Abstract

Some cognitive training studies have reported working memory benefits that generalize beyond the trained tasks, whereas others have only found task-specific training effects. What brain networks are associated with general training effects, rather than task-specific effects? We investigated this question in the context of working memory training using the COGITO data set, a longitudinal project including behavioral assessments before and after 100 days of cognitive training in 101 younger (20–31 years) and 103 older (65–80 years) adults. Pre- and postassessments included verbal, numerical, and spatial measures of working memory. It was therefore possible to assess training effects on working memory at a general latent ability level. Previous analyses of these data found training-related improvements on this latent working memory factor in both young and old participants. fMRI data were collected from a subsample of participants (24 young and 15 old) during pre- and post-training sessions. We used independent component analysis to identify networks involved in a perceptual decision-making task performed in the scanner. We identified five task-positive components that were task-related: two frontal networks, a ventral visual network, a motor network, and a cerebellar network. Pre-training activity of the motor network predicted latent working memory performance before training. Additionally, activity in the motor network predicted training-related changes in working memory ability. These findings suggest activity in the motor network plays a role in task-independent working memory improvements and have implications for our understanding of working memory training and for the design and implementation of future training interventions.

INTRODUCTION

Improving cognitive abilities such as working memory or perceptual speed through practice has been pursued in a large number of cognitive training interventions (for reviews, see Noack, Lövdén, & Schmiedek, 2014; Lustig, Shah, Seidler, & Reuter-Lorenz, 2009; Verhaeghen, Marcoen, & Goossens, 1992; Baltes & Lindenberger, 1988). These research programs involve training some cognitive skill through practice, followed by a measurement of improvement at the end of this training period. Not surprisingly, intense cognitive training almost always leads to improved performance on the tasks being trained, but evidence for more general effects on the underlying cognitive process itself has been mixed.

One type of evidence has come from studies that have reported transfer effects, that is, improvement on tasks that were different from the tasks that were trained. Jaeggi and colleagues (Jaeggi et al., 2010; Jaeggi, Buschkuehl, Jonides, & Perrig, 2008) reported improvements in fluid processing following working memory training (however, see Chooi & Thompson, 2012, for a failure to replicate these findings). Karbach and Kray (2009) examined if task-switching training could lead to the transfer of executive training gains and found substantial performance gains in untrained executive tasks and also in fluid intelligence. A wide range of studies including the ACTIVE study (Willis et al., 2006), the NeuroRacer study (Anguera et al., 2013), and the Synapse study (Park et al., 2014) all reported training benefits that transferred to tasks that were significantly different from the trained tasks (often called far transfer). Other studies have administered multiple-task training approaches through video games and demonstrated transfer to some, but not all, untrained tasks, including divided attention (Baniqued et al., 2014) and working memory and perceptual speed (Baniqued et al., 2015).

On the other hand, many other studies show limited improvements beyond the practiced tasks (Redick et al., 2013; van Muijden, Band, & Hommel, 2012; Owen et al., 2010), and a recent meta-analysis of working memory training studies concluded that there was no evidence of far transfer across studies (Melby-Lervåg, Redick, & Hulme, 2016). Furthermore, none of the studies that did report transfer were able to shed light on the underlying mechanisms that gave rise to the observed general effects.

An alternative way to assess general, task-independent training effects is to look for effects at the latent ability level. For example, the authors of the COGITO study (Schmiedek, Lövdén, & Lindenberger, 2010) collected multiple measures of the same underlying construct (e.g., working memory) and then examined training effects on a latent variable derived from these multiple measures. In that study, 101 younger and 103 older adults practiced six perceptual speed tasks: three working memory tasks and three episodic memory tasks for approximately 100 daily sessions. Participants also undertook lengthy pre- and post-training assessments. Multiple tasks were tested in each content domain, which made it possible to assess effects at a latent ability level, rather than just task-specific improvements on individual trained tasks. And the study did find training-related improvements at the level of latent cognitive abilities, specifically in a latent variable associated with general working memory ability. A 2-year follow-up also found reliable long-term training effects (Schmiedek, Lövdén, & Lindenberger, 2014).

A subset of the COGITO study participants also participated in brain imaging sessions before and after training, making it possible to study the neural substrates of latent ability changes. That was the goal of this study. Specifically, we used independent component analysis (ICA) to identify neural networks engaged while COGITO participants performed a perceptual decision-making task (Kühn et al., 2011). We first examined whether the baseline (pre-training) activity in any of these networks related to latent working memory ability before training. We then examined whether activity in any of these networks was related to training-related changes in working memory ability.

We focused on baseline activity for two reasons. First, training interventions are costly to administer, so ideally one would like to be able to predict who will benefit most from an intervention before the fact (i.e., based on baseline variables), allowing interventions to be personalized for different people. This approach has been adopted in a number of previous training studies that have linked baseline properties of the brain to training-related gains (e.g., Iordan et al., 2018; Gallen et al., 2016; Arnemann et al., 2015). The second reason is simply that there often is not much variability in training-related brain changes across people, making it difficult to relate them to other variables. For example, in the COGITO data set, there is 71% more variance in the pre-training neural activity data than in the neural change measures.

METHODS

Participants

The full sample of the COGITO study included 101 younger adults (51.5% women; mean age = 25.12 years, SD = 2.72, range = 20–31 years) and 103 older adults (49.5% women; mean age = 70.79 years, SD = 4.15, range = 64–80 years). Full demographic details of this sample can be found in Schmiedek et al. (2010). Participants were recruited through newspaper advertisements, word of mouth, and flyers circulated in Berlin, Germany.

A subset of the parent sample participated in imaging sessions before and after training, with data obtained from 24 younger adults (54.2% women; mean age = 25.2 years, SD = 3.2, range = 20.5–31.1 years) and 15 older adults (60.0% women; mean age = 69.47 years, SD = 4.0, range = 64.0–80 years). The imaging sample was comparable to the full sample in age and cognitive status (see Table 1). All participants in the imaging sample were right-handed, had normal or corrected-to-normal vision, and reported no history of diabetes, neurological or psychiatric conditions, drug/alcohol abuse, or cardiovascular disease (with the exception of treated hypertension). No use of drugs acting on the central nervous system (e.g., antiepileptic or antidepressants) was reported. Older participants were screened for dementia using the Mini-Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975) with a cutoff score of 26 for the imaging subsample. This cutoff has been demonstrated to be optimal for detecting dementia (O'Bryant et al., 2008). The ethics committee of the Otto von Guericke University of Magdeburg approved the imaging study, and the ethics committee of the Max Planck Institute for Human Development, Berlin, approved the behavioral parent study. All participants provided written informed consent.

