In accordance with the Declaration of Helsinki (1991), Written informed consent was obtained from each subject. This study was approved by the Ethics Committee of Tohoku University.

Fifty-five healthy, right-handed individuals (42 men and 13 women) participated in this study. The mean age was 21.7 years (standard deviation [SD], 1.4). All subjects were university students or postgraduate students. Females were included in this study as was the case with almost all of the intervention studies of this kind. All subjects had normal vision, none had a history of neurological or psychiatric illness, and none reported any recent use of psychoactive drugs or antipsychotic drugs. A history of psychiatric illnesses or recent drug use was assessed with our laboratory's routine questionnaire in which each subject answered questions about whether they had or have any of a list of illnesses and also listed drugs they had taken recently. Handedness was evaluated using the Edinburgh Handedness Inventory [21]. Written informed consent was obtained from each subject. This study was approved by the Ethics Committee of Tohoku University. Several participants participated in the pre- and post- MRI and psychological experiments at the same time. Participants were randomly assigned to either the intervention group (the IATWMMC or the placebo group) or to the no-intervention group. But participants in the same intervention group period were all assigned to the same training protocol group (the IATWMMC group or the placebo group). This means for example, participants in the period of January 7^th–January 14^th are assigned to the IATWMMC group when they are assigned to the intervention group, but participants in the period of January 21^st–January 28^th are assigned to the placebo group when they are assigned to the intervention group. In addition, participants chose the periods in which they wanted to participate by themselves. None of the participants were notified that there were three groups (rather than just the intervention group and the no-intervention group) until after the post MRI and psychological experiments. In other words, participants in the placebo controlled group did not know they were practicing placebo training tasks until the end of the experiment. The IATWMMC group consisted of 18 participants (13 men and five women) and the mean age of the IATWMMC group was 21.9 years (standard deviation [SD], 1.5). The placebo group consisted of 18 participants (twelve men and six women) and the mean age of the placebo group was 21.6 years (standard deviation [SD], 1.6). The no-intervention group consisted of 19 participants (17 men and two women) and the mean age of the no-intervention group was 21.7 years (standard deviation [SD], 1.3). The IATWMMC and placebo intervention and no-intervention groups did not differ significantly (P>0.2, ANOVA) in basic background characteristic such as age, sex, and the score of Raven's Advanced Progressive Matrix [22], which measures cognitive ability that is central to general intelligence [23]. One participant in the IATWMMC group and one participant in the placebo intervention group terminated their training prior to completion. Furthermore, another participant in the IATWMMC group repeated intentional mistakes during the training. These three subjects were excluded from the further analysis of the effects of the training.

The experimental and placebo training programs were computerized, in-house developed Borland C++ programs that consisted of mental multiplication tasks and mental addition tasks. Participants in the experimental and placebo training groups undertook five days of training within six days. Training each day lasted about four hours which usually included two 10-minuite breaks. All participants were MRI scanned and took psychological tests immediately before and after this six-day period. This means the participants were MRI scanned and took psychological tests in one day and one week after that. The no-intervention group did not receive any training or perform any specific activity during the period separating the two MRI sessions.

The mental multiplication task in the IATWMMC is an adaptive training of mental multiplication calculations. The program for the mental multiplication task in the IATWMMC group is designed to assist participants' mental calculation abilities by allowing subjects to check whether the intermediate result of their mental multiplication is correct. The “intermediate result” of mental multiplication refers to the answer in each column when multiplication problems are solved in the following way: when the problem is 37×45, the intermediate result of the first column is 37×5 = 185, and that of the second column is 37×4 = 148). Subjects are asked to solve mental multiplication problems in a normal way (as Japanese or English speakers do computations on paper (see [24]) in their minds and not to solve problems in any other way. Subjects must continue the task until they get the correct answer. After they get the final correct answer, they are asked to give the intermediate answers to the problem without looking at the problem in order to rule out the use of any other possible strategies that do not solve the problems in a normal way. Giving up on or experiencing too many failures during the calculation, or running out of time (one hour), are conditions considered to be failures. There is a 1 hour time limit in this task, which means there is no virtually no time limit in this task. This measure was taken so that the time (not the subjects' abilities) did not prevent subjects from reaching the solution. If participants answer correctly, the problems become more difficult (the task starts from two-digit times two-digit multiplication and then becomes two-digit times three-digit multiplication and then three-digit times three-digit multiplication and then three-digit times four-digit mental multiplication). Two failures in a row make the problems less difficult. The computerized task for mental addition is programmed for the intensive adaptive or progressive training of mental addition calculations. Ten two-digit numbers are presented one by one and the subjects are asked to add them. If they get the correct answer, the interstimulus interval (ISI) becomes shorter [ISI becomes (original ISI)×(0.9)×(0.9)×(0.9)]. Five wrong answers in a row makes the problems less difficult [ISI becomes (original ISI)×(10/9)]. The two trainings were interleaved. The subjects in the placebo intervention group perform similar tasks, except the difficulty of the tasks doesn't change from the initial points (two-digit times two-digit multiplication in the mental multiplication task, ten-seconds-ISI in the mental addition task). In this placebo mental multiplication task, making three mistakes in one problem is considered to be a failure and the next problem appears. The IATWMMC group's subjects know the level of the tasks they are doing and the placebo intervention group's subjects receive feedback on their current performance (accuracy) in every training session during the training. Thus, subjects of the placebo training go through the same amount of training as the IATWMMC group's subjects and receive feedback on their task performance. WM training without intensive adaptive training does not cause an increase in general WM capacity [25]. After the experiment, the individuals complete a questionnaire to ascertain each subject's subjective feelings about the effects of the training, the subjective fatigue that subjects felt during the task [which was measured by a visual analogue scale (VAS)], the strategies that they used while performing the training task, and so on.

