With the recent surge in use of transcranial direct current stimulation (tDCS) has come a growing uncertainty about the reliability of this neuromodulatory technique. TDCS, a form of non-invasive electrical brain stimulation, hinges on a simple premise: hypo-polarization of a cortical area should increase neuronal excitability, while hyper-polarization should induce the opposite effect. Within the motor domain, this rationale has been largely supported at the neurophysiological level: when primary motor areas are hypo-polarized by positive current administered during anodal stimulation (A-tDCS), motor-evoked potentials (MEPs) recorded from peripheral muscles tend to increase in magnitude, indicating a boost in cortical excitability. In contrast, hyper-polarization of these areas via negative current administered during cathodal stimulation (C-tDCS) tends to diminish the amplitude of MEPs, indicating cortical inhibition (Nitsche and Paulus, 2000; see also, Fregni et al., 2006; Furubayashi et al., 2008; Jefferson et al., 2009; Stagg et al., 2009). Extended to the cognitive domain, it was thus assumed that improvement of a cognitive function could be achieved by anodal stimulation of the substrate underlying that function. Conversely, cathodal stimulation of the underlying substrate should lead to decrements in that function.

Several recent findings have called into question this basic premise—on which the design and interpretation of all tDCS studies hinge—thereby creating a wave of confusion. Of most pressing importance for the future of brain stimulation research, the effects of tDCS are demonstrably sensitive to seemingly subtle variations in task, stimulation parameters, and characteristics of the individuals being tested. In one well-known example from the motor domain, Batsikadze et al. (2013) charted the effects of stimulation intensity (1 mA vs. 2 mA) for both A- and C-tDCS of primary motor cortex. A-tDCS at the highest stimulation intensity produced an increase in MEPs while A-tDCS at 1 mA produced no significant change relative to baseline amplitude. Unexpectedly, C-tDCS at 2 mA also induced an excitatory effect (compared to the suppressed MEPs observed for C-tDCS at the level of 1 mA). Thus, the excitatory and inhibitory effects of tDCS may depend not only on the polarity of current, but also on the intensity level of stimulation. In accordance with these findings, a number of meta-analyses and review articles offer in-depth discussions of other sources of variation associated with tDCS both within and outside the motor domain (Nitsche et al., 2008; Jacobson et al., 2012; Filmer et al., 2014; Horvath et al., 2014; Li et al., 2015; Price et al., 2015). These factors include: electrode position and size (e.g., Bikson et al., 2010), timing of task relative to stimulation period (e.g., Nozari et al., 2014), duration of stimulation (e.g., Nitsche and Paulus, 2001), cognitive demand involved in task (e.g., Antal et al., 2007; Gill et al., 2015), skull thickness and subcutaneous fat content (Datta et al., 2012), and a genetic polymorphism associated with prefrontal dopamine (Plewnia et al., 2013; Nieratschker et al., 2015).

In acknowledging the challenges faced by the field of tDCS research, our aim is not to encourage the abandonment of this tool but rather to stress the value of more comprehensive experimental approach. We propose that a thorough exploration of the stimulation parameter space is essential to facilitating cross-study comparison and ultimately working toward a reliable model of tDCS effects. These steps are especially crucial for tDCS investigations in the cognitive domain, for which, relative to the motor domain, there exists greater variability in experimental design and potentially greater complexity in the neural systems engaged at task. Below, we highlight the extent of methodological variation within one sub-field of cognitive tDCS research (cognitive control), thereby motivating our own experimental approach.