Table 1. 
Demographics of Younger and Older Groups for Full Sample and Imaging Subsample
 Younger AdultsOlder Adults
COGITO Full Sample (n = 101)fMRI Subsample (n = 24)Differences between Younger SubgroupsaCOGITO Full Sample (n = 103)fMRI Subsample (n = 15)Differences between Older Subgroupsa
Male/female 49/52 11/13 0.09 (0.76) 50/53 6/9 0.51 (0.47) 
Age, y 25.12 (2.72) 24.38 (3.12) 1.54 (0.13) 70.79 (4.15) 69.47 (4.10) 1.34 (0.18) 
Digit symbol 60.32 (9.54) 60.67 (8.26) 0.20 (0.84) 43.64 (9.01) 44.00 (6.93) 0.17 (0.87 
Raven Matrices 0.53 (0.21) .60 (0.23) 1.78 (0.08) 0.24 (0.15) 0.26 (0.14) 0.48 (0.63) 
MMSE N/A N/A N/A 28.36 (1.22) 28.33 (1.35) 1.34 (0.18) 
 Younger AdultsOlder Adults
COGITO Full Sample (n = 101)fMRI Subsample (n = 24)Differences between Younger SubgroupsaCOGITO Full Sample (n = 103)fMRI Subsample (n = 15)Differences between Older Subgroupsa
Male/female 49/52 11/13 0.09 (0.76) 50/53 6/9 0.51 (0.47) 
Age, y 25.12 (2.72) 24.38 (3.12) 1.54 (0.13) 70.79 (4.15) 69.47 (4.10) 1.34 (0.18) 
Digit symbol 60.32 (9.54) 60.67 (8.26) 0.20 (0.84) 43.64 (9.01) 44.00 (6.93) 0.17 (0.87 
Raven Matrices 0.53 (0.21) .60 (0.23) 1.78 (0.08) 0.24 (0.15) 0.26 (0.14) 0.48 (0.63) 
MMSE N/A N/A N/A 28.36 (1.22) 28.33 (1.35) 1.34 (0.18) 
a

To determine whether there were differences between the members of the full COGITO sample that were included in the imaging group and those not included in the imaging group, Pearson's chi-square test was calculated for categorical variables (i.e., sex) and two-sample t tests were performed for continuable variables.

Study Design

Full details and parameters of the COGITO study are provided in Schmiedek et al. (2010). The COGITO study consisted of an initial assessment using a large battery of cognitive tests and self-report questionnaires, followed by a longitudinal intervention consisting of approximately 101 one-hour training sessions (younger adults: mean sessions = 100.8, SD = 2.6, range = 87–109; older adults: mean sessions = 101.0, SD = 2.7, range = 90–106) during which 12 tasks were practiced, followed by a postintervention assessment that was identical in content to the initial, pre-training assessment. MRI scans were acquired pre- and post-training, shortly after behavioral testing had been completed. Participants were paid between €1450 and €1950, depending on the number of training sessions they completed.

Materials and Procedure

Pretest and Posttest

Participants completed 10 testing sessions before the intervention phase and another 10 testing sessions afterward. Each testing session consisted of approximately 2–2.5 hr of comprehensive cognitive test batteries and self-report questionnaires. The test battery included the 12 daily training tasks (detailed in the next section), plus three near-transfer working memory tasks (animal span, 3-back numerical and spatial memory updating), three far-transfer working memory tasks (reading span, counting span, and rotation span), a word pair episodic memory task, a set of Raven Matrices, and 27 tests from the Berlin Intelligence Structure Test, a paper and pencil battery designed to measure fluid intelligence (Gf), episodic memory, and perceptual speed. As we did not analyze data from tasks only administered during testing sessions, we do not describe them further.

Daily Training Tasks

In each daily training session, participants practiced 12 different computerized tasks, which are presented in Figure 1. These tasks tapped three different ability domains: perceptual speed, episodic memory, and working memory. Furthermore, within each domain, there was a numerical task, a verbal task, and a figural–spatial task. For perceptual speed, these were three 2-choice reaction tasks (CRTs; odd vs. even numbers for numerical; consonants vs. vowels for verbal; symmetric vs. asymmetric figures for figural–spatial) and three comparison tasks (two strings of digits [numerical], two strings of consonants [verbal], two three-dimensional figures [figural–spatial]). For episodic memory, participants had to memorize number–word pairs (numerical), word lists (verbal), or object positions in a grid (figural–spatial). Working memory tasks were adapted versions of the numerical memory updating (numerical; Salthouse, Babcock, & Shaw, 1991), alpha span (verbal; Craik, 1986), and spatial n-back tasks (figural–spatial; Cohen et al., 1997). To maximize and even out the cognitive challenge of these tasks across individuals, while also maintaining motivation, difficulty levels for the episodic memory and working memory tasks were individualized using presentation times based on pretest performance. Following this tailoring, presentation times were kept constant for each individual across the intervention period. Training sessions were carried out in small groups of two to five participants, each at a PC, using the keyboard, mouse, and specialized button boxes.

Figure 1. 

Twelve daily training tasks that were practiced by participants during the training intervention.

Figure 1. 

Twelve daily training tasks that were practiced by participants during the training intervention.

Perceptual speed: CRTs.

The stimulus layout was the same for all three CRT tasks. Imagine the seven horizontal and vertical lines making up the number “8” as displayed on a desktop calculator. Stimuli consisted of subsets of those seven lines making numbers for the numerical CRT, letters for the verbal CRT, and line figures for the figural–spatial CRT (Figure 1). Stimuli were presented and then masked with the “calculator 8” figure, with additional lines extending in all 10 possible directions. Participants had to determine whether the stimulus was odd or even (numerical task), a consonant or a vowel (verbal task), or symmetric or asymmetric (figural–spatial task) as quickly as possible. Based on pre-training assessment, a fast and a slow masking time (from a possible 24, 47, or 94 msec) were chosen for each participant. Each task run contained 2 slow trials and 20 fast trials, and two runs of each of the three CRTs were included in each practice session.

Perceptual speed: Comparison tasks.

