Anodal tDCS applied during multitasking training leads to transferable performance gains

Cognitive training can lead to performance improvements that are specific to the tasks trained. Recent research has suggested that transcranial direct current stimulation (tDCS) applied during training of a simple response-selection paradigm can broaden performance benefits to an untrained task. Here we assessed the impact of combined tDCS and training on multitasking, stimulus-response mapping specificity, response-inhibition, and spatial attention performance in a cohort of healthy adults. Participants trained over four days with concurrent tDCS – anodal, cathodal, or sham – applied to the left prefrontal cortex. Immediately prior to, 1 day after, and 2 weeks after training, performance was assessed on the trained multitasking paradigm, an untrained multitasking paradigm, a go/no-go inhibition task, and a visual search task. Training combined with anodal tDCS, compared with training plus cathodal or sham stimulation, enhanced performance for the untrained multitasking paradigm and visual search tasks. By contrast, there were no training benefits for the go/no-go task. Our findings demonstrate that anodal tDCS combined with multitasking training can extend to untrained multitasking paradigms as well as spatial attention, but with no extension to the domain of response inhibition.

To address these issues, we trained participants in a multitasking paradigm as they underwent concurrent tDCS. Participants received anodal, cathodal, or sham tDCS to the left PFC. Immediately prior to training, 1 day after training, and 2 weeks after training, participants completed a battery of four tasks. The untrained tasks were selected to measure the potential for "near" and "far" transfer 26 . For near transfer, a multitasking paradigm identical to that which participants trained upon was used, but with different stimuli. Previous research has shown that multitasking training alone does not lead to transfer to the same task with new stimuli 4 . For far transfer, visual search and go/no-go tasks were employed. These tasks tap distinct cognitive processes to multitasking (visual search -spatial attention; go/no-go -response inhibition), but visual search has been linked to activity in left prefrontal cortex 27 , while go/no-go is more commonly associated with activity in right PFC 28,29 . Hence, the two tasks vary in the extent to which the neural substrates overlap with the trained multitasking paradigm. By assessing whether the effects of combined training and tDCS generalise to one or both tasks, we were able to examine the extent and condition of training transfer following combined tDCS and cognitive control training.

Results
To examine the key effects of stimulation and training on reaction times the data for each of the four tasks were submitted to ANOVAs with the between-subjects factor of stimulation group, and within-subjects factors including session (e.g. pre-and post-training). In addition, the baseline performance (pre-training) across the anodal, cathodal, and sham stimulation groups was subjected to a Bayesian ANOVA to provide stringent evidence for the null hypothesis (no difference before training between the stimulation groups).
Accuracy. Performance accuracy was analysed for all tasks, with the exception of go responses in the go/ no-go task, as for these responses performance was at ceiling. For each of the tasks, stimulation group did not interact with any other factors (F < 2.35, p > 0.1 for all), with one exception: for the visual search task, from pre-to post-training, anodal tDCS improved accuracy for set size 8 only, whereas cathodal stimulation led to an improvement for set size 16 only (F(4,112) = 2.46, p = 0.049, η 2 p = 0.08). These effects were not present when comparing pre-training performance to the follow-up session (F(4,108) = 1.83, p = 0.129, η 2 p = 0.06). All analysis that follow were conducted on reaction times. Trained on multitasking paradigm. As predicted from previous work 16,30,31 , across the four training sessions there were substantial reductions in reaction times for both single-and dual-task trials (see Fig. 2; main effect of training session: F(3,168) = 23.359, p < 0.001, η 2 p = 0.294). However, there were no differences in performance across the three stimulation groups (session × group interaction: F(6, 168) = 1.371, p = 0.229, η 2 p = 0.047), and no stimulation effect that varied between the single-and dual-task trials (session × group × trial