Following tDCS stimulation, polarization occurs both in the soma and synaptic terminals of pyramidal and non-pyramidal tract neurons (Stagg and Nitsche, 2011; Rahman et al., 2013), with cell compartment polarization being polarity specific with proximity to the anode or cathode. Studies examining effects of tDCS on motor cortex find that the cortical excitability is increased under the anode but decreased under the cathode (Nitsche and Paulus, 2000; Stagg et al., 2009). Studies examining effects of tDCS on cognitive tasks generally show heightened functionality with an anode over targeted cortex, with weaker or null effects when the cathode was over targeted cortex (Jacobson et al., 2012).

To date, only a few studies have investigated effects of tDCS on real world-relevant cognitive multi-tasks—the focus of our research efforts. Decision-making was altered by bilateral stimulation of the dorsolateral PFC (dlPFC) (Fecteau et al., 2007). A “threat detection” task developed to train military personnel to detect concealed objects benefited from placing the anode over either F10 or P4 (contralateral shoulder cathode). Those locations were selected based on fMRI BOLD signals before and after training to an “expert” level of performance. Anodal tDCS to both sites improved threat detection performance compared to sham (Clark et al., 2012). A replication of this study by our group used signal detection theory to demonstrate that performance improvements on this task were attributable to increased perceptual sensitivity and not to changes in response bias. Performance improvements were retained for at least 24 h after stimulation (Falcone et al., 2012). This small literature, while suggesting that tDCS may be useful in training in real-world settings, has been limited by studies that: (a) used tasks that were mastered relatively quickly; (b) did not separately assess subtasks of multi-tasks; (c) did not compare a range of stimulation montages.

In concert with the growing recognition that cognition emerges from core large-scale brain networks (Bressler and Menon, 2010), recent evidence suggests that certain resting state networks have an important role in cognitive training. Corbetta et al. (2008) argued from a large body of neuroimaging work that the dorsal attention network plays a role in focusing visuospatial attention on relevant events while ignoring irrelevant events by suppressing the ventral network to prevent reorienting. Although this hypothesis was developed based on evidence from attention tasks, it has recently been found to be relevant to training effects. Nodes in the dorsal attention network [including intraparietal sulcus (IPS) and superior parietal cortex (SPC)] selectively undergo structural and functional changes following both motor and cognitive training (Draganski et al., 2004; Olesen et al., 2004; Dahlin et al., 2008; Lewis et al., 2009; Scholz et al., 2009; Takeuchi et al., 2010; Strenziok et al., 2014). This evidence suggests that focusing attention in the face of distraction is important in successful training, regardless of the specific training task. The right hemisphere ventral attention network [with nodes in right ventral frontal (VFC) and right temporo-parietal junction (TPJ)] is important for redirecting focused attention (Corbetta et al., 2008). The ventral attention network is claimed to be suppressed when attention is focused, but transiently activated along with the dorsal network when attention is redirected in response to a task-relevant event (Corbetta et al., 2008). Consistent with that hypothesis, changes in nodes of the ventral attention network have also been observed following training (Lee et al., 2012; Prakash et al., 2012). Relevant to the present study, tDCS selectively alters functional connectivity in resting state networks (Polanía et al., 2011), including the dorsal attention network (Keeser et al., 2011).

Considered together, there is growing evidence that both dorsal and ventral attention networks play an important role in cognitive training, suggesting that tDCS could be effective in facilitating training to the extent that it heightens activity in those networks.

To assess effects of tDCS on the initial phase of multi-task training, we tested hypotheses on the effect of stimulating specific resting state networks during the first 30 min of training on component subtasks of a multi-task typically learned over weeks. We selected the Space Fortress (SF) task (Mane and Donchin, 1989) for the following reasons: (a) SF is a complex task requiring integration of spatial, motor, and executive function demands, making it a good simulation of real-world multi-tasks; (b) SF is demanding and typically learned over a number of weeks, making it a good simulation of tasks that require considerable time to master; (c) SF subtasks provide separate measures of different task demands, allowing hypotheses to be tested about the relation between site of stimulation and performance on subtask measures during training; (d) we recently found that SF training altered functional connectivity between the dorsal attention network and temporal association cortex (Strenziok et al., 2014), consistent with other findings of altered connectivity involving the dorsal network following cognitive training (Lewis et al., 2009; Takeuchi et al., 2010).