For each comparison task, participants had to make the same or a different judgment as quickly as possible. For the numerical version, two strings of five numbers each appeared on the left and right of the screen. These two strings were either the same or differed by just one number. The verbal task was the same as the numerical task but contained strings of five consonants. For the figural version of the task, two “fribbles”—three-dimensional colored objects consisted of several parts—were presented on the left and right of the screen, and participants decided as quickly as possible if the objects were the same or different (Figure 1). (Fribble images were used courtesy of Michael J. Tarr, Brown University, tarrlab.org/.) In each practice session, two runs, each consisting of 40 items, were included for each of the three task variants.

Episodic memory: Number–noun pairs.

Lists of 12 pairs of two-digit numbers and plural nouns (e.g., 22 dogs) were presented sequentially with presentation time ranging from 1000 to 4000 msec based on pre-training performance. Following presentation of the list, the nouns were presented again, in a random order, and participants were asked to enter the number that the noun was paired with during its initial presentation (Figure 1). Two runs of 12 number–noun pairs were included in each daily session.

Episodic memory: Word lists.

Lists of 36 nouns were presented sequentially, with presentation time ranging from 1000 to 4000 msec, based on pre-training performance. Word lists were balanced for length, emotional valence, and imageability. Following presentation, participants were instructed to enter the first three letters of each word in the correct order using the keyboard (Figure 1). Performance scores were based on the number of words recalled, multiplied by the accuracy of the order (ranging from 0 for reverse order to 1 for perfect order). Two word list runs were included in each daily training session.

Episodic memory: Object position memory.

Sequences of 12 photographs of real-life objects were displayed in different positions on a 6 × 6 grid. Presentation time varied from 1000 to 4000 msec based on each individual's pre-training performance. Following presentation, the objects appeared at the bottom of the screen, and participants were asked to move the objects to the correct locations using the computer mouse (Figure 1). Each daily training session included two sequences of 12 photographs.

Working memory: Memory updating numerical.

Four digits were presented simultaneously for 4000 msec. After an ISI of 500 msec, a sequence of eight updating operations was presented below individual digits in the set. These updating operations were additions and subtractions ranging from −8 to +8. Participants were instructed to apply each operation to the digit immediately above and memorize the updated result. The presentation time for each operation was based on pre-training performance, with possible presentation times ranging from 500 to 2750 msec. At the end of the run, participants were required to report the four updated digits (Figure 1). Eight runs of this task were included in each daily training session.

Working memory: Alpha span.

Ten upper case consonants were presented sequentially together with a number below each letter. Participants were required to decide as quickly as possible whether the number corresponded to the position of the current letter in the alphabet within the set of letters presented up to that point (Figure 1). Five of the 10 position numbers were correct, and if they were incorrect, they differed from the correct position by ±1. Presentation time for letters was based on pre-training performance and ranged from 750 to 3000 msec. Eight runs of 10 letters were included in each daily session.

Working memory: 3-Back spatial.

A sequence of 39 black dots appeared at varying locations in a 4 × 4 grid. Participants were asked if each dot was in the same position as the dot three steps prior in the sequence (Figure 1). Dots were presented in random locations, with the following constraints: 12 dots were targets (i.e., positioned in the same location as three dots previous in the sequence), dots did not appear in the same location in consecutive steps, three dots each were presented in the same position as dots two, four, five, or six steps earlier. For all participants, dots were presented for 500 msec, and ISI was varied between 500 and 2500 msec, based on pre-training performance. Four runs of the 3-back spatial task were included in each daily training session.

Behavioral Task in MRI Scanner

While in the scanner, participants performed a two-choice RT task, similar to the perceptual speed CRTs performed during the training stage (Figure 2) and as described in Kühn et al. (2011). Participants saw a fixation stimulus for 400 msec, followed by a target stimulus for 600 msec, which was then masked for 50 msec. The task had two stimulus categories—numerical and verbal. Participants were asked to make a decision regarding the stimulus presented. During numerical blocks, participants were asked to decide whether the number presented was odd or even. During verbal blocks, participants were asked if the letter was a consonant or a vowel. Participants responded by pressing a button. The task consisted of 53-sec blocks, which alternated between verbal and numerical stimuli. These task blocks were intermixed with 16-sec fixation blocks. Each task block included 16 trials (eight odd/eight even or eight consonants/eight vowels) presented with jittered ISIs of 2000–8000 msec. In total, for each session, eight verbal and eight numerical blocks were presented over four runs, giving a total of 128 trials per category. There were two scanning sessions, one before the training intervention and one after.

Figure 2. 

Perceptual decision-making choice RT task performed in the fMRI scanner.

Figure 2. 

Perceptual decision-making choice RT task performed in the fMRI scanner.

MRI Procedures

Images were collected using a 3-T Magnetom Trio MRI scanner system (Siemens Medical Systems) using an eight-channel RF head coil. Functional images were collected using a T2*-weighted EPI sequence sensitive to BOLD contrast (repetition time = 2000 msec, echo time = 30 msec, image matrix 64 × 64, field of view = 224 mm, flip angle = 80, slice thickness = 3.5 mm, distance factor = 0%, voxel size = 3.5 × 3.5 × 3.5 mm3, 32 axial slices). One hundred forty-seven image volumes were acquired per run. Anatomical images were acquired using a T1-weighted sagittal 3-D spoiled gradient-echo sequence (repetition time = 24 msec, echo time = 8 msec, acquisition matrix = 256 × 256 × 256, field of view = 250 × 250 mm2, flip angle = 30, slice thickness = 1.5 mm) on a GE Signa system (General Electric).

fMRI Data Analysis

Preprocessing.

Imaging data were preprocessed using SPM12 software (Wellcome Department of Cognitive Neurology, London, United Kingdom). The first three volumes from each run were excluded to avoid T1 saturation effects. Data were slice time corrected, and a mean image for all EPI volumes was created, to which the individual volumes were spatially realigned through rigid body transformations. The high-resolution structural image was coregistered to the mean image of the EPI series. Realigned images were then spatially normalized to Montreal Neurological Institute (MNI) standardized space using the EPI template image and spatially smoothed using an 8-mm FWHM filter. Low-frequency drifts in the time domain were removed by modeling the time series for each voxel by a set of discrete cosine functions to which a cutoff of 224 sec was applied. Final voxel resolution was 3 × 3 × 3 mm3.

Independent Component Analysis.