For pre- and post- training evaluation, a battery of neuropsychological tests and questionnaires was administered. This battery includes the following contents. [A] Raven's Advanced Progressive Matrices [RAPM; 22], a nonverbal reasoning task; [B] the arithmetic task in the WAIS-III [26], a WM task using mental calculation; [C] the digit symbol task in the WAIS- III [26], a processing speed task; [D] the Stroop task (Hakoda's version) [27], which measures response inhibition and impulsivity. This version of the Stroop task is a matching-type requiring subjects to choose and check correct answers, unlike the traditional oral naming task. The test consists of two control tasks, a Stroop task and a reverse-Stroop task. Two independent measures, reverse-Stroop interference rate and Stroop interference rate, are calculated. [E] The S-A creativity test [28], which measures creativity. A detailed discussion of the psychometric properties of this instrument and how it was developed is found in the technical manual of this test [28]. The test is used to evaluate creativity through divergent thinking [28] and it involves three types of tasks. The first task requires subjects to generate unique ways of using typical objects. The second task requires subjects to imagine desirable functions in ordinary objects. The third task requires subjects to imagine the consequences of ‘unimaginable things’ happening. The S-A test scores the four dimensions of the creative process (fluency, originality, elaboration, and flexibility). In this study, the sum of the graded scores of the four dimensions was used in the analysis. For more details including the psychometric properties of this test, sample answers to the questionnaire, and the manner in which they were scored, see our previous works [29], [30]. [F] Arithmetic tasks, which are similar to the ones constructed by Grabner et al. [31]. These tests measure multiplication performance consisting of two forms of one-digit times one-digit multiplication problems (a simple arithmetic task with numbers between 2 and 9) and two forms of two-digit times two-digit multiplication problems (a complex arithmetic task with numbers between 11 and 19) . The two forms of each task are the same, but the numbers used in the problems are different. Each form of the simple arithmetic task is presented with a time limit of 30 s and each form of the complex arithmetic task is presented with time limits of 60 s. [G] Letter mental rotation task, in which a pair of Japanese letters (one, a normal word and the other, either a rotated normal letter or a rotated mirrored image), are presented and participants are asked to judge whether the two presented letters would be the same or not after they are rotated. [H] The letter span task, a verbal WM task. This test is administered like the Digit span task [26], except that instead of digits, Japanese letters are used. This measure was taken to rule out the possibility that the expected improvement in the span task following our training resulted because participants became habituated to remembering numbers. [I] Trail making tests A and B, which measures cognitive flexibility [32]. No other cognitive tests were used in this study. Questionnaires that were designed to assess mainly the traits of subjects were collected from the subjects but not described in this study because they were apparently not designed to assess the effects of a five-day intervention. Except self-report questionnaires, all neuropsychological assessments were performed by postgraduate and undergraduate students who were kept blind to the group membership of participants.

All MRI data acquisition was conducted with a 3-T Philips Intera Achieva scanner. Using a MPRAGE sequence, high-resolution T1-weighted structural images (240×240 matrix, TR = 6.5 ms, TE = 3 ms, FOV = 24 cm, 162 slices, 1.0 mm slice thickness) were collected. In this study, only these T1-weighted structural images were analyzed. The diffusion-weighted data were acquired only in the pre- MRI experiment by using a spin-echo EPI sequence. Arterial spin labeling images were obtained only in the pre- MRI experiment. All the participants are assigned to our on-going study to investigate the association among brain images, cognitive functions, and their age-related changes. The images that were taken in the pre- MRI experiment were used in our previous study [30] and are going to be used in our future study, but not in this study. Furthermore, functional MRI data were obtained while the subjects were performing the N-back task and mental calculation task in the pre- MRI and post- MRI scans, but functional MRI data were not analyzed in this study. The details of parameters in these scans were not described in this study, since these scans were not used in this study. However, for the details of diffusion-weighted data, see our previous work [30].