Broadly construed, cognitive control underlies our capacity to interact flexibly with our surroundings in a goal-directed manner. More precisely, this term refers to processes such as the selection and maintenance of relevant information, shifting between tasks, and the inhibition of prepotent responses (Miller and Cohen, 2001). A combination of lesion and functional neuroimaging studies have implicated a network of cortical and subcortical brain regions as the seat of these essential functions. In particular, a host of tDCS studies have stimulated prefrontal cortex in order to affect performance on tasks of working memory, set-shifting and inhibitory control. Experimenters have examined these processes via different stimulation sites (e.g., F7- contralateral mastoid placement: Nozari et al., 2014; Fz- left cheek: Hsu et al., 2011; F3- right supraorbital (RSO): Ohn et al., 2008; the crossing point between T3-Fz and F7-Cz-RSO: Cattaneo et al., 2011; the crossing point between T4-Fz and F8-Cz: Ditye et al., 2012), and at different current intensities (e.g., 1 mA: Fregni et al., 2005; 1.5 mA: Nozari et al., 2014; 2.0 mA: Vanderhasselt et al., 2013). On top of these differences, a recent review of polarity effects on executive function (Jacobson et al., 2012) found that only half of the tDCS studies surveyed examined both cathodal and anodal effects. In some cases, experimenters did not include a sham stimulation condition as a control (e.g., Ditye et al., 2012). While studies generally show a boost in cognitive control during A-tDCS administered to the prefrontal cortex, it is unclear whether an equal and opposite effect would be obtained during C-tDCS under otherwise identical experimental conditions. Moreover, experimenters rarely probe dose-dependent changes in stimulation intensity (but see, Iyer et al., 2005; Hoy et al., 2013; Horvath et al., 2016), leaving open the question of whether behavioral effects might change, such as by flipping directions or diminishing across intensity levels (as in Batsikadze et al., 2013).

Against this backdrop, in which we have an abundance of data but great variation in how those data were obtained, the field is thus faced with a host of interpretation issues. To illustrate: Hoy et al. (2013) reported that anodal stimulation of left prefrontal cortex significantly improved participants’ speed on a simple work memory task (the “2-back”), but not on a more difficult version of the task (the “3-back”). They did not test the effects of cathodal stimulation. In contrast, Fregni et al. (2005) reported improved accuracy but not speed on 3-back task performed during anodal stimulation. They also tested the effects of cathodal stimulation, but found no significant effect. Finally, Zaehle et al. (2011) reported polarity-dependent changes in accuracy on a 2-back task (accuracy was superior for A-tDCS relative to C-tDCS), but polarity-independent changes in RT (reaction time was equally facilitated for A-and C-tDCS relative to sham). Though these three studies are, in fact, more closely related than is typical (i.e., they made use of the F3 electrode montage and comparable current intensity), their remaining dissimilarities still make it challenging to pinpoint the source of non-overlapping results. Can they be traced to important differences in task structure, timing of stimulation, duration of stimulation, current polarity, or some combination of these factors? In an era when replicability is increasingly a focus in psychological research (Open Science Collaboration, 2015; Anderson et al., 2016), careful consideration of these features is critical.

The present study represents a crucial step toward disentangling two basic but still crudely understood stimulation parameters within the domain of cognitive research: current polarity and stimulation intensity. While we propose the systematic variation of stimulation parameters constitutes an important contribution to the field, we fully acknowledge the limitations of this approach. Specifically, between-subject manipulations, which are useful in minimizing stimulation timing and task familiarity effects, may require prohibitively large sample sizes. Indeed, this limitation becomes even more essential when considering that the behavioral effects of tDCS may be quite small (Minarik et al., 2016).

Bearing in mind this trade-off between a comprehensive approach and lowered statistical power, we examine here performance on the Eriksen Flanker, a cognitive control task that taps into the capacity for selective attention and response inhibition (Eriksen and Eriksen, 1974), and one whose rapid pace enables us to collect a rich data set (over 550 trials). As this task has been associated with activation in prefrontal cortex (e.g., Casey et al., 2000; Ullsperger and von Cramon, 2001; Bunge et al., 2002), we selected an electrode montage thought to target prefrontal cortex in the left hemisphere: F3- RSO. In particular, we chose this montage due to its common use in studies of tDCS during various cognitive control tasks (Fregni et al., 2005; Ohn et al., 2008; Hoy et al., 2013). While the Flanker task has also been employed in a handful of other tDCS studies (Weiss and Lavidor, 2012; Nozari et al., 2014; Zmigrod et al., 2016), our design enables us to examine both the effects of current polarity (i.e., A-tDCS and C-tDCS relative to sham stimulation) and dose-dependent stimulation (i.e., to ask whether cognitive control increases or decreases monotonically from 1, to 1.5, to 2 mA). Specifically, we test the dual hypotheses that A-tDCS will improve performance on the Eriksen Flanker in a dose-dependent manner while C-tDCS will worsen performance in this way. To this end, we extract an index of cognitive control by comparing, for each participant, reaction times for trials that require response inhibition relative to those that do not strongly engage this process. By charting the parameter space of tDCS during this task, we hope to open the door to further methodological research while also serving as a launching pad for exciting theoretical questions about the behavioral consequences of suppressing and exciting cognitive control capacities.