Data were analyzed using a group ICA approach, implemented using the GIFT software package (mialab.mrn.org/software/gift/index.html) to identify spatially independent and temporally coherent networks. Data were first concatenated across all runs and all participants. Calhoun, Adali, Pearlson, and Pekar (2001) demonstrate that performing ICA analyses on concatenated group data yields largely similar results to performing separate ICAs on individual participants. Concatenating across groups is a standard use in the GIFT ICA toolbox, including investigations of schizophrenia (Meda, Stevens, Folley, Calhoun, & Pearlson, 2009; Calhoun, Kiehl, Liddle, & Pearlson, 2004), major depressive disorder (Abbott et al., 2013), attention-deficit/hyperactivity disorder (Hoekzema et al., 2014), and epilepsy (Kay et al., 2013). This technique has also been employed extensively within the aging literature by a number of groups (Salami, Pudas, & Nyberg, 2014; De Vogelaere, Santens, Achten, Boon, & Vingerhoets, 2012; Onoda, Ishihara, & Yamaguchi, 2012). Next, data reduction was carried out through a PCA stage. Data were then decomposed into 26 mutually independent components using the Infomax algorithm (Bell & Sejnowski, 1995). The number of components was determined using the minimum description length (MDL) criterion adjusted to account for correlated samples. The ICA analysis produces a spatial map detailing the brain regions included in each network (i.e., component) and an associated average time course of the BOLD signal change in that network based on the data from the full group. Time courses and spatial maps are then back-reconstructed for each participant and each run. This back-reconstruction captures the individual differences in the expression of the components and allows the testing of hypotheses regarding group differences in network performance and intraindividual network changes over time.

To identify and display brain regions associated with each network, spatial maps were entered into second-level group analyses (SPM one-sample t test) and overlaid on structural images, using a p < .05 family-wise error, k = 20 voxel threshold. To understand the degree to which each network was task-engaged, the average time course of each component was modeled in a multiple regression as a function of regressors describing the timing of the task paradigm (i.e., onsets and duration of verbal and numerical blocks convolved with the hemodynamic response). Fitting this model provided estimates (beta weights) of the extent to which each regressor was associated with the associated network (specifically, how strongly weighted that regressor was in the best fitting model of that network's time series). The beta weights calculated thus reflect the degree to which the network was engaged by the task in the scanner. We then averaged across the verbal and numerical beta weights to obtain a single value of task relatedness per network.

Statistical Analysis

A confirmatory factor analysis was conducted on the pre- and posttest working memory measures using R to derive latent working memory variables. Latent variables are comparable to factors in factor analysis, that is, they are not measured but instead represent the variance that is shared by the measured observed variables. In this study, working memory is a latent variable, defined by the individual tasks—numerical memory updating, alpha span, and spatial 3-back. Because the individual tasks were administered before and after cognitive training, we were able to derive pre-training and post-training latent working memory factors. To ensure the pre- and post-training working memory variables were equivalent, factor structures were fixed.

To investigate our first research question—the relationship between working memory ability and network activity before training—separate regression analyses were performed for each of the networks of interest. Similar regression analyses were performed to examine the relationship between pre-training network activity and training-related changes in working memory ability—our second research question. To control for the confounding effect of age in our data set—as data were collected from a young group (aged 20–31 years) and an older group (aged 64–80 years)—age was included as a nuisance covariate in all regressions. To address the concern of multiple comparisons, regression analyses were only considered significant if they reached a significance threshold of p < .01 (Bonferroni correction: α = .05, divided by 5, since five statistical tests were performed).

RESULTS

Independent Components

Twenty-six independent components were estimated using group ICA. Of the 26, 10 components that were related to movement or reflected other artifacts were identified (following criteria provided by Griffanti et al., 2017) and eliminated from further analysis. Specifically, we eliminated components that had a large number of small clusters (“confetti”) rather than a small number of large clusters, that did not seem to discriminate between gray matter and white matter/cerebrospinal fluid or that exhibited a lot of activation in ventricles or white matter, that exhibited a ring-like appearance around the edge of the brain, that had a lot of activation in areas of susceptibility induced signal loss (frontal sinuses, ear canals), or that showed banding patterns in the slice direction or streaks along the phase encoding direction. Of the remaining 16 components, five anatomically plausible, task-positive networks were selected for further analysis, all of which had a task relatedness of R2 > .15. This threshold represented a trade-off. We wanted to restrict the analysis to a relatively small number of the most task-relevant components so that we would have the most power to observe an effect within those networks. We also wanted to include all clearly relevant networks so important findings would not be overlooked. Components were therefore sorted from most to least task related and judged whether each network could be interpreted as crucial to the performance on the scanner task. The five components that were selected for further analysis were all easily interpretable as neural networks that seemed necessary to performing the task and are shown in Figure 3. Selected networks included two frontal networks, a ventral visual network, a motor network, and a cerebellar network. Table 2 contains MNI coordinates of local maxima for these five components. Starting with the sixth most task-related component, it became less obvious how each component should be interpreted and its functional role in the task, and therefore, these components were not included in our analyses.

Figure 3. 

Task positive components in the perceptual decision-making task.

Figure 3. 

Task positive components in the perceptual decision-making task.

Table 2. 
MNI Coordinates for Selected Components
Region (Brodmann's area)MNINo. of VoxelsT
xyz
Left frontal component 
Left precentral gyrus (6) −36 29 955 12.64 
Right precuneus (7) 30 −64 44 93 9.05 
Right middle frontal gyrus (46) 48 32 20 39 7.83 
  
Right frontal component 
Right precentral gyrus (6) 60 26 591 11.14 
  
Ventral visual component 
Inferior occipital gyrus (19) 42 −82 −4 3359 14.21 
Claustrum 36 17 −7 32 9.69 
Inferior parietal lobule (40) −33 −28 38 235 8.40 
Precuneus (7) 24 −46 47 44 8.25 
Cingulate gyrus (32) 26 29 37 7.92 
Inferior frontal gyrus (9) 48 26 69 7.57 
Precentral gyrus (6) −54 26 26 6.93 
Paracentral lobule (31) −9 −7 50 37 6.87 
  
Motor component 
Left inferior parietal lobule (40) −39 −31 47 587 12.34 
Right medial frontal gyrus (6) −4 50 275 10.00 
Right precentral gyrus (6) 27 −7 53 350 9.33 
Left precuneus (7) −49 54 38 6.97 
  
Cerebellum component 
Right cerebellum −73 −19 560 9.90 
Region (Brodmann's area)MNINo. of VoxelsT
xyz
Left frontal component 
Left precentral gyrus (6) −36 29 955 12.64 
Right precuneus (7) 30 −64 44 93 9.05 
Right middle frontal gyrus (46) 48 32 20 39 7.83 
  
Right frontal component 
Right precentral gyrus (6) 60 26 591 11.14 
  
Ventral visual component 
Inferior occipital gyrus (19) 42 −82 −4 3359 14.21 
Claustrum 36 17 −7 32 9.69 
Inferior parietal lobule (40) −33 −28 38 235 8.40 
Precuneus (7) 24 −46 47 44 8.25 
Cingulate gyrus (32) 26 29 37 7.92 
Inferior frontal gyrus (9) 48 26 69 7.57 
Precentral gyrus (6) −54 26 26 6.93 
Paracentral lobule (31) −9 −7 50 37 6.87 
  
Motor component 
Left inferior parietal lobule (40) −39 −31 47 587 12.34 
Right medial frontal gyrus (6) −4 50 275 10.00 
Right precentral gyrus (6) 27 −7 53 350 9.33 
Left precuneus (7) −49 54 38 6.97 
  
Cerebellum component 
Right cerebellum −73 −19 560 9.90 

Table reports local maxima.