Data pre-processing of the morphological data was performed with VBM2 software [33], an extension of SPM2. Default parameter settings were used [33]. In order to reduce the scanner-specific bias, we created a customized GM anatomical template from the pre-intervention data of all the participants in this study. To facilitate optimal segmentation, we estimated normalization parameters using an optimized protocol [11]. In addition, we performed a correction for volume changes (modulation) by modulating each voxel with the Jacobian determinants derived from spatial normalization, allowing us to also test for regional differences in the absolute amount of GM [34]. Subsequently, all images were smoothed by convolving them with an isotropic Gaussian Kernel of 10 mm full-width at half maximum. Finally, the signal change in regional gray matter volume (rGMV) between pre- and post- intervention images was computed at each voxel for each participant. In this computation, we included only voxels that showed GMV values>0.10 in both pre- and post- scans to avoid possible partial volume effects around the borders between GM and WM as well as between GM and CSF. The resulting maps representing the rGMV change between the pre- and post- MRI experiments (rGMV post – rGMV pre) were then forwarded to the group level analysis, described in the next section.

Finally, to ensure that there were no potential differences between the groups in the amount of rGMV and that training related changes cannot be explained by pre-training differences in gray matter, we compared pre-training rGMV between the IATWMMC group and the combined control groups. We used the ANCOVA option of SPM5 for this analysis with no covariates (which is equal to ANOVA). In the group analysis, we included only voxels that showed a GM value>0.10 to avoid the possibility of partial volume effects.

The behavioral data were analyzed using the statistic software SPSS 16.0 (SPSS Inc., Chicago, IL). Since our primary interest is only the superiority (or beneficial effects) of the intervention training, in our behavioral analysis test-retest changes in the group of interest were compared to the test-retest changes in the control group using one-tailed tests (P<0.05) as was performed in previous studies [3], [35]. However, two-tailed tests were used in behavioral measures in which the definition of ‘superiority’ was not clear; namely, for the inverse Stroop interference rate which shows age related decline [36] and for an increase in the Stroop interference rate in patients with schizophrenia [37]. The two-tailed tests were also used to compare group differences in the changes in the creativity test scores, which are associated with impaired selective attention systems, psychosis and cognitive disinhibition [38], [39], [40]. Creativity seems as an obvious positive trait, however there are tremendous amount of literatures that show creativity is associated with psychopathologies and impaired selective attention system (for the full discussion of this matter, see [2]). Especially, WMC and creativity show a lot of opposing psychological, pathological, pharmacological and genetic characteristics (for detail, see [2]). For example, the prevalent genotype that is associated with lower WMC [41] is associated with increased creativity [42]. On the other hand, Ritalin (methylphenidate) administration significantly decreased symptoms of attention deficit hyperactive disorder and creativity [43] while improving WMC [44]. In the psychological and morphological analyses, first the placebo group was compared with the no-intervention group using one-way analyses of covariance (ANCOVAs) with the difference between pre- and post- test measures as dependent variables and pretest scores as covariates [in the VBM analyses, total gray matter volume in the pre-measurement, pretest scores of general intelligence (RAPM) measures, and two measures of WM (the arithmetic task of WAIS-III and the letter span task), were used as covariates] to exclude the possibility that any pre-existing difference of measurement between the groups affected the result of each measure. Complex arithmetic ability was not included as a covariate because it is relatively little to do with WM compared with the arithmetic task of WAIS-III (note the latter is a typical WM task in that subjects remember told complex information and do mental calculation based on the information, even though the task is called “arithmetic”, while the former is not the case and the two tasks are very much different in relationship with WM). In this kind of randomized controlled interventional study, ANCOVA can answer the following question [45]. “If the groups were equivalent on the pretest, would there be a significant difference between the groups on the posttest?” Thus, ANCOVA should be used in this kind of randomized controlled interventional study [45]. No significant effects were found for any of the psychological measures (P>0.05) and morphological data analyzed as described below [P>0.05, and corrected at the non-isotropic adjusted cluster level [46] with an underlying voxel level of P<0.005]. Thus, since we could not find evidence of a difference between the change of the placebo group and that of the no-intervention group, these two control groups were combined in all subsequent analyses as was performed in a previous WM training study [47]. After that, in another set of ANCOVAs with the same variables, the IATWMMC group, was compared with the combined control group.

