Behavioral response conflict arises in the color-word Stroop task and triggers the cognitive control network. Midfrontal theta-band oscillations correlate with adaptive control mechanisms during and after conflict resolution. In order to prove causality, in two experiments, we applied transcranial alternating current stimulation (tACS) at 6 Hz to the dorsolateral prefrontal cortex (DLPFC) during Stroop task performance. Sham stimulation served as a control in both experiments; 9.7 Hz tACS served as a nonharmonic alpha band control in the second experiment. We employed generalized linear mixed models for analysis of behavioral data. Accuracy remained unchanged by any type of active stimulation. Over both experiments, the Stroop effect (response time difference between congruent and incongruent trials) was reduced by 6 Hz stimulation as compared to sham, mainly in trials without prior conflict adaptation. Alpha tACS did not modify the Stroop effect. Theta tACS can both reduce the Stroop effect and modulate adaptive mechanisms of the cognitive control network, suggesting midfrontal theta oscillations as causally involved in cognitive control.
In the face of conflicting information, human beings are capable of adjusting their executive control to resolve conflict and perform the appropriate behavior.
During this process, the cognitive control network first detects conflict, then selects and monitors behaviors for attaining a goal. Multiple brain regions jointly exercise inhibitory control when task demands are high to override stimulus-driven behavior. Generally, cognitive control is measured by performance in conflict tasks, like the Stroop task, in which conflicting task-irrelevant information has to be suppressed for responding correctly [
In the Stroop color-word task (SCWT), participants indicate the ink color of a color-word while not responding to its semantic meaning. Responses are faster when the semantic meaning and ink color match (congruent, low-conflict, e.g., “Blue” in blue ink) compared to a mismatch (incongruent, high-conflict, e.g., “Blue” in red ink). This response time difference is a function of the congruence and named after its discoverer
Previous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) studies found that several brain regions are activated during the Stroop task, including the dorsal Anterior Cingulate Cortex (dACC), the dorsolateral prefrontal cortex (DLPFC), and the posterior parietal cortex (PPC) [
Responses in incongruent trials, which are preceded by incongruent trials (iI), are faster than in incongruent trials, which are preceded by congruent trials (cI). Conversely, responses in cI are slower than in cC [
Transcranial alternating current stimulation (tACS) allows us to causally infer function of oscillatory networks [
In this study, we have chosen tACS with a frequency of 6 Hz based on previous electrophysiological results. Generally, these electrophysiological studies are in line with and corroborate the findings of neuroimaging studies in the Stroop task. The dACC has been shown to be the generator of mediofrontal negativity in the theta (4–8 Hz) range marked by a stronger negative potential around 450 ms in the incongruent condition [
Furthermore, dACC and left DLPFC couple in the theta phase between conflict detection and resolution [
Similarly to our approach, a previous work has also targeted the left DLPFC with a theta-range tACS during decision-making requiring cognitive control [
To investigate the efficacy of tACS on conflict processing, we have used the drift diffusion model for conflict tasks (DMC). The DMC is a newly developed extension of the classical drift diffusion model (DDM) [
Generally, cognitive processing in conflict tasks is studied by behavioral measures like response time and accuracy, which are influenced by a trade-off between speed and accuracy of response. Cognitive models allow decomposing the response time and accuracy into several parameters underlying the decision process. The DDM models the cognitive processes underlying two-alternative forced choice tasks by assuming that participants start to accumulate for either alternative over the time of the trial. The accumulation of evidence begins at the start of the trial, and as soon as it reaches a certain threshold for one alternative, a decision is being made. Due to noisy sensory input, the accumulation is a stochastic process which occasionally results in error trials. Aside from the decision process, the time needed for nondecisional processes is also accounted for.