Latent Working Memory Factors

Data from the full behavioral sample was used to determine two first-order latent factors of working memory—one pre-training and one post-training. Each of these latent factors was based on data from the same tasks that were practiced daily—the alpha span, n-back, and memory updating tasks. Factor structures were fixed for the pre-training and post-training latent working memory factors. Standardized factor loadings for numerical memory updating, alphabet span, and spatial n-back tasks ranged from .72 and .85 (all ps < .001) on these pre- and post-training latent working memory factors. We used a paired sample t test to confirm that the working memory latent variable improvements following cognitive training that were reported in Schmiedek et al. (2010) extended to the subsample of participants who took part in the imaging session (t38 = 2.37, p < .05). Pre-training working memory ability and training-related working memory change were extremely highly correlated (r = .98, p < .001).

Pre-training Relationships between Working Memory and Component Activation

To examine if pre-training activation of any of the five identified networks predicted pre-training latent working memory ability, regression models were fit for each of the five networks. Each regression model included a regressor based on network activity (average beta weight for the ICA component when fit by the task regressors) and also included age as a nuisance covariate. The results are presented in Table 3 and Figure 4.

Table 3. 
Relationships between Component Activation and Latent Working Memory Ability
Network/ComponentR2Pre-training Working Memory AbilityTraining-related Change in Working Memory Ability
tptp
Left frontal 0.17 1.24 .22 1.09 .28 
Right frontal 0.31 1.08 .29 1.02 .31 
Ventral visual 0.22 0.25 .81 −0.06 .95 
Motor 0.16 2.83 <.01a 3.00 <.01a 
Cerebellum 0.27 1.41 .17 1.60 .12 
Network/ComponentR2Pre-training Working Memory AbilityTraining-related Change in Working Memory Ability
tptp
Left frontal 0.17 1.24 .22 1.09 .28 
Right frontal 0.31 1.08 .29 1.02 .31 
Ventral visual 0.22 0.25 .81 −0.06 .95 
Motor 0.16 2.83 <.01a 3.00 <.01a 
Cerebellum 0.27 1.41 .17 1.60 .12 
a

Survives correction for multiple comparisons.

Figure 4. 

Scatter plots showing age-corrected relationships between ICA betas and (A) pre-training working memory ability and (B) change in working memory performance over training (“+” = young participants; “○” = old participants).

Figure 4. 

Scatter plots showing age-corrected relationships between ICA betas and (A) pre-training working memory ability and (B) change in working memory performance over training (“+” = young participants; “○” = old participants).

Activity in the motor network, but in none of the other four task-related networks, predicted pre-training latent working memory ability (t36 = 2.88, p < .01), surviving correction for multiple comparisons. Post hoc analyses suggested that the relationship was also significant in the young participants alone (t22 = 2.11, p < .05) but was only marginally significant in the old participants alone (t13 = 1.77, p = .10). Comparison of the beta values for each group suggests that the relationships did not significantly differ (t35 = 0.38, p = .70).

Relationships between Training-related Working Memory Change and Neural Activation

Separate regression analyses were also performed for each network to determine if training-related changes in latent working memory were predicted by pre-training activation in any of the networks. Each regression model included network activity and age covariates. The results are presented in Table 3 and Figure 4.

Activity in the motor network, but in none of the other four task-related networks, predicted training-related change in working memory ability (t36 = 3.00, p < .01). Once again, post hoc analyses suggested that this relationship was also significant in the young group alone (t22 = 2.53, p < .05) but did not reach significance in the old group (t13 = 1.31, p = .21), although the size of the effect did not differ significantly between the young and old (t35 = 0.10, p = .92).

To rule out the idea that improvements in processing speed account for training-related changes in working memory, the regression analysis (across the full young and older sample) was repeated, with the inclusion of the mean RT change following training on a perceptual speed task. With the inclusion of this variable, there was still a significant relationship between working memory change and network activation (t35 = 2.88, p < .01).

DISCUSSION

Using ICA and confirmatory factor analysis, we investigated the relationship between the activity of a set of task-positive neural networks and general working memory ability. We investigated this relationship in the COGITO data set—a data set in which a previous analysis of behavioral performance indicated that participants demonstrate not only training-related improvements on individual working memory tasks but also improvements on a latent factor representing more general working memory ability (Schmiedek et al., 2010). These improvements were observed in both old and young participants. We investigated the neural substrates of this latent ability and of the training-related change in the latent ability.

First, using ICA, we identified five task-positive functional networks. We then examined the degree to which activity in these networks was related to working memory ability. We found that activation of a motor network significantly predicted working memory ability before training. Activity in this motor network also significantly predicted change in working memory ability as a result of cognitive training, consistent with a number of studies that have linked characteristics of the brain at baseline (i.e., before training interventions begin) with cognitive gains resulting from training (Iordan et al., 2018; Gallen et al., 2016; Arnemann et al., 2015).

When we examined these relationships separately in older and younger groups, only our findings in the younger group reached significance. However, when we compared regression values between the young and older groups, the relationship was similar. We therefore hypothesize that our nonsignificant findings in the older group were due to a small sample size.

