regression on resting state Tokyo Shinri Corporation, 1969.
correlations: Are anti-correlated networks introduced? Song M, Zhou Y, Li J, Liu Y, Tian L, Yu C, et al. Brain spontaneous
NeuroImage, 44(3): 893e905, 2009. functional connectivity and intelligence. NeuroImage, 41(3):
Necka E. Creativity and attention. Polish Psychological Bulletin, 30: 1168e1176, 2008.
85e98, 1999. Stavridou A and Furnham A. The relationship between
Oken BS. Placebo effects: Clinical aspects and neurobiology. Brain, psychoticism, trait-creativity and the attentional mechanism
131(11): 2812e2823, 2008. of cognitive inhibition. Personality and Individual Differences,
Oken BS, Flegal K, Zajdel D, Kishiyama S, Haas M, and Peters D. 21(1): 143e153, 1996.
Expectancy effect: Impact of pill administration on cognitive Steele SD, Minshew NJ, Luna B, and Sweeney JA. Spatial working
performance in healthy seniors. Journal of Clinical and memory deficits in autism. Journal of Autism and Developmental
Experimental Neuropsychology, 30(1): 7e17, 2008. Disorders, 37(4): 605e612, 2007.
Oldfield RC. The assessment and analysis of handedness: The Stenger VA. Technical considerations for BOLD fMRI of the
Edinburgh inventory. Neuropsychologia, 9(1): 97e113, 1971. orbitofrontal cortex. In Zald DH and Rauch SL (Eds), The
Olesen PJ, Westerberg H, and Klingberg T. Increased prefrontal Orbitofrontal Cortex. London: Oxford University Press, 2006:
and parietal activity after training of working memory. Nature 423e446.
Neuroscience, 7(1): 75e79, 2004. Takeuchi H, Sekiguchi A, Taki Y, Yokoyama S, Yomogida Y,
Osaka M and Nishizaki Y. Brain and Working Memory (in Japanese). Komuro N, et al. Training of working memory impacts
Kyoto, Japan: Kyoto University Press, 2000. structural connectivity. Journal of Neuroscience, 30(9):
Owen AM, McMillan KM, Laird AR, and Bullmore E. N-back 3297e3303, 2010a.
working memory paradigm: A meta-analysis of normative Takeuchi H, Taki Y, Hashizume H, Sassa Y, Nagase T, Nouchi R,
functional neuroimaging studies. Human Brain Mapping, 25(1): et al. Cerebral blood flow during rest associates with general
46e59, 2005. intelligence and creativity. PLoS One, 6(9): e25532, 2011a.
Petersen ET, Lim T, and Golay X. Model-free arterial spin labeling Takeuchi H, Taki Y, Hashizume H, Sassa Y, Nagase T, Nouchi R,
quantification approach for perfusion MRI. Magnetic Resonance et al. Effects of training of processing speed on neural
in Medicine, 55(2): 219e232, 2006a. systems. Journal of Neuroscience, 31(34): 12139e12148, 2011b.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007
20 c o r t e x x x x ( 2 0 1 2 ) 1 e2 0

Takeuchi H, Taki Y, Hashizume H, Sassa Y, Nagase T, Nouchi R, Tanaka K, Okamoto K, and Tanaka H. Manual of New Tanaka B
et al. Failing to deactivate: The association between brain Type Intelligence Test. Tokyo: Kaneko Syobo, 2003.
activity during a working memory task and creativity. Tang YY, Ma Y, Wang J, Fan Y, Feng S, Lu Q, et al. Short-term
NeuroImage, 55(2): 681e687, 2011c. meditation training improves attention and self-regulation.
Takeuchi H, Taki Y, Hashizume H, Sassa Y, Nagase T, Nouchi R, Proceedings of the National Academy of Sciences, 104(43):
et al. The association between resting functional connectivity 17152e17156, 2007.
and creativity. Cerebral Cortex, in press. Taubert M, Lohmann G, Margulies DS, Villringer A, and Ragert P.
Takeuchi H, Taki Y, and Kawashima R. Effects of working Long-term effects of motor training on resting-state networks
memory training on cognitive functions and neural systems. and underlying brain structure. NeuroImage, 4(15): 1492e1498,
Reviews in the Neurosciences, 21(6): 427e450, 2010b. 2011.
Takeuchi H, Taki Y, Nagase T, Sassa Y, Hashizume H, and Uchida S and Kawashima R. Reading and solving arithmetic
Kawashima R. The effects of the training of simple processing problems improves cognitive functions of normal aged people:
speed tasks on brain structures. In 16th Annual Meeting of the A randomized controlled study. Age, 30(1): 21e29, 2008.
Organization for Human Brain Mapping. Barcelona, Spain, 2010c. Umemoto T, Osaka R, Kuraishi S, Sumida K, and Fujimoto M.
Takeuchi H, Taki Y, Sassa Y, Hashizume H, Sekiguchi A, Report of Kyodai SX intelligence test. Kyoikushinrigakunenpo (in
Fukushima A, et al. Regional gray matter volume of Japanese), 2(32): 138, 1963.
dopaminergic system associate with creativity: Evidence Weinberger DR, Berman KF, and Zec RF. Physiologic dysfunction
from voxel-based morphometry. NeuroImage, 51(2): 578e585, of dorsolateral prefrontal cortex in schizophrenia: I. Regional
2010d. cerebral blood flow evidence. Archives of General Psychiatry,
Takeuchi H, Taki Y, Sassa Y, Hashizume H, Sekiguchi A, 43(2): 114, 1986.
Fukushima A, et al. White matter structures associated with Westerberg H, Hirvikoski T, Forssberg H, and Klingberg T. Visuo-
creativity: Evidence from diffusion tensor imaging. spatial working memory span: A sensitive measure of
NeuroImage, 51(1): 11e18, 2010e. cognitive deficits in children with ADHD. Child
Takeuchi H, Taki Y, Sassa Y, Hashizume H, Sekiguchi A, Neuropsychology, 10(3): 155e161, 2004.
Fukushima A, et al. Working memory training using mental Whitfield-Gabrieli S, Thermenos HW, Milanovic S, Tsuang MT,
calculation impacts regional gray matter of the frontal and Faraone SV, McCarley RW, et al. Hyperactivity and
parietal regions. PLoS One, 6(8): e23175, 2011d. hyperconnectivity of the default network in schizophrenia
Takeuchi H, Taki Y, Sassa Y, Hashizume H, Sekiguchi A, and in first-degree relatives of persons with schizophrenia.
Fukushima A, et al. Regional gray matter density associated Proceedings of the National Academy of Sciences, 106(4):
with emotional intelligence: Evidence from voxel-based 1279e1284, 2009.
morphometry. Human Brain Mapping, 32(9): 1497e1510, 2011. Wingfield A, Stine EAL, Lahar CJ, and Aberdeen JS. Does the
Taki Y, Hashizume H, Sassa Y, Takeuchi H, Wu K, Asano M, et al. capacity of working memory change with age? Experimental
Correlation between gray matter density-adjusted brain Aging Research, 14(2): 103e107, 1988.
perfusion and age using brain MR images of 202 healthy Yokoyama K. POMS Shortened Version (in Japanese). Tokyo:
children. Human Brain Mapping, 32(11): 1973e1985, 2011. Kanekoshobo, 2005.

Please cite this article in press as: Takeuchi H, et al., Effects of working memory training on functional connectivity and cerebral
blood flow during rest, Cortex (2012), http://dx.doi.org/10.1016/j.cortex.2012.09.007