In the group level imaging analysis, we tested for group wise differences in the change in rGMV. We used a factorial design option in SPM5. In these analyses, the effects of the interventions, estimated by comparing changes in pre- to post- measures as described above, were compared between the groups at each voxel with total gray matter volume in the pre-measurement, pretest scores of measures of general intelligence (RAPM), and two measures of WM (the arithmetic task of WAIS-III and the letter span task), as covariates. Two measures of WM were included in covariates to rule out the possibility that pre-existing differences in WM affected the extent of WM-training-induced change in rGMV. In the analysis, images representing the changes of rGMV (computed as described above) were compared between groups. Also in the group level imaging analysis, after it was confirmed that there were no significant differences between the effects of placebo training and the effects of no-intervention on rGMV, the two control groups were combined. Then the differences between the effects of IATWMMC and those of the combined control groups were investigated.

In this study, the level of statistical significance was set at P<0.05, and corrected for multiple comparisons at the whole brain level using the non-isotropic adjusted cluster level [46] with an underlying voxel level of P<0.005. Non-isotropic adjusted cluster-size tests can be applied to data known to be non-isotropic (in another words, not uniformly smooth), such as VBM data [46]. Simulation-based validation of this test has been performed [46], and now it is widely used (in the case of the interventional study, see [48]). In this non-isotropic cluster-size test of random field theory, a relatively higher cluster determining thresholds combined with higher smoothing values of more than 6 voxels are recommended [49]. In this non-isotropic cluster-size test, statistical thresholds were determined based on random field theory [46].

Additionally, we performed simple regression analyses in the IATWMMC group, using the difference between pre- and post- test measures of the letter span task (which is a verbal WM task that showed IATWMMC-related improvement after training in the behavioral analysis described below) and mean rGMV changes in the clusters identified as significant in the analysis of the group comparison, to test for possible correlations between rGMV change and performance change.

Furthermore, to show firmly that preexisting group differences in rGMV did not affect the finding in the group level imaging analysis, we extracted mean value of rGMV changes in pre- to post- measures in the significant clusters in the group level whole brain imaging analysis (ANCOVA) described above as well as that mean rGMV values of pre- measure in these significant clusters. Then, we performed the ANOVA to compare group differences of mean changes in pre- to post- measures in the significant clusters in the group level whole brain imaging analysis (the analysis which did not take preexisting rGMV differences between groups into account) and also we performed ANCOVA to compare group differences of mean changes in pre- to post- measures in the significant clusters with mean rGMV values of pre- measure in these significant clusters as a covariate (the analysis which took preexisting rGMV differences between groups into account). Then we compared significance of two results and saw if the results substantially changed when the mean rGMV values of pre- measures in the significant clusters were taken into account.

All groups were comparable on the relevant background characteristics of age, sex and general intelligence (no significant differences for P>0.20, ANOVA). Practice of IATWMMC resulted in a significant increase in trained mental addition task performance (for the shortest ISI of the task solved correctly by the subjects) from the first day of training to the last day of training (paired-t, P<0.001; Fig. 1A). Practice-related performance increased in the IATWMMC group. Practice of IATWMMC resulted in a significant increase in trained mental multiplication task performance (the highest level of the task subjects solved correctly) from the first day of training to the last day of training (paired-t, P<0.001; Fig. 1B).

With regard to the effects of the training on other measures, first the placebo group was compared with the no-intervention group using one-way analyses of covariance (ANCOVAs). The difference between pre- and post- test measures were used as dependent variables and the pretest scores were used as covariates to exclude the possibility that any pre-existing difference in measurement between the groups affected the result of each measure. In all psychological measures, planned statistical tests (one-tail or two-tail) were conducted based on our hypotheses (see Methods for details), as were performed in the previous studies [3], [35]. No significant effects were found for any of the psychological measures (P>0.05) or morphological data (P>0.05 corrected at non-isotropic adjusted cluster level). Thus, since we could not find any evidence of a difference between the placebo group and the no-intervention group, these two control groups were combined in all subsequent analyses. This was also performed in the previous WM training study [47]. In another set of ANCOVAs with the same variables, the IATWMMC group was compared with the combined control group.

Behavioral results comparing the combined control group, and the IATWMMC group showed a significantly larger pre- to post- test increase for performance of a complex arithmetic task (P = 0.049), for performance of the letter span task (P = 0.002), and for reverse Stroop interference (P = 0.008) in the IATWMMC group. The IATWMMC group showed a significantly larger pre- to post- test decrease in creativity test performance (P = 0.007) (for all the results of the psychological measures, see Table 1). Also the IATWMMC group showed a statistical trend of increase in the mental rotation task (P = 0.064). These significant behavioral results remained significant or showed statistical trends when the analyses were performed without data from the no-intervention group, though unsurprisingly the P value increased in some tests (for all the results, see Table 1).

VBM analysis revealed that, compared with a test-retest decrease in the combined control group, the IATWMMC showed a statistically significantly larger decrease in the rGMV of the bilateral dorsolateral prefrontal cortex (DLPFC), the regions in the bilateral parietal cortices, and the left superior temporal gyrus [(rGM pre - rGM post) IATWMMC - (rGM pre -rGM post) combined control; Fig. 2,3 and Table 2]. There were no significant IATWMMC related increases in rGMV (P>0.05, corrected for multiple comparisons at the non-isotropic adjusted cluster level, with an underlying voxel level of P<0.005 uncorrected) when compared with the test-retest increase in the combined control groups.

Using a questionnaire gathered after the training, 10 out of 16 subjects from the IATWMMC group and an almost similar proportion (12 out of 17 subjects) from the placebo training group reported their subjective feelings about the effects of the training (most of the reports were related to task, memory, or mental calculation). The questionnaires also asked each subject about their fatigue during the training using VAS. The mean level of subject fatigue in the IATWMMC group was 7.49 (out of 10) points, while the mean level of subject fatigue in the placebo intervention group was 6.74 (out of 10) points. There were no significant differences in subject fatigue between the two training groups (P>0.1). These data, together with our experimental design and the present results, indicate that any subjective feelings of the training effects (or any other factor the placebo intervention group may have had, such as fatigue, commitment to the training, feedback from the training performance, and contact with experimenters) did not lead to an improvement in the performance of the untrained tasks in this kind of study. In our study, monetary reward was given to the subjects of each group in the same way that subjects of noninterventional fMRI studies are recruited; based on how long they participate in the experiment.

The analysis showed no significant regional differences in rGMV between the IATWMMC group and the combined control group (P>0.1, corrected at the nonisotropic adjusted cluster level).

Simple regression analyses using the difference between pre- and post- test measures of the letter span task in the IATWMMC group and mean rGMV changes in the clusters that showed the significant effects of IATWMMC revealed there was a significant negative correlation only in a cluster in the left superior temporal gyrus (P = 0.011, t = −2.94; Fig. 3). The results indicate that the more rGMV decreased in subjects of the IATWMMC group following the training, the more subjects improved on the letter span task.

The lack of correlation in the fronto-parietal clusters (which play a key role in WM) may be because of the nonlinearity of the training induced gray matter change [17], [20]. Also, note number of subjects in the intervention group (N = 16) is apparently not suitable for investigating the number of possible nonlinear relationships. Consistent with this notion, VBM studies have rather consistently failed to identify the linear relationship between gray matter change and intervention-related variables [13], [17], [20], [50].

The P values of ANOVA (to-tailed) that compare the group differences (between the IATWMMC group and the combined control group) of mean values of pre- to post- changes of rGMV in significant clusters of (a) the right DLPFC, (b) the left DLPFC (the more anterior one), (c) the left parietal cortex (d) the left DLPFC (the more posterior one) (e) the left superior temporal gyrus and (f) the right parietal cortex were 0.0002, 0.0014, 0.0014, 0.0003, 0.00008 and 0.003 respectively. Note this ANOVA did not take preexisting rGMV differences values into account. On the other hand, the P values of ANCOVA (two-tailed) that compare the group differences (between the IATWMMC group and the combined control group) of mean values of pre- to post- changes of rGMV with mean values of pre- measure of rGMV in these clusters as a covariate in significant clusters (a)–(f), were 0.0002, 0.0006, 0.0016, 0.0004, 0.00011 and 0.004, respectively (note this analysis is taking preexisting rGMV differences in significant clusters into account). These comparisons show preexisting differences of rGMV in these significant clusters merely affected the group differences of pre- to post- changes of rGMV in these clusters.

When the data from the no-intervention group was removed from the analyses, the same statistical patterns were observed. The P values of ANCOVA (two-tailed) that compared the group differences (between the IATWMMC group and the placebo group) of mean values of pre- to post- changes of rGMV with mean values of pre- measure of rGMV in these clusters as a covariate in significant clusters (a)–(f) were 0.007, 0.012, 0.020, 0.004, 0.001, and 0.029, respectively (note that this analysis takes preexisting rGMV differences in significant clusters into account).