type interaction: F(6, 168) = 0.403, p = 0.876, η 2 p = 0.014). Comparing the data from pre-to post-training, there was a clear multitasking cost, with longer reaction times for the dual-than single-task trials (mean difference = 333 ms; main effect of trial type: F(1, 56) = 859.004, p < 0.001, η 2 p = 0.939). In addition, for all the stimulation groups, there was a marked improvement in reaction times from pre-to post-training (Fig. 3A, mean improvement = 98 ms, main effect of session: F(1, 56) = 59.185, p < 0.001, η 2 p = 0.514). The dual-task trials showed greater benefit from training (122 ms) than single-task trials (73 ms; session × trial type interaction: F(1, 56) = 16.479, p < 0.001, η 2 p = 0.227). Again, performance improvement with training was comparable for the three stimulation groups (session × group interaction: F(2, 56) = 0.297, p = 0.744, η 2 p = 0.011), and this was true for both the single-and dual-task trials (session × group × trial type interaction: F(2, 56) = 0.018, p = 0.982, η 2 p = 0.001). To assess for any longer term benefits of training and stimulation, data from the pre-training and follow-up sessions were compared (see Fig. 3B). Fifty-six of the original participants completed the follow-up session two weeks after the final training session (anodal group: 20, cathodal group: 18, sham group: 18). Of those participants, one had a missing data set for the trained task. Of import, even after 2 weeks without training, the benefits to performance from training were still apparent, with all three groups showing a reduction in reaction times for the follow-up session (main effect of session: F(1, 52) = 5.939, p = 0.018, η 2 p = 0.103), and this improvement was again greater for the dual-task trials (56 ms) than the single-task trials (17 ms; session × trial type interaction: F(1, 52) = 9.418, p = 0.003, η 2 p = 0.154). There did not appear to be any added benefit from the application of tDCS, with comparable performance improvements for the three groups (all effects of stimulation: p > 0.3, η 2 p < 0.05).
Performance was compared from pre-training to the follow-up session to determine whether the training and stimulation benefits endured (see Fig. 4B). Whilst all groups showed improvement in the speed of responses (46 ms; main effect of session: F(1, 53) = 10.775, p = 0.002, η 2 p = 0.169), the added performance benefits for the anodal stimulation group had reduced and were no longer significant (session × stimulation group interaction: F(2, 53) = 1.397, p = 0.256, η 2 p = 0.05) relative to cathodal and sham stimulation. Visual search task. To address the possibility of far transfer, performance for the visual search task was compared from pre-to post-training (see Fig. 5A). Overall, there were substantial improvements in performance for all stimulation groups (mean improvement = 109 ms, main effect of session: F(1, 56) = 76.901, p < 0.001, η 2 p = 0.579). As expected, reaction times increased with distractor set size (mean difference between set size 8 and 16 = 469 ms, main effect of set size: F(2, 112) = 364.25, p < 0.001, η 2 p = 0.867). Given our previous results 16 , we predicted that anodal tDCS would lead to greater performance improvement, compared with sham and cathodal stimulation, for the larger set sizes. An ANOVA was conducted to compare visual search performance from preto post-training for the largest set size only (16). There was greater improvement in performance for the anodal (166 ms) relative to the sham (103 ms) and cathodal (78 ms) groups for set size 16 (session × stimulation group interaction: F(2, 56) = 3.2, p = 0.048, η 2 p = 0.1). This finding replicates our previous work, but with a different trained task than that used previously.
Looking at the change in performance from pre-training to the follow-up session (Fig. 5B), all groups showed improvement in speed of responses (main effect of session: F(1, 53) = 100.796, p < 0.001, η 2 p = 0.655), and this improvement was greater for the larger set sizes (session × set size interaction: F(2,106) = 11.078, p < 0.001, η 2 p = 0.173). Stimulation group again modulated performance (session × stimulation group × set size interaction: F(4, 106) = 2.598, p = 0.04, η 2 p = 0.089), with greater performance improvement at set size 16 for the anodal (217 ms) than the cathodal (125 ms) and sham (162 ms) groups. Hence, the performance enhancement for the anodal tDCS group at the largest set size, in our far transfer measure, remained present two weeks after cessation of training.