We compared stimulation montages that were selected based on the neurocognitive hypothesis of attentional reorienting (Corbetta et al., 2008) and modeled to predict regional brain current flow (Datta et al., 2009). Overall, we hypothesized that tDCS would exert selective effects on subtasks of the multi-task very early in training. Specifically, we hypothesized that tDCS aimed at either right or left IPS in the dorsal attention network (C4, C3, respectively), associated with focusing visuospatial attention, would facilitate learning ship control and ship velocity subtasks that require attention to be focused on flying the ship. We also hypothesized that tDCS aimed at ventral prefrontal (F10) and TPJ (C4) in the right hemisphere ventral attention network, associated with redirecting attention, would facilitate learning of speed and points subtasks. Those two subtasks require redirection of attention from flying the ship toward symbols on the screen. As both dorsal and ventral networks are claimed to be active if a task-relevant event triggers reorientation while attention is focused (Corbetta et al., 2008), we predicted that the strongest benefits on redirecting attention from control and velocity to speed and points subtasks would be when both dorsal and ventral networks are simultaneously stimulated by right parietal tDCS (C4) inducing electrical fields over both IPS and TPJ. We tested our hypotheses by random assignment to sham or active right or left hemisphere ventral prefrontal (F9, F10) or parietal stimulation (C3, C4) during the first 30 min of a 45 min training session on the SF task. We acknowledge that uncertainties exist in the relation between the tDCS-induced electrical field and the brain regions and networks activated as a consequence and our approach was based on above-reviewed evidence of the role of the dorsal and ventral attention networks in training effects.

A total of 100 undergraduate students were recruited to participate in this study. Participants were recruited through George Mason University's participant recruitment system and students received course credit for their participation. All participants were right-handed, based on self-report and on experimenter observation of the hand used to fill out the questionnaires. Participants had normal or corrected to normal vision (20/30) based on the Rosenbaum pocket screener, no history of psychiatric or mental illness based on a questionnaire, and were not taking psychoactive medication. All participants provided written consent with procedures approved by the IRB of George Mason University. Participants completed a questionnaire to report the average number of hours spent per week playing different types of video games, including first-person shooter-type games. This is important in light of evidence that people who frequently play first-person shooter-type games have enhanced attentional abilities (Green and Bavelier, 2012).

The laboratory testing room was equipped with a computer with a Windows operating system and a 17 inch monitor. A joystick was operated by the right hand and a three-button mouse was operated by the left hand. Participants were seated at an average distance of 24 inches from the monitor but were allowed to adjust the distance and chair position to their comfort.

The Space Fortress task was developed by cognitive psychologists to study learning of complex and important real-world tasks. We briefly describe this complex task here, but details of design and scoring can also be found in Gopher et al. (1989, 1994). The goal of this game is to destroy the Space Fortress in the center of the screen (Figure 1) while dealing with “friend” and “foe” mines and monitoring resources (i.e., number of missiles). The player uses a joystick in the right hand to navigate their ship while firing missiles at the Space Fortress, which becomes vulnerable to destruction after being hit by missiles 10 times with at least 250 ms between any two hits. After the Space Fortress becomes vulnerable, the player must destroy the Fortress by firing two missiles in quick succession (less than 250 ms between shots). The vulnerability of the Fortress is reset to 0 if any two shots are made within 250 ms of each other prior to becoming vulnerable or if the player's space ship is destroyed. Each trial was one game lasting 3 min. Performance measurements include total score, which is the sum of 4 subscores, each assessing a different component of the game; control subscore, velocity subscore, speed subscore, and points subscore.

Before participants started Game 2, electrodes were positioned on the head and connected to the stimulation unit (ActivaDose II, ActivaTek). Electrode placement followed the 10–20 EEG system. tDCS was delivered via two 2 × 2 in. electrodes fitted with saline soaked sponges (resulting in sponge contact area of 1.25 × 1.25 in.), with the anode placed over the target area and cathode placed on the contralateral upper arm. Use of an extracephalic reference has been investigated for its potential to influence brain stem autonomic functions. In a recent study, no differential effect of real or sham stimulation with an extracephalic reference was found on heart rate, respiratory rate, blood pressure, or sympathetic tone (Vandermeeren et al., 2010).

Participants were randomly assigned to receive either sham stimulation (0.2 mA with anode at F10) or 2 mA stimulation with anode placed at (a) C3, (b) C4, (c) F9, or (d) F10 (cathode on contralateral arm). Previous studies have demonstrated the safety of 2 mA stimulation on a range of tasks (Keeser et al., 2011; Chib et al., 2013; Clark and Parasuraman, 2014), including complex, real-world tasks (Fecteau et al., 2007; Clark et al., 2012; Falcone et al., 2012). Furthermore, studies comparing dosage levels have demonstrated performance modulation with 2 mA, but not with 1 mA (Boggio et al., 2006; Teo et al., 2011; Moos et al., 2012) or 0.6 mA (Clark et al., 2012) compared to sham. Participants were blinded as to whether they were administered active or sham stimulation.

