WM plays a significant role in many cognitive tasks and is controlled by broad frontoparietal networks accessible to tDCS. These features have made it attractive to researchers interested in applying tDCS to studies of cognition. This approach has been successful. The most consistent set of WM data comes from the use of n-back tasks paired with left dorsolateral prefrontal cortex (DLPFC) stimulation in healthy (Ohn et al., 2008; Andrews et al., 2011; Mulquiney et al., 2011; Zaehle et al., 2011) and special populations (Boggio et al., 2006; Jo et al., 2009); for a recent meta-analysis see (Brunoni and Vanderhasselt, 2014); see Table 1. In short, these findings reveal a robust pattern: anodal tDCS to the left DLPFC improves verbal n-back task performance when compared to either sham or cathodal tDCS. This pattern of tDCS-linked WM improvement remains constant across a variety of stimulus intensities, durations, and participant populations. Indeed, recent meta-analysis looking at these studies identified a reliable reaction time improvement during active tDCS in those studies using the n-back and stimulating the DLPFC (Brunoni and Vanderhasselt, 2014, but see also Jacobson et al., 2012).

This consistency highlights at least one notable exception. When we tested healthy older adults in a verbal and visual 2-back task we replicated the anodal tDCS benefit, but only in those with more education (Berryhill and Jones, 2012). Surprisingly, those with less than a college-degree received no benefit from tDCS. Instead their performance revealed a nearly equal and opposite decrement in performance. Furthermore, regardless of whether the stimuli were verbal or visual and whether stimulation site were the left or the right the differences in education predicted whether tDCS helped or hurt WM performance. This was the first paper, to our knowledge, showing that an inhomogeneous population significantly modulated tDCS effects in different directions. A recent paper has used current modeling to clarify that participants only benefited on a verbal 3-back WM task when the tDCS applying DLPFC during a verbal 3-back task current modeling indicates that those who benefited showed significant modulation in the DLPFC current, but those who did not improved did not show significant DLPFC modulation (Kim et al., 2014). They attributed these data to morphological differences in brain structure and where current flow went. This finding reveals the need for refining stimulation targeting by registering an individual's MRI scans with tDCS electrode placement as is the practice in TMS research. However, their data also showed that those who started with higher WM performance garnered greater benefits from tDCS.

There are several other papers pairing tDCS with different WM tasks (e.g., old/new recognition, recall, change detection) and parietal stimulation sites. These data are less clear. For example, we found that anodal tDCS to the right posterior parietal cortex (PPC: P4) selectively interfered with WM probed by old/new recognition, but not with WM probed by recall (Berryhill et al., 2010). We subsequently found that when participants performed two WM tasks of different difficulty levels in the same session, that tDCS effects were only apparent for the more challenging task (Jones and Berryhill, 2012). Importantly, here, again, there were significant and opposing patterns in the data such that young adults with high WM span benefited but those with low WM span performance was impaired after cathodal or anodal tDCS to the right PPC. In contrast, a similar study also applying anodal tDCS to the right PPC reported that those with low WM performance on the sham day performed better during the anodal session on a challenging change detection task (Tseng et al., 2012). However, this study lacked an independent measure of WM capacity to segment their participants. Instead, behavioral performance during the sham session was used. Thus, the observed effects are contaminated by regression to the mean because poor performers during the sham session were likely to perform better at another session regardless of tDCS presence. However, here, again are several data points indicating that population differences predict the direction and magnitude of tDCS effects on WM.

A second issue buried in these data is that effects are apparent when tasks are difficult. Apart from the WM papers just noted there is at least one other analysis that has found that tDCS effects were apparent only when the task demands were difficult. More specifically, in an associative memory task participants learned face-name and place-name pairs and received left or right anodal tDCS. In younger and older adults, tDCS provided a performance benefit only when the participants struggled to produce the correct face or place name, as evidenced by long reaction times (Ross et al., 2010, 2011). One way to think about this in terms of tDCS is that the extra stimulation can serve as a tipping factor. This is consistent with our understanding that tDCS induces changes the changes in neuronal excitability—cells become more (anodal) or less (cathodal) likely to fire action potentials. When tasks are easy, the outcome is clear and the addition of tDCS does not change performance. However, when tasks or even trials are very difficult, tDCS effects emerge. When designing a task to pair with tDCS, it is worth ensuring that the task demands are sufficiently challenging for participants and/or that the more challenging trials can be isolated for separate analysis.