In DMC, evidence accumulation is the sum of a controlled process (naming of color) and another, automatic process (recognition of semantic meaning). These processes are summed, either leading to (slower) faster responses in (in)congruent trials. The distribution in time of the automatic process is a gamma density function, peaking early during the trial and decaying afterwards. Therefore, the DMC is well suited as it accounts for both the RT distributions and accuracies of conflict tasks as Stroop, Simon, or Eriksen flanker task [
We aimed to externally modulate theta power in the left DLPFC and to thereby causally change the function of the cognitive control network. We employed tACS in the theta range (6 Hz) with a high-definition (HD) electrode montage over the left DLPFC in two experiments, in order to entrain the cortical control network [
As mentioned above, we employed GLMM and the newly developed DMC to analyse the effects of tACS on response times and accuracy and also the interaction with the congruency effect [
We hypothesized that the cortical control network can be exogenously entrained (via the left DLFPC) by theta tACS. This would result in increased theta power during and after conflict resolution. With longer phase-coupling between the dACC and the left DLPFC, all trials would show activation patterns similar to those in the incongruent trial. This would induce higher cognitive control for the next trial, comparable to the iC or iI conditions of the Gratton effect. Therefore, we predicted a reduced Stroop effect in the active condition compared to the controls. We expected trials which are preceded by a congruent trial to be more strongly affected by stimulation (reduced Stroop effect) as normally they show no conflict adaptation mediated by theta phase-coupling. Consequently, in DMC, the influence of the automatic process on the decision-making should be reduced.
The participants consisted of 22 healthy, right-handed, and native German-speaking adult volunteers, who have normal or corrected-to-normal vision and gave their written informed consent to join the study. They were measured in two experimental groups. The first group consisted of 10 participants (8 females, mean age:
The experiments were double-blinded, placebo-controlled, and executed in a within-subject design. Experiment 1 (
Participants performed a Stroop color-word task (SCWT) [
Each session started with a minimum of 50 practice trials (termination rule: 18 of the last 20 trials correct), and the following main phase consisted of 300 congruent and incongruent trials in a randomized order. The length of a trial was 1.5 s; the mean interstimulus interval lasted 0.5 s (Chi-squared distribution, range 0.3 s–0.7 s) during which a gray fixation cross (hue) was shown. The participants were instructed to respond as quickly and accurately as possible. The SCWT lasted for 20 minutes (Figure
The color-word Stroop task. After practice trials, the participants performed 600 trials within one session while being stimulated by tACS. They responded as quickly and accurately as possible during the 1.5 s of a single trial. Congruent and incongruent trials appeared equally often and were subcategorized depending on the preceding trial.
Stimulation was delivered by a CE-certified neuroConn multichannel stimulator (neuroConn GmbH, Ilmenau, Germany) throughout the main experimental phase [
The high-definition (HD) montage centered over AF3 according to the international 10-10 EEG system with four return electrodes. The return electrodes were positioned over F5, F2, Fp2, and AF7 as in earlier studies targeting the DLPFC [
The HD tACS montage for stimulation of the left dorsolateral prefrontal cortex and the modelled electric field strength. (a) The central electrode of the HD montage is centered over AF3. Two pairs of return electrodes form equilateral triangles of 6 cm side length with the central electrode. The distance between both pairs is 10 cm. The return electrodes are located over F5, Fp2, F2, and AF7. (b) The electric field strength is maximal (0.35 mV/mm) over the left prefrontal cortex including the DLPFC. The graphics and electric field strength modelling are derived from SimNIBS 2.0.1.
Sinusoidal tACS of 1 mA (peak-to-baseline) intensity and 6 Hz frequency was applied throughout the 20 min duration of the WCST in the active stimulation condition (including 10 s ramp-up and ramp-down periods). Similarly, 9.7 Hz was used as an active control stimulation in the alpha range in the second experiment. Sham stimulation was limited to 30 s (including 10 s ramp-up and ramp-down periods) during the beginning and the end of the SCWT in order to blind the participants while not influencing task performance. The impedances were kept below 15 k
The DMC fitting and the organization of behavioral datasets were done in Python. All statistical testing were conducted in R [
Generalized linear mixed models (GLMMs) are increasingly utilized to analyse complex research designs [
Parsimonious GLMMs were run on nontransformed RTs of correctly answered trials using an identity-linked Inverse Gaussian distribution as recommended by Lo and Andrews [
The random effects in the final parsimonious model included intercepts for participants and word-color, with slopes of current trial congruency for word-color and within-participant slopes of current trial congruency and stimulation. The random effects account for variance in the data which arises as, for instance, every participant balances the speed-accuracy trade-off differently, which leads to individual response time and accuracy distributions. The categorical two-level fixed effects stimulation (sham, 6 Hz), congruencies of current and preceding trials (both: congruent, incongruent), was sum-coded numerically for the first experiment. In the second experiment, the stimulation (sham, 6 Hz, 9.7 Hz) was also sum-coded numerically, allowing the effect of the active stimulations to be individually compared