Sham stimulation was administered by first ramping the current up to 2mA and then slowly decreasing the current to 0.2 mA where it remained. This ramp-up/ramp-down method for sham stimulation has been shown to be indistinguishable from active stimulation to the participants (Ambrus et al., 2012) and has been used in several previous studies (Fecteau et al., 2007; Nelson et al., 2014).

The timed 30 min of stimulation automatically ended mid-way through Game 9 (the third total emphasis game). After completion of this game, the electrodes were removed and the participant completed the remainder of the training and post-training assessments without stimulation (see Figure 2). A mood questionnaire in which participants rated their current feelings of fatigue, boredom, and sadness was completed both before and after stimulation to assess any changes in affect due to stimulation. A sensation questionnaire was completed by participants at three timepoints (at ramp-up, after 5 min, and after 15 min of stimulation) during both active and sham stimulation. This required rating on a scale of 1–10 the degree of “itching,” “burning/heating,” and “tingling”-each assessed separately. Due to variability in the first assessment, which was given during the ramp-up (for active conditions) or ramp-up/ramp-down (during sham condition) period, sensation ratings from the latter two assessments were used to evaluate any potential effect of sensation on performance.

Each stimulation montage (anode at F9, F10, C3, C4 to cathode on contralateral arm) was modeled with Finite Element Method (FEM) models to predict regional brain current flow (Figure 3). High-resolution head models were created from a previously segmented adult male based on a T1 MRI scan with a 1 mm isotropic resolution. Models of sponges and electrodes were positioned and resampled into the image volume before a voxel-based volumetric mesh was generated using ScanCAD and ScanIP (Simpleware Ltd., Exeter, UK). This mesh was imported into a FEM solver (COMSOL 3.5a, Dassault Systèmes Corp., Waltham, MA) modeling electrostatic physics. One of nine conductivities were assigned to the various materials: skin (0.465 S/m), fat (0.025 S/m), skull (0.01 S/m), cerebrospinal fluid (1.65 S/m), gray matter (0.276 S/m), white matter (0.126 S/m), air (1e-15 S/m), electrode (5.99e7 S/m), saline-soaked sponge (1.4 S/m). The field equation (Lapace, ∇ · (σ∇V) = 0) was solved with boundary conditions set to: insulated on the skin surface, ground on the cathode surface, inward current density on the anode surface (see details in Datta et al., 2009).

The FEM model for F9 and F10 anodes (cathode on contralateral arm) generates electrical fields in lateral and ventral aspects of the temporal and prefrontal cortices (Figure 3), with only minor spread to the ventral surface of the contralateral hemisphere. The FEM model for the C3 and C4 anodes (cathode on contralateral arm) generates electrical fields with a focus over the TPJ, and including lateral temporal and parietal cortices, largely within the stimulated hemisphere. Thus, current flow patterns were specific to the electrode montage (Bikson et al., 2010). Based on these models, we targeted networks with electrode locations as follows: (a) the dorsal attention network with C3 and C4 (left and right IPS, respectively); (b) the right hemisphere ventral attention network with F10 (right ventral prefrontal) and C4 (right TPJ). C4 may stimulate both right IPS and TPJ, as will be discussed below.

An omnibus mixed-design ANOVA was conducted on total score with stimulation group as the between subjects factor and game as the within subjects factor. Typically all subscores are negative during the first sessions of training before becoming positive (e.g., Gopher et al., 1989; Prakash et al., 2012). Since total score is the sum of those scores, it is more negative than the others (Figures 4–6). Total score performance improved over training games (Figure 4, main effect of game [F_{(9.71, 727.89)} = 17.75, p < 0.0001, partial eta squared = 0.027]. The groups differed significantly from each other [F_{(4, 75)} = 3.61, p = 0.009, partial eta squared = 0.151]. Post-hoc pairwise comparisons between stimulation groups showed that the C4 group showed better performance than the sham group (p = 0.004). Group × Game was not significant.

Because we hypothesized that stimulation aimed at dorsal and ventral networks would differentially influence performance of the subtasks (control, velocity, speed and points), we planned separate analyses of each subscore. Each subscore is described in the Methods. Stimulation group was the between subjects factor while game was the within subjects factor in each of the following analyses.

We had hypothesized that control and velocity subtasks (both involved flying the spaceship) would benefit selectively from stimulating the dorsal attention network (anodes at C3 and C4). This was based on literature showing alteration in the dorsal attention network with cognitive training (e.g., Lewis et al., 2009; Takeuchi et al., 2010; Strenziok et al., 2014), and based on evidence that tDCS selectively increased activity in dorsal attention networks (Keeser et al., 2011). Contrary to that prediction, we found that control and velocity subscores (and total score) benefited from C4 but not C3 stimulation. Moreover, for the velocity subscore this effect increased as stimulated training proceeded. Simple effects testing revealed that the effects of C4 stimulation were strongest after stimulation had started and several games had been played (Table 3).