A third emerging issue that becomes more apparent when reviewing the tDCS-WM papers is that the effect sizes tend to be small and the studies are underpowered. For example, although early studies report significant effects with 10 participants, more recent papers tend to include 20 or more participants. One speculation is that early studies tapped homogeneous populations, presumably available lab personnel, to participate and this meant resulted in more consistent performance and tDCS patterns.

A related concern is that as laboratories become more comfortable with the tDCS technique they are subject to added noise from poorly motivated participants. Although this is a problem for many experimental techniques, it may be particularly relevant if unmotivated participants do not engage during challenging tasks or during challenging trials. As mentioned above, the subtle response-shifts induced by tDCS may be particularly sensitive to contamination from poorly motivated participants. However, this notion must be considered as speculative because there are no explicit data testing the role that motivation plays in tDCS designs, although we are currently testing this hypothesis.

Apart from WM, there is broad use of tDCS to investigate wide-ranging cognitive topics (for a recent review see Coffman et al., 2012). A real challenge is the diversity in experimental paradigms and tasks makes it difficult to identify consistent patterns in the tDCS literature. Furthermore, a recent review paper highlighted “foundational” problems associated with tDCS and the impact of variability across participants, issues associated with cognitive set and performance, the reliability of effects over time and current dynamics (Horvath et al., 2014; see also Lopez-Alonso et al., 2014). However, in a few areas of upper-level cognitive domains some consistency is beginning to emerge. In Table 2, we provide an admittedly incomplete survey of cognitive studies employing tDCS reporting significant effects in healthy adults. As in WM, the majority of cognitive studies target frontal stimulation sites and thus, it may not be surprising that the papers describing significant effects of tDCS relate to upper-level cognitive tasks (e.g., response inhibition, memory, decision making). From these scattered findings, it can be difficult to predict the direction of effects in a tDCS study and it can be difficult to know why the effects are as they are. As noted above, these are the studies with positive findings and likely there are many other null findings that would be informative for the research population. In the following section, we raise several points to consider when developing tDCS studies of cognition.

Equal and opposite effects in different populations can easily obscure effects. Researchers interested in relating structure-function relationships can reduce the noise in their data by targeting a homogeneous population. More practically, armed with the knowledge that population differences are pertinent it is advisable to include measures of relevant factors such as WM capacity in a WM study. This permits incorporating some demographic or other factor in the analysis. This is particularly relevant for applied and translational applications of tDCS in development for general use. For this purpose, it is essential to identify population differences and use that information to predict who will garner the greatest benefit from tDCS.

The effect sizes in cognitive studies of tDCS are modest. Thus, it is important to counter low power by enrolling sufficiently large cohorts. Because population differences have been reported showing equal and opposite effects of tDCS, it is likely others exist. Such differences may be adding significant noise and obscuring positive findings.

The effects of tDCS are subtle. It is unlikely that tDCS could significantly influence supraliminal response patterns. As such, it is during near-threshold events that tDCS effect become apparent—for instance, when the task is really difficult. Experimental design should include tasks that are adaptive such that all participants are performing an effortful task. Analyses should be designed to permit separation of easy trials (e.g., high accuracy, fast responses) from more difficult trials.

This issue is related to the use of challenging tasks. We suspect that tDCS is particularly sensitive to participants with low motivation. This may be a particular problem when testing freely available undergraduate volunteers who value course credit more than the research. Low motivation may matter because tDCS effects are subtle and seem to shape performance only over the range in the heart of response functions where there is variability in the response outcome. In other words, tDCS will not change someone's response when it is 100%, but it may push responses from 49% in one direction to 52%.

For each condition and each participant we estimated WM capacity by calculating Cowan's K: [K = set size^* (hit rate—false alarm rate)]. The first question was to determine whether cathodal tDCS to the right parietal lobe interfered with WM grouping mechanisms. The K-values were subjected to repeated-measures ANOVA with the factors of condition [2-ungrouped (2-UG), 3-grouped (3-G), 3-ungrouped (3-UG)] and tDCS session (active, sham). The data replicated the expected behavioral grouping benefit [F_{(2, 64)} = 5.39, p = 0.007, η²_p = 0.14, β = 0.83] such that capacity was significantly larger in the 3-G condition compared to the 2-UG condition (3-G = 1.49, 2-UG = 1.29, p = 0.004, Bonferroni corrected here and throughout); see Figure 2. Additional pairwise comparisons indicated that there were no differences between the 2-ungrouped and 3-ungrouped conditions (3-UG = 1.38, p = 0.26) or between the 3-ungrouped and 3-grouped conditions (p = 0.44). In other words, grouping numerically enhanced performance on the 3-grouped condition, but not significantly. However, the answer to the primary question was that there was no significant effect of tDCS on VWM performance as there was no main effect of tDCS [F_{(1, 32)} = 0.07, p = 0.79] and no significant condition × tDCS interaction [F_{(2, 64)} = 0.18, p = 0.83].