Here, we systematically probed the effects of current polarity and stimulation intensity on participants’ ability to perform a task of inhibitory cognitive control. Results indicated nearly at-ceiling levels of accuracy on the Eriksen Flanker across stimulation parameters. Statistical modeling of RT data clearly showed an effect of current polarity for the anodal condition relative to sham (i.e., an overall greater speed up of RTs). Most compellingly, we have demonstrated that A-tDCS to left prefrontal cortex (via the F3-RSO electrode montage) facilitated the deployment of cognitive control resources when applied concurrently with task. This evidence was clear from the reduced difference in RTs between congruent and incongruent trials (i.e., a smaller interference effect).

Nonetheless, it is important to note that we observed significant differences in the physical sensations experienced during anodal and sham stimulation. This finding runs contrary to the bulk of the tDCS literature, which overwhelmingly reports no difference in physical perception between stimulation conditions. While it is possible that differences in physical sensation, not cortical excitability, account for observed effects in behavior, including this variable in our statistical models did not on the whole dramatically alter our results. We suggest here that participants experiencing A-tDCS may have reported increased sensitivity to stimulation because they were devoting fewer cognitive resources to perform the task required. A related possibility is that overall enhanced attentional capacity during A-tDCS may have induced learners to attend more to their physical environment. Indeed, challenging cognitive control tasks have been shown to attenuate pain intensity (Bantick et al., 2002; Valet et al., 2004; Buhle and Wager, 2010). Consonant with our findings, pain reduction in one study was shown in low working memory capacity but not high working memory capacity individuals, suggesting that attenuation of pain scales with individual differences in cognitive control (Nakae et al., 2013).

While it is best to be cautious in interpreting null effects, it is also useful to review the experimental manipulations that showed no effect on cognitive control. First, we observed no significant interaction between stimulation intensity (1–2 mA) and trial type (congruent vs. incongruent). In other words, cognitive control capacities were not influenced in a dose-dependent manner. A post hoc analysis did reveal a dose-dependent increase in overall RTs for the cathodal condition, but this trend did not apply to the RT difference between trial types. Most strikingly, we found no evidence that cathodal stimulation differed reliably from sham, either in significantly improving or impeding Flanker performance. While such results suggest weaker reliability of cathodal stimulation, we stress that the relatively small sample size of the current study precludes us from making definitive claims about its efficacy (Minarik et al., 2016).

Nonetheless, previous studies contextualize this null finding. Specifically, Zmigrod et al. (2016) demonstrated a significantly increased Flanker interference effect (reduced cognitive control) during cathodal stimulation of right prefrontal cortex (electrode montage: F4-RSO; current intensity: 2 mA), but no such effect on Simon task performance (demonstrating specificity of right PFC to stimulus-stimulus rather than stimulus–response conflict). Similar to the present findings, Nozari et al. (2014) found no evidence from either accuracy or RT measures that C-tDCS to left prefrontal cortex mediated the strength of the Flanker interference effect (electrode montage: F7-right mastoid; current intensity: 1.5 mA). Thus, it appears that cathodal stimulation of right but not left DLPFC stimulation may impede Flanker performance, but precisely why this behavioral modulation is specific to these particular stimulation conditions (C-tDCS to the right hemisphere) remains an open question. One possibility is that the Flanker task more strongly recruits right relative to left prefrontal cortex (Hazeltine et al., 2000). Indeed, recent evidence suggests that patients with right prefrontal damage showed greater Flanker interference effects than those with similar damage in the left hemisphere (Geddes et al., 2014). However, other neuroimaging studies point to more diffuse, bilateral prefrontal involvement during the Flanker (Ullsperger and von Cramon, 2001; Bunge et al., 2002; Durston et al., 2003), suggesting that the left hemisphere does assume a processing burden during tasks of inhibitory control.