Go/no-go task. There were improvements in the reaction times for all groups (Fig. 6A, Table 2. Mean reaction times (ms) for each session, for each of the four tasks and relevant conditions. For the go/no-go task, reaction times are presented for the 'go' trials only. The follow-up sessions comprised a slightly reduced N for the cathodal and sham groups (highlighted in italics).

Discussion
Our findings reveal transferable performance gains when anodal tDCS is applied to the left PFC during multitasking training. Anodal tDCS enhanced performance for two of the three transfer tasks: the multitasking paradigm that utilised untrained stimulus-response mappings, and a visual search task. Transfer effects were still evident  at the two-week follow up for the visual search task. By contrast, for the go/no-go task, which involves inhibitory control and relies on activity within brain regions outside of those typically associated with multitasking 28,29 , there was no significant benefit from training and anodal tDCS.
It is possible that the performance gains shown by participants were not due to combined tDCS and training, but to tDCS alone. We believe this is unlikely, however, as a recent study conducted by our group using an analogous approach revealed no evidence for performance gains after multiple sessions of tDCS delivered over left PFC without a concurrent training task 16 . Given the relationship between decision-making and multitasking it seems unlikely tDCS alone would lead to the current pattern of results. In addition, our results are unlikely to be attributable to a general arousal effect, as the benefits we observed were specific to the application of anodal tDCS; cathodal stimulation did not lead to any significant training effects. Moreover, the fact that performance was not enhanced for the go/no-go task indicates that the influence of tDCS was not sufficiently general to improve all tested tasks (e.g., general modulations to sustained attention, or motivation). It is also unlikely the results reflect an effect of tDCS on low level perceptual or motor processes, as all tasks included visual or auditory sensory inputs and manual motor outputs, but not all were affected by combined training and stimulation. In addition, we have previously found modulations to training with tDCS to be unrelated to stimulus modality 22,31 . Finally, through response modelling, we have linked the modulations from combined training and tDCS to central task processes (evidence accumulation/drift rates, Filmer, et al. 16 ).
We chose the left PFC as the target region because this area has previously been implicated in cognitive control, multitasking, and training 16,22,31,32 . Our study provides the first causal evidence for this region's role in multitasking training, and converges with our earlier finding that left PFC plays a causal role in single-task decision-making training 16,31 . Of course, it must be considered that in the present study we targeted only one region of the cortex. Hence it is possible training benefits might be found with stimulation of other regions, and this should be the focus of future investigations. In previous work, we found that left-but not right-PFC  stimulation yielded training and transfer benefits following combined single-task decision making training and anodal tDCS 16 . When taken together with the relevant neuroimaging literature, which suggests a key role for the left PFC specifically in cognitive control and multitasking 30,33-35 , it seems most likely that combined training and brain stimulation effects will be specific to the left PFC, but this remains to be determined.
We did not find any significant modulation of performance for the trained task with anodal tDCS. This is a different pattern of results to an earlier study conducted by our group, in which we observed trained-on task benefits from anodal stimulation 16 . It is possible that any benefits of stimulation for the trained task were masked by the training benefits present without tDCS, and consequently enhancement was only detected for the transfer tasks. Moreover, whereas Filmer et al. 16 employed a relatively difficult 6-response alternative training task, here we used an easier 3-response alternative task, which might have reduced the potential for further improvement above and beyond training alone. Finally, it is also possible that the current study was somewhat underpowered. Although we determined the sample size a priori, based on our own recent findings from a paired tDCS and training study 16 , it is possible that the training effect in the present study was smaller and simply required a larger sample to be detected. Regardless, we can conclude with a high degree of certainty that one of the critical effects of anodal tDCS over left PFC is to generalise the training benefit to a broader, but not universal, set of tasks/processes. That is to say, while the present results are statitistically significant, but modest in terms of effect size, they converge with the observations of Filmer et al. 16 in demonstrating sustained transfer of practice following left PFC tDCS.