Our concerns regarding participant homogeneity prompted dividing participants into high and low WM capacity groups based on the independent WM measure of combined forward + backward digit span scores (high N = 16: combined digit span score >13; low N = 17: combined digit span score <13). The inclusion of WM capacity in the ANOVA did not change the pattern of the results. The grouping benefit remained [F_{(2, 62)} = 5.26, p = 0.008, η²_p = 0.15, β = 0.82]. As before, pairwise comparisons reveled significant differences between the 2-ungrouped and the 3-grouped condition (p = 0.005), but not between the 2-ungrouped and the 3-ungrouped conditions (p = 0.26) or between the 3-grouped and the 3-ungrouped conditions (p = 0.47). As before, there was no main effect of tDCS [F_{(1, 31)} = 0.05, p = 0.83], and no two- or three-way interactions (all p's > 0.13). In summary, there was no difference in VWM performance in response to grouping via common region between high and low WM capacity individuals. Additionally, diverging from our predictions, there was no effect of cathodal tDCS relative to sham tDCS.

Here, we tested the hypothesis that cathodal tDCS targeting parietal regions involved in grouping processes would disrupt these processes. Contrary to our predictions, tDCS did not modulate grouping benefits in VWM. Furthermore, WM capacity did not contribute to the current results as cathodal tDCS failed to interrupt grouping benefits to WM performance in either high or low WM capacity participants. These findings were unexpected because we previously interrupted WM using cathodal tDCS to the right parietal lobe in a VWM recognition task (Berryhill et al., 2010). One possibility to account for these null results is that participants may not have been effortfully engaged in the task. This interpretation is supported by low VWM performance (e.g., mean K-values for the 2-UG, 3-UG, 3-G conditions: 1.29, 1.38, 1.49 items compared to the ~1.9, 2.3, 2.6 items per condition reported by Xu and Chun, 2007). Anecdotally, in our previous successful study, the participants were largely graduate students known to the experimenter, rather than undergraduates interested in obtaining course extra-credit.

Alternatively, previous fMRI findings revealed bilateral IPS activity during this task (Xu and Chun, 2007). Thus, some might argue that a unilateral stimulation protocol might not have sufficient power to prevent some contralateral compensatory mechanism. However, we think that this is unlikely for several reasons. First, there is evidence supporting right IPL in attending to stimuli across both visual hemifields (e.g., Sheremata et al., 2010; Szczepanski and Kastner, 2013), making it more likely to see disrupted performance in the VWM task after right lateralized stimulation. Second, and perhaps of greater relevance, we previously interrupted VWM using the identical tDCS protocol but different VWM tasks (Berryhill et al., 2010; Tanoue et al., 2013). Thus, we suspect that participants' engagement was the most important factor in this particular experiment.

The University Institutional Review Board approved all protocols. Volunteers completed the Adult ADHD Self-Report Scale (ASRS—v1.1; Kessler et al., 2005). This short screen was developed to identify ADHD symptomology in adults and has been validated in adult (Adler et al., 2006) and college-aged (Fuller-Killgore et al., 2013) populations. Scores were derived from the six-screener questions in Part A of the ASRS, the most predictive of ADHD. Questions probed the frequency with which participants forgot appointments, completed tasks, or felt distracted Responses were converted from verbal labels (e.g., “never,” “sometimes,” “rarely,” “often,” “very often”) to point values (1–5). Questions 1–3 required doubling the point value when answers of “sometimes,” “often,” or “very often” were recorded; questions 4–6 required doubling the point value when answers “often” or “very often” were recorded. Four or more answers requiring doubling meet the criteria for “high likelihood” of ADHD (Kessler et al., 2005). To identify participants we recruited participants high (ASRS scores 39–58) and low (ASRS scores <19) in ADHD symptomology. This high ADHD group met the criteria of high likelihood of ADHD. Thirty-six right-handed, normal, neurotypical participants were subsequently enrolled (age 18–37, 14 male). Participants were screened to ensure they were not taking medications that modulate the excitability of the brain (e.g., stimulants). Participants completed anodal, cathodal and sham tDCS sessions, in counterbalanced order across 3 separate days with a minimum washout period of 24 h. Current was administered using a commercial stimulator (Eldith Magstim GMbH, Ilmenau, Germany). To target the right pre-SMA, we placed the electrode 2 cm to the right of FZ (10–20 system) and the reference electrode was placed on the contralateral cheek. Participants received 10 min of 1.5 mA tDCS during anodal and cathodal sessions and during sham stimulation current was ramped up and down for the initial and final 20 s of the period. Participants completed practice trials of each task during stimulation and after the electrodes were removed, they began the experimental trials of the tasks described below.