We did not predict that activity in a motor network would be the most consistent predictor of pre-training working memory performance and training-related working memory improvements. We initially expected that working memory ability would be most related to brain networks associated with executive functioning—for example, the left and right frontal networks identified using ICA. Nevertheless, the association with motor network activity is consistent with a number of previous studies. For example, Awh et al. (1996) and Braver et al. (1997) reported load-dependent neural activity in motor, premotor, and supplementary motor areas during working memory tasks. They hypothesized that this association may reflect changes in motor readiness, with systems adapting to deal with higher working memory loads by maintaining tonic increases of activation within motor areas. Meta-analyses of working memory tasks, such as the n-back paradigm, have also revealed robust working memory activation in lateral and medial premotor areas, including the supplementary motor area (Owen, McMillan, Laird, & Bullmore, 2005), particularly during nonverbal tasks (Rottschy et al., 2012). Furthermore, lesions involving the supplementary motor area have been associated with working memory deficits, with patients able to temporarily sustain information but not mentally manipulate it (Cañas, Juncadella, Lau, Gabarrós, & Hernández, 2018).

One potential explanation for the association with the motor network is that training effects that influence working memory at the latent ability level (rather than at the individual task level) primarily reflect improvements in the efficiency with which information in working memory can be manipulated and rehearsed, which might be expected to depend on the motor network. Specifically, individuals who can manipulate information in working memory more efficiently might be able to rehearse more items, compared with individuals who manipulate that information more slowly.

This interpretation is consistent with many working memory models that propose that parts of the motor system (especially supplementary motor areas, premotor areas, and FEFs) play a role in (internal) information manipulation that is analogous to the (external) manipulation of objects in the environment by explicit movement (Mackey & Curtis, 2017; Ikkai & Curtis, 2011; Curtis, 2006; Chein & Fiez, 2001; Frank, Loughry, & O'Reilly, 2001; Smith & Jonides, 1998). This view has now received empirical support from studies of neurological patients with damage to the motor system (e.g., Cañas et al., 2018; Müller & Knight, 2006; Gabrieli, Singh, Stebbins, & Goetz, 1996), from work in nonhuman primates (e.g., Carpenter, Baud-Bovy, Georgopoulos, & Pellizzer, 2018), from neuroimaging studies (Mackey & Curtis, 2017; Ikkai & Curtis, 2011; Curtis, 2006; Chein & Fiez, 2001), and even from infant studies that have related working memory performance to motor control (Gottwald, Achermann, Marciszko, Lindskog, & Gredebäck, 2016).

The limitations regarding this study should be noted. First, although behavioral data were obtained from over 200 participants, only a subset of those individuals completed pre- and post-training fMRI sessions, which meant that our imaging sample was composed of only 39 participants. Although our factor analysis used the behavioral data from all 204 participants, a larger imaging sample would have allowed a more robust analysis of the relationship between behavioral measures and neural measures. A larger imaging sample would also have allowed an analysis of age-related differences, rather than simply controlling for age, as we did in the reported analysis. Additionally, although the COGITO data set included a no-training control group, members of the control group did not complete imaging sessions, and so we were unable to incorporate control group data into our analysis.

Another limitation is that the measures of pre-training working memory and training-related working memory change are highly correlated, in fact, so highly correlated as to be almost indistinguishable. Accordingly, we cannot distinguish whether activity in the motor network is associated with pre-training working memory ability, with working memory change, or with both (although the answer cannot be neither).

A further limitation was that the task performed in the scanner—a perceptual decision-making task—was quite simple and omitted many important cognitive processes. It would be informative to analyze imaging data from a more complex cognitive task to determine if activation of the motor network during a more complex task is similarly predictive of working memory ability.

To summarize, this study uncovered a significant association between working memory ability and motor network activity and between training-related changes in working memory ability and motor network activity. These associations suggest that training-related changes in the speed with which information in working memory can be manipulated may play an important role in training-related changes in working memory ability.

Reprint requests should be sent to Molly Simmonite, Department of Psychology, University of Michigan, 530 Church Street, Ann Arbor, MI 48018, or via e-mail: mollysimmonite@gmail.com.