Typically, when one set of stimulus-response mappings is trained upon, there is no transfer of the training benefit to a multitasking paradigm with different stimuli 4,36 . In the current study, the application of anodal tDCS generalised the benefits of training, suggesting that learning under these conditions is not bound to specific stimulus-response mappings. Furthermore, the transferred training gains to visual search, conceptually replicating our earlier finding 16 , supports the broad transfer of learning via tDCS. Indeed, visual search has been linked to activity in the left PFC 27 , and yet the other far transfer task, the go/no-go paradigm, which did not show any transfer benefits, is more commonly associated with the right PFC 28,29 . A potential explanation for our findings, therefore, is that tDCS modulates activity within the left PFC which in turn facilitates performance for tasks that tap the same prefrontal cortical area. Alternatively, there might be a single mechanism for the performance enhancement that encompasses dual-and single-task processes, as well as visual search, but not response inhibition. This mechanism could represent a task-general process such as adaptive tuning of neurons to evidence accumulators for the relevant context or task 37 .
The transferable benefit of anodal tDCS remained present 2 weeks after cessation of training for the visual search task, but was less persistent for the control multitasking paradigm. Thus the longevity of the transferable benefits reported here was variable. The current findings are broadly in line with our previous study 16 , in which a response selection task showed a weakened (non-significant) benefit at 2-week follow-up, but a visual search task showed sustained (and significant) performance enhancement. It is possible that increasing the number of training sessions, or modifying the tDCS parameters (e.g., increasing stimulation intensity) might lead to longer lasting effects for near transfer tasks. It is also possible that visual search provides a more sensitive measure of the processes modified by combined anodal tDCS and training.
In conclusion, we here mapped the extent of improvement in an aspect of cognitive control via tDCS applied during training. We show that left prefrontal stimulation combined with concurrent training can improve performance for cognitive control operations associated with multitasking, and this improvement generalises to both near and far cognitive operations. Importantly, performance gains were generalised, but not universal -there was no improvement in response inhibition -illustrating important boundary conditions to the effects of combined training and brain stimulation. Transfer effects were specific to anodal tDCS, which may influence activity within the left PFC. Collectively, this study demonstrates the potential for tDCS as a tool for enhancing training outcomes in individuals with impaired cognition, and specifically operations associated with cognitive control that are particularly susceptible to the effects of disease and ageing.

Materials and Methods
Participants. Eighty-seven right-handed participants from The University of Queensland took part in the study. All participants had normal hearing, and normal or corrected to normal vision, and successfully passed safety screening, which included questions relating to medication use. Four participants were excluded due to missing data, and a further 24 were excluded for poor pre-training performance (>40% errors). Exclusion criteria were established a priori, and exclusions based on participants' pre-training performance occurred without knowledge of the training outcome. The final sample consisted of 59 participants, split into three groups (20,20 and 19 participants in each). Allocation to group was pseudo-random for most of the experiment, with participants being allocated to one of the three groups in turn. However, the last few participants were allocated to specific groups to ensure approximately equivalent group sizes following exclusions. The mean age of the final sample was 21 years (SD = 2 years; 51 women), and was comparable for the anodal (21 years, 18 women), cathodal (21 years, 18 women), and sham (21 years, 15 women) groups. The University of Queensland Human Research Ethics committee approved the study, all participants gave informed, written consent, and the study was run in accorandance with the Declaration of Helsinki.