Performance on each task was subjected to a 2 ADHD group (low, high symptomology) × 3 tDCS condition (sham, anodal, cathodal) repeated measures ANOVA. Performance accuracy on the Go/No-Go task revealed no significant main effect of group [F_{(1, 17)} = 1.01, p = 0.33], no main effect of tDCS [F_{(2, 34)} = 0.996, p = 0.38], and no interaction [F_{(2, 34)} = 1.145, p = 0.33]; see Figure 3A. The reaction time data were also analyzed and they followed the same pattern of null results {group [F_{(1, 17)} = 0.246, p = 0.626]; tDCS [F_{(2, 34)} = 0.060, p = 0.942], interaction [F_{(2, 34)} = 1.643, p = 0.208]}. Four participants were eliminated from the OSPAN and 2-back WM tasks because they pressed the same button for all responses. Performance on the OSPAN task followed the same pattern, with no significant main effects of group [F_{(1, 15)} = 1.635, p = 0.22], or tDCS [F_{(2, 30)} = 0.151, p = 0.860], and no interaction [F_{(2, 30)} = 0.213, p = 0.810]; see Figure 3B. This pattern was also true for performance accuracy on the spatial 2-back WM task. There were no main effects of group [F_{(1, 15)} = 0.015, p = 0.904] or tDCS [F_{(2, 30)} = 0.466, p = 0.632] and no significant interaction [F_{(2, 30)} = 0.750, p = 0.481]; see Figure 3C.

Here, we failed to replicate an anodal tDCS benefit on a response inhibition task performance after stimulation to the pre-SMA (Hsu et al., 2011), and to extend it into those with high/low ADHD symptomology using the go/no-go task. Others have had luck with response inhibition and tDCS, particularly in suppressing responses (reviewed in Juan and Muggleton, 2012). We may have tapped a particularly heterogeneous population and used an insufficiently difficult response inhibition task that did not allow us to discriminate between easy and challenging trials. We may have selected a stimulation site that was too lateralized and an intensity that did not effectively reach the pre-SMA. These issues may have clouded tDCS effects. Future studies should include a more difficult response inhibition task, as participants approached ceiling. In the Hsu et al. (2011) paper, the stop-signal task they employed required more finely titrating the timing of the stop signal to ensure that the task difficulty was tailored to the participant. Finally, because the participants completed a self-assessment rather than complete a clinical interview and ADHD diagnosis, a different pattern of results may emerge in a clinically diagnosed ADHD population.

In the two WM tasks, the OSPAN and spatial 2-back tasks, likewise, we did not observe any significant modulation after tDCS. These later results were expected, but difficult to interpret as site-specific tDCS effects given the null findings in the primary task of interest, the go/no-go task.

The considerations we noted above point us toward an emerging research question. Why do group differences matter? The underlying mechanism of tDCS remains unclear, although long-term neuroplasticity is indicated (Rosenkranz et al., 2000; Nitsche et al., 2003). Some pharmacological work in humans indicates that blocking sodium or calcium channels prevents longer-lasting effects of anodal tDCS and antagonizing NMDA receptors prevents longer-lasting effects of anodal or cathodal tDCS, at least in motor cortex (Nitsche et al., 2003). Furthermore, at a larger scale, the way current flows through a particular individual's brain certainly varies and has tremendous implications for applied and experimental use of tDCS (Bikson et al., 2013). Incorporating other variables, including the structural and functional connectivity of stimulated networks and their temporal dynamics will also improve experimental success when applying tDCS. We suspect that the answers to these questions will also clarify why some groups benefit more than others and why homogeneous groups provide more consistent data. For an individualized approach to tDCS application, the underlying molecular mechanism must be fully clarified and this in turn will advance our understanding at the network level. At this point, another way to counter these differences is simply through brute power and by running sufficient participants.

The role of task difficulty is important. As noted in the introduction, when a participant knows the answer definitively tDCS is unlikely to have an effect. This means tasks that are too easy do not show any effect of tDCS. It is only in the challenging, borderline cases where tDCS serves as a tipping factor. As we age, we encounter more of these near-threshold cases and this population may be the one to benefit most from tDCS-linked cognitive interventions. In particular, in the realm of WM, small improvements can benefit quality of life and are worth the effort to gain them.

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.