REFERENCES

Abbott
,
C. C.
,
Lemke
,
N. T.
,
Gopal
,
S.
,
Thoma
,
R. J.
,
Bustillo
,
J.
,
Calhoun
,
V. D.
, et al
(
2013
).
Electroconvulsive therapy response in major depressive disorder: A pilot functional network connectivity resting state fMRI investigation
.
Frontiers in Psychiatry
,
4
,
10
.
Anguera
,
J. A.
,
Boccanfuso
,
J.
,
Rintoul
,
J. L.
,
Al-Hashimi
,
O.
,
Faraji
,
F.
,
Janowich
,
J.
, et al
(
2013
).
Video game training enhances cognitive control in older adults
.
Nature
,
501
,
97
101
.
Arnemann
,
K. L.
,
Chen
,
A. J.-W.
,
Novakovic-Agopian
,
T.
,
Gratton
,
C.
,
Nomura
,
E. M.
, &
D'Esposito
,
M.
(
2015
).
Functional brain network modularity predicts response to cognitive training after brain injury
.
Neurology
,
84
,
1568
1574
.
Awh
,
E.
,
Jonides
,
J.
,
Smith
,
E. E.
,
Schumacher
,
E. H.
,
Koeppe
,
R. A.
, &
Katz
,
S.
(
1996
).
Dissociation of storage and rehearsal in verbal working memory: Evidence from positron emission tomography
.
Psychological Science
,
7
,
25
31
.
Baltes
,
P. B.
, &
Lindenberger
,
U.
(
1988
).
On the range of cognitive plasticity in old age as a function of experience: 15 years of intervention research
.
Behavior Therapy
,
19
,
283
300
.
Baniqued
,
P. L.
,
Allen
,
C. M.
,
Kranz
,
M. B.
,
Johnson
,
K.
,
Sipolins
,
A.
,
Dickens
,
C.
, et al
(
2015
).
Working memory, reasoning, and task switching training: Transfer effects, limitations, and great expectations?
PLoS One
,
10
,
e0142169
.
Baniqued
,
P. L.
,
Kranz
,
M. B.
,
Voss
,
M. W.
,
Lee
,
H.
,
Cosman
,
J. D.
,
Severson
,
J.
, et al
(
2014
).
Cognitive training with casual video games: Points to consider
.
Frontiers in Psychology
,
4
,
1010
.
Bell
,
A. J.
, &
Sejnowski
,
T. J.
(
1995
).
An information-maximization approach to blind separation and blind deconvolution
.
Neural Computation
,
7
,
1129
1159
.
Braver
,
T. S.
,
Cohen
,
J. D.
,
Nystrom
,
L. E.
,
Jonides
,
J.
,
Smith
,
E. E.
, &
Noll
,
D. C.
(
1997
).
A parametric study of prefrontal cortex involvement in human working memory
.
Neuroimage
,
5
,
49
62
.
Calhoun
,
V. D.
,
Adali
,
T.
,
Pearlson
,
G. D.
, &
Pekar
,
J. J.
(
2001
).
A method for making group inferences from functional MRI data using independent component analysis
.
Human Brain Mapping
,
14
,
140
151
.
Calhoun
,
V. D.
,
Kiehl
,
K. A.
,
Liddle
,
P. F.
, &
Pearlson
,
G. D.
(
2004
).
Aberrant localization of synchronous hemodynamic activity in auditory cortex reliably characterizes schizophrenia
.
Biological Psychiatry
,
55
,
842
849
.
Cañas
,
A.
,
Juncadella
,
M.
,
Lau
,
R.
,
Gabarrós
,
A.
, &
Hernández
,
M.
(
2018
).
Working memory deficits after lesions involving the supplementary motor area
.
Frontiers in Psychology
,
9
,
765
.
Carpenter
,
A. F.
,
Baud-Bovy
,
G.
,
Georgopoulos
,
A. P.
, &
Pellizzer
,
G.
(
2018
).
Encoding of serial order in working memory: Neuronal activity in motor, premotor, and prefrontal cortex during a memory scanning task
.
Journal of Neuroscience
,
38
,
4912
4933
.
Chein
,
J. M.
, &
Fiez
,
J. A.
(
2001
).
Dissociation of verbal working memory system components using a delayed serial recall task
.
Cerebral Cortex
,
11
,
1003
1014
.
Chooi
,
W.-T.
, &
Thompson
,
L. A.
(
2012
).
Working memory training does not improve intelligence in healthy young adults
.
Intelligence
,
40
,
531
542
.
Cohen
,
J. D.
,
Perlstein
,
W. M.
,
Braver
,
T. S.
,
Nystrom
,
L. E.
,
Noll
,
D. C.
,
Jonides
,
J.
, et al
(
1997
).
Temporal dynamics of brain activation during a working memory task
.
Nature
,
386
,
604
608
.
Craik
,
F. I. M.
(
1986
).
A functional account of age differences in memory
. In
F.
Klix
&
H.
Hagendorf
(Eds.),
Human memory and cognitive capabilities: Mechanisms and performances
(pp.
409
422
).
Amsterdam
:
Elsevier
.
Curtis
,
C. E.
(
2006
).
Prefrontal and parietal contributions to spatial working memory
.
Neuroscience
,
139
,
173
180
.
De Vogelaere
,
F.
,
Santens
,
P.
,
Achten
,
E.
,
Boon
,
P.
, &
Vingerhoets
,
G.
(
2012
).
Altered default-mode network activation in mild cognitive impairment compared with healthy aging
.
Neuroradiology
,
54
,
1195
1206
.
Folstein
,
M. F.
,
Folstein
,
S. E.
, &
McHugh
,
P. R.
(
1975
).
“Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician
.
Journal of Psychiatric Research
,
12
,
189
198
.
Frank
,
M. J.
,
Loughry
,
B.
, &
O'Reilly
,
R. C.
(
2001
).
Interactions between frontal cortex and basal ganglia in working memory: A computational model
.
Cognitive, Affective & Behavioral Neuroscience
,
1
,
137
160
.
Gabrieli
,
J. D. E.
,
Singh
,
J.
,
Stebbins
,
G. T.
, &
Goetz
,
C. G.
(
1996
).
Reduced working memory span in Parkinson's disease: Evidence for the role of frontostriatal system in working and strategic memory
.
Neuropsychology
,
10
,
322
332
.
Gallen
,
C. L.
,
Baniqued
,
P. L.
,
Chapman
,
S. B.
,
Aslan
,
S.
,
Keebler
,
M.
,
Didehbani
,
N.
, et al
(
2016
).
Modular brain network organization predicts response to cognitive training in older adults
.
PLoS One
,
11
,
e0169015
.
Gottwald
,
J. M.
,
Achermann
,
S.
,
Marciszko
,
C.
,
Lindskog
,
M.
, &
Gredebäck
,
G.
(
2016
).
An embodied account of early executive-function development: Prospective motor control in infancy is related to inhibition and working memory
.
Psychological Science
,
27
,
1600
1610
.
Griffanti
,
L.
,
Douaud
,
G.
,
Bijsterbosch
,
J.
,
Evangelisti
,
S.
,
Alfaro-Almagro
,
F.
,
Glasser
,
M. F.
, et al
(
2017
).
Hand classification of fMRI ICA noise components
.
Neuroimage
,
154
,
188
205
.
Hoekzema
,
E.
,
Carmona
,
S.
,
Ramos-Quiroga
,
J. A.
,
Richarte Fernández
,
V.
,
Bosch
,
R.
,
Soliva
,
J. C.
, et al
(
2014
).
An independent components and functional connectivity analysis of resting state fMRI data points to neural network dysregulation in adult ADHD
.
Human Brain Mapping
,
35
,
1261
1272
.
Ikkai
,
A.
, &
Curtis
,
C. E.
(
2011
).
Common neural mechanisms supporting spatial working memory, attention and motor intention
.
Neuropsychologia