Why did right parietal stimulation selectively affect ability to control the ship? We consider several explanations.

We also observed a benefit of F9 stimulation on the control subscore, but not the velocity subscore. F9 lies over left ventral frontal cortex, not associated with either dorsal or ventral attention networks. This result is not consistent with our hypothesis.

A different pattern of results was seen with the other two subtasks—speed and points. We had hypothesized that because processing irregularly-timed speed and points symbols would require redirection of attention from flying the ship to processing those symbols, the need to intermittently respond to symbols while flying the spaceship would benefit from stimulation of the right hemisphere ventral network. On the points subscore, we found no effects of tDCS. On the speed subscore, the C4 anode group performed better than the sham. The latter could be consistent with an attentional explanation, but also with a motor explanation as the left hand responded. The speed measure depends on (a) remembering which three letters had been displayed at the start of each game to indicate which mines were designated “foe” mines for that game, and (b) taking appropriate action when mines suddenly appeared. The speed measure also showed benefits of a left ventral PFC anode (F9), insofar as that group performed better than the other groups, except for C4. We speculate that the working memory load associated with retaining and updating memory of “foe” mines from game to game may have benefited from stimulation of left PFC, previously found to benefit working memory (e.g., Hoy et al., 2013; Meiron and Lavidor, 2013). However, both left and right PFC stimulation at F3 and F4 have been found to have other effects, e.g., more cautious driving in a simulator (Beeli et al., 2008) and reduced risk taking (Fecteau et al., 2007).

Our finding that multi-task cognitive training benefited from right parietal stimulation is consistent with the growing evidence that attention networks are important in cognitive training. The dorsal attention network has been consistently found to undergo structural and/or functional change following training on juggling (Draganski et al., 2004; Scholz et al., 2009), working memory (Takeuchi et al., 2010), visual search (Lewis et al., 2009), and auditory perception (Strenziok et al., 2014). Two studies observed that reliance on the dorsal attention network decreased as perception improved following perceptual training (Lewis et al., 2009; Strenziok et al., 2014). The ventral attention network has also been implicated in SF training (Lee et al., 2012; Prakash et al., 2012; Strenziok et al., 2014). Those findings are consistent with the literature showing a decreased reliance on control and attentional processing after hours or days of training (reviewed in Kelly and Garavan, 2005). Considered together, this evidence suggests the importance of stimulating very early in training—as in the present study—when visuospatial attention is most active (Kelly and Garavan, 2005; Strenziok et al., 2014) and when plasticity mechanisms favor conversion from short-term to long-term memory (Jones et al., 2001; Maroteaux et al., 2014).

This evidence from the present study that right parietal stimulation selectively facilitated learning of control, velocity, and speed subtasks of the SF multi-task leads to the interpretation that simultaneous stimulation of dorsal and ventral attention networks benefited SF training because of increased efficiency in redirecting visuospatial attention from one subtask to another. Our findings are also broadly consistent with previous work from Kramer's lab showing that SF training led to reduced activation of ventral frontal regions (a node in the ventral attention network) after training (Lee et al., 2012; Prakash et al., 2012). Those two studies assessed activation patterns at the end of training. We modulated activation at the beginning of training. To the extent that SF training requires frequent reorienting of visuospatial attention, the ventral attention network may be important early in training—producing our results—but become less important as the task is learned—leading to the reduced activation of the ventral attention network seen after SF training in Kramer's studies (Lee et al., 2012; Prakash et al., 2012).

How do we reconcile our findings of the benefits of simultaneous stimulation of dorsal and ventral attention networks on ship control measures with findings from other training tasks found to induce changes specifically in the dorsal attention network (Lewis et al., 2009; Lövdén et al., 2010; Takeuchi et al., 2010; Strenziok et al., 2014)? We speculate that cognitive training that does not require multi-tasking (e.g., the perception training tasks used by Lewis et al., 2009 and Strenziok et al., 2014) may rely on focused attention and hence on the dorsal attention network. On the other hand, training that does require multi-tasking, like SF, may rely more on the ability to redirect attention controlled by the ventral attention network (Lee et al., 2012; Prakash et al., 2012), found to be activated in conjunction with the dorsal attention network (Corbetta et al., 2008). It should be noted that while this explanation based on the role of attention in cognitive training is consistent with findings from ship control and velocity subscores, it does not explain the effectiveness of left ventral PFC (F9) stimulation for the speed measure.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.