Stimulation protocol. The stimulation protocol closely followed that of Filmer et al. 16 who observed transferable gains following single-task decision-making training paired with anodal stimulation of left lateral PFC. Each participant completed four sessions on consecutive days, with tDCS (Neuroconn DC-Stimulator) delivered to the left PFC. The stimulation site was localised via the EEG 10-20 system 38 , with the target electrode placed 1 cm posterior to F3 16,22,31 . The reference electrode was placed over the contralateral orbitofrontal region (see Fig. 7A). Both electrodes were 5 × 5 cm saline soaked sponges. The stimulation intensity was 0.7 mA, resulting in a current density of 0.028 mA/cm 2 . We employed this stimulation intensity as we had previously found it was effective in promoting effects with single-task training 16 . The experiment followed a single blinded, between-group design, with each of the three groups of participants receiving a different type of stimulation -anodal, cathodal, or sham. The anodal and cathodal groups received stimulation for a total of 13 minutes, including 30 seconds of the current ramping up and down. The sham group received a total of 1 minute 15 seconds of stimulation, including a 30 second ramp up and down 16,22,31,39 . Trained multitasking paradigm. Participants trained on a sensory-motor paradigm consisting of three different trial types: single visual task, single auditory task, and dual-task (see Fig. 7B). For the single task trials, either a coloured circle (visual) or a sound (auditory) was presented for 200 ms, and participants had to respond as quickly and accurately as possible to the identity of the stimulus. Responses were made via a keyboard. There were three different possible colours (red: RGB 237 32 36, dark green: RGB 10 130 65, dark blue: RGB 44 71 151), and three possible complex tones 16,22,30,31,33 , each associated with a different key on the keyboard. For the dual-task trials, both a sound and a colour were presented simultaneously. Participants were instructed to respond to both stimuli as quickly and as accurately as possible. The three trial types were randomly intermixed within blocks of trials.
Transfer tasks. The experiment also consisted of three tasks to assess transfer of training effects (see Fig. 7C-E). The first transfer task was an alternate version of the trained multitasking paradigm, but with different visual and auditory stimuli. The visual stimuli consisted of symbols (&, #, and %), and the auditory stimuli were three complex tones that were different from those used in the trained task 16,22,30,31,33 . This task tested the extent to which the training and stimulation parameters generalised to different sensory-motor mappings within the broad multitasking context. As noted above, when such training is performed alone, without concurrent brain stimulation, no transfer is observed when stimuli in the multuitasking paradigm are changed (e.g., Garner et al., 2014). The second transfer measure assessed spatial attention with a visual search task, where participants had to locate and respond to a letter 'T' amongst distractor 'L' stimuli. Once located, participants indicated whether the T was oriented through 90 or 270 degrees. The number of distractor 'L's was 8, 12, or 16, and these trial types were randomly intermixed. The third transfer task was a go/no-go paradigm assessing the cognitive control operation of inhibition, which is distinct from that tapped by multitasking 40 . For this task, an abstract line shape was shown on each trial (see Fig. 7E) and participants had to press the 'g' key on the keyboard if it was the 'go' stimulus, or withhold their response if it was the specific 'no-go' stimulus. The go stimulus was presented on 75% of trials. Participants were instructed to respond as quickly and as accurately as they could in all three tasks, with responses made via a keyboard.
ScIenTIfIc REPoRTS | 7: 12988 | DOI:10.1038/s41598-017-13075-y Procedure. An outline of the experimental sessions is shown in Fig. 7A. During the experiment, participants sat approximately 70 cm from a 19″ CRT monitor with a refresh rate of 100 Hz. They initially completed all four tasks -including practice and a baseline measurement for each (baseline measures: trained task = 240 trials, transfer multitasking paradigm 1 = 240 trials, visual search = 240 trials, go/no-go = 80). The trained task, transfer multitasking paradigm, and visual search took approximately 10 minutes to complete, and the go/no-go took approximately 5 minutes to complete. The order in which the tasks were completed was controlled across participants using a Latin square. The electrodes were then fitted and the first training session completed. For the first 10 seconds of stimulation, participants were asked to sit quietly and were reminded of the response mappings for the two tasks. For the following 10 seconds participants were shown the three response keys for the sound stimuli, and could hear the sounds if they pressed a relevant key to remind themselves of the stimulus-response mappings. After this, participants had another 10 seconds where the three response keys for the colour task were shown, with the relevant colour displayed underneath each key. The main task began at the same time as the current ramp-up ceased. The task then lasted for just over 12 minutes, and consisted of 240 trials (80 single-auditory, 80 single-visual, and 80 dual-task). The task was completed 10 seconds into the current ramp down, and participants sat quietly for the remaining 20 seconds of the ramp down. For each of the next three days, participants again undertook the trained task with concurrent stimulation. For the post-training session (Day 5) and the follow up session (2 weeks later), participants completed all four tasks again, but with no brain stimulation.