In general, our findings concur with the observation that, within the broader cognitive domain, the consequences of cathodal stimulation are more varied compared to anodal stimulation (Jacobson et al., 2012). It is worth stressing that this general pattern might be traced to the observation that high-level cognitive tasks are likely to engage diffuse swathes of the brain. For example (and as indicated above), fMRI recordings of brain activation during the Flanker task have implicated widespread frontal and posterior parietal regions that are often bilaterally distributed (e.g., the middle frontal gyrus, the precentral gyrus, inferior frontal gyrus, precuneus, superior parietal lobule, etc., Zhu et al., 2010). Thus, while a stimulation-induced increase in activity in one of these areas might be enough to improve behavioral performance, suppression of an area might have unpredictable effects (likely due to compensatory recruitment of other regions). Second, whereas both A-tDCS and C-tDCS involve NMDA-receptor mediated effects, anodal stimulation effects also require sodium channel function in motor cortex (Liebetanz et al., 2002). The extent to which receptor mediated effects generalize to cognition remain to be seen. Genetically mediated individual differences in response to C-tDCS may offer a third explanation for the unreliable effects of C-tDCS. Nieratschker et al. (2015) showed that a genetic polymorphism associated with prefrontal dopamine levels predicted individual differences in behavioral response to C-tDCS. Specifically, cathodal stimulation was found to reduce cognitive control abilities in COMT 166 Val–Val homozygotes but not in Met-allele carriers (Nieratschker et al., 2015). To be clear, Plewnia et al. (2013) also showed that Met–Met homozygotes under anodal stimulation were impaired in their set-shifting abilities; however, a smaller percentage of the population are Met-allele carriers (Auton et al., 2015). Pinpointing the underlying sources of varied responses to C-tDCS, including relevant genetic determinants, is an exciting and imperative area of future research.

In sum, we have begun to disentangle the contribution of two key stimulation parameters: current polarity and intensity. Using a behavioral task that demanded the suppression of prepotent responses, we offer convincing evidence that the former, current polarity, is a robust predictor of inhibitory control abilities but that the latter, current intensity, is not. Intriguingly, this effect was specific to anodal stimulation of left prefrontal cortex, which induced a boost in cognitive control. Thus, our initial hypotheses were only partially confirmed: current polarity indeed influenced performance on a task of cognitive control. However, contrary to our expectations, this effect was specific to the anodal stimulation condition and RT interference effects did not unfold in a dose-dependent manner.

While our efforts to probe the stimulation parameter space represent an important step forward, we stress that the present approach was by no means exhaustive. Nonetheless, by providing a template for how the stimulation parameter space might be mapped, we open up the possibility for future research to build substantially on the findings reported here. Particularly in light of the null effects observed during cathodal stimulation, one clear next step is to examine whether these results would be overturned with a much larger sample size (i.e., perhaps C-tDCS of prefrontal cortex has a weaker, but significant behavioral impact). One might also ask whether current polarity is influenced by stimulation timing or precise electrode placement. Through increased understanding of the impact of parameter selection, a more consistent picture may emerge across cognitive tasks, and only then will truly generalizable inferences be forthcoming.

This study was carried out in accordance with the recommendations of the University of Pennsylvania Institutional Review Board (IRB) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the IRB.

EK, ZB, RH, and ST-S formulated the experiment. EK and ZB implemented the experiment, supervised data collection and performed data analysis. RH, JM, NT, and ST-S provided input on stimulus materials and data analysis. EK and ZB wrote the original draft of the paper, with revisions and additional content contributions from RH, JM, NT, and ST-S.

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.