,
49
,
1428
1434
.
Iordan
,
A. D.
,
Cooke
,
K. A.
,
Moored
,
K. D.
,
Katz
,
B.
,
Buschkuehl
,
M.
,
Jaeggi
,
S. M.
, et al
(
2018
).
Aging and network properties: Stability over time and links with learning during working memory training
.
Frontiers in Aging Neuroscience
,
9
,
419
.
Jaeggi
,
S. M.
,
Buschkuehl
,
M.
,
Jonides
,
J.
, &
Perrig
,
W. J.
(
2008
).
Improving fluid intelligence with training on working memory
.
Proceedings of the National Academy of Sciences, U.S.A.
,
105
,
6829
6833
.
Jaeggi
,
S. M.
,
Studer-Luethi
,
B.
,
Buschkuehl
,
M.
,
Su
,
Y.-F.
,
Jonides
,
J.
, &
Perrig
,
W. J.
(
2010
).
The relationship between n-back performance and matrix reasoning—Implications for training and transfer
.
Intelligence
,
38
,
625
635
.
Karbach
,
J.
, &
Kray
,
J.
(
2009
).
How useful is executive control training? Age differences in near and far transfer of task-switching training
.
Developmental Science
,
12
,
978
990
.
Kay
,
B. P.
,
DiFrancesco
,
M. W.
,
Privitera
,
M. D.
,
Gotman
,
J.
,
Holland
,
S. K.
, &
Szaflarski
,
J. P.
(
2013
).
Reduced default mode network connectivity in treatment-resistant idiopathic generalized epilepsy
.
Epilepsia
,
54
,
461
470
.
Kühn
,
S.
,
Schmiedek
,
F.
,
Schott
,
B.
,
Ratcliff
,
R.
,
Heinze
,
H.-J.
,
Düzel
,
E.
, et al
(
2011
).
Brain areas consistently linked to individual differences in perceptual decision-making in younger as well as older adults before and after training
.
Journal of Cognitive Neuroscience
,
23
,
2147
2158
.
Lustig
,
C.
,
Shah
,
P.
,
Seidler
,
R.
, &
Reuter-Lorenz
,
P. A.
(
2009
).
Aging, training, and the brain: A review and future directions
.
Neuropsychology Review
,
19
,
504
522
.
Mackey
,
W. E.
, &
Curtis
,
C. E.
(
2017
).
Distinct contributions by frontal and parietal cortices support working memory
.
Scientific Reports
,
7
,
6188
.
Meda
,
S. A.
,
Stevens
,
M. C.
,
Folley
,
B. S.
,
Calhoun
,
V. D.
, &
Pearlson
,
G. D.
(
2009
).
Evidence for anomalous network connectivity during working memory encoding in schizophrenia: An ICA based analysis
.
PLoS One
,
4
,
e7911
.
Melby-Lervåg
,
M.
,
Redick
,
T. S.
, &
Hulme
,
C.
(
2016
).
Working memory training does not improve performance on measures of intelligence or other measures of “far transfer”: Evidence from a meta-analytic review
.
Perspectives on Psychological Science
,
11
,
512
534
.
Müller
,
N. G.
, &
Knight
,
R. T.
(
2006
).
The functional neuroanatomy of working memory: Contributions of human brain lesion studies
.
Neuroscience
,
139
,
51
58
.
Noack
,
H.
,
Lövdén
,
M.
, &
Schmiedek
,
F.
(
2014
).
On the validity and generality of transfer effects in cognitive training research
.
Psychological Research
,
78
,
773
789
.
O'Bryant
,
S. E.
,
Humphreys
,
J. D.
,
Smith
,
G. E.
,
Ivnik
,
R. J.
,
Graff-Radford
,
N. R.
,
Petersen
,
R. C.
, et al
(
2008
).
Detecting dementia with the mini-mental state examination in highly educated individuals
.
Archives of Neurology
,
65
,
963
967
.
Onoda
,
K.
,
Ishihara
,
M.
, &
Yamaguchi
,
S.
(
2012
).
Decreased functional connectivity by aging is associated with cognitive decline
.
Journal of Cognitive Neuroscience
,
24
,
2186
2198
.
Owen
,
A. M.
,
Hampshire
,
A.
,
Grahn
,
J. A.
,
Stenton
,
R.
,
Dajani
,
S.
,
Burns
,
A. S.
, et al
(
2010
).
Putting brain training to the test
.
Nature
,
465
,
775
778
.
Owen
,
A. M.
,
McMillan
,
K. M.
,
Laird
,
A. R.
, &
Bullmore
,
E.
(
2005
).
N-back working memory paradigm: A meta-analysis of normative functional neuroimaging studies
.
Human Brain Mapping
,
25
,
46
59
.
Park
,
D. C.
,
Lodi-Smith
,
J.
,
Drew
,
L.
,
Haber
,
S.
,
Hebrank
,
A.
,
Bischof
,
G. N.
, et al
(
2014
).
The impact of sustained engagement on cognitive function in older adults: The Synapse Project
.
Psychological Science
,
25
,
103
112
.
Redick
,
T. S.
,
Shipstead
,
Z.
,
Harrison
,
T. L.
,
Hicks
,
K. L.
,
Fried
,
D. E.
,
Hambrick
,
D. Z.
, et al
(
2013
).
No evidence of intelligence improvement after working memory training: A randomized, placebo-controlled study
.
Journal of Experimental Psychology: General
,
142
,
359
379
.
Rottschy
,
C.
,
Langner
,
R.
,
Dogan
,
I.
,
Reetz
,
K.
,
Laird
,
A. R.
,
Schulz
,
J. B.
, et al
(
2012
).
Modelling neural correlates of working memory: A coordinate-based meta-analysis
.
Neuroimage
,
60
,
830
846
.
Salami
,
A.
,
Pudas
,
S.
, &
Nyberg
,
L.
(
2014
).
Elevated hippocampal resting-state connectivity underlies deficient neurocognitive function in aging
.
Proceedings of the National Academy of Sciences, U.S.A.
,
111
,
17654
17659
.
Salthouse
,
T. A.
,
Babcock
,
R. L.
, &
Shaw
,
R. J.
(
1991
).
Effects of adult age on structural and operational capacities in working memory
.
Psychology and Aging
,
6
,
118
127
.
Schmiedek
,
F.
,
Lövdén
,
M.
, &
Lindenberger
,
U.
(
2010
).
Hundred days of cognitive training enhance broad cognitive abilities in adulthood: Findings from the COGITO study
.
Frontiers in Aging Neuroscience
,
2
,
27
.
Schmiedek
,
F.
,
Lövdén
,
M.
, &
Lindenberger
,
U.
(
2014
).
Younger adults show long-term effects of cognitive training on broad cognitive abilities over 2 years
.
Developmental Psychology
,
50
,
2304
2310
.
Smith
,
E. E.
, &
Jonides
,
J.
(
1998
).
Neuroimaging analyses of human working memory
.
Proceedings of the National Academy of Sciences, U.S.A.
,
95
,
12061
12068
.
van Muijden
,
J.
,
Band
,
G. P. H.
, &
Hommel
,
B.
(
2012
).
Online games training aging brains: Limited transfer to cognitive control functions
.
Frontiers in Human Neuroscience
,
6
,
221
.
Verhaeghen
,
P.
,
Marcoen
,
A.
, &
Goossens
,
L.
(
1992
).
Improving memory performance in the aged through mnemonic training: A meta-analytic study
.
Psychology and Aging
,
7
,
242
251
.
Willis
,
S. L.
,
Tennstedt
,
S. L.
,
Marsiske
,
M.
,
Ball
,
K.
,
Elias
,
J.
,
Koepke
,
K. M.
, et al
(
2006
).
Long-term effects of cognitive training on everyday functional outcomes in older adults
.
JAMA
,
296
,
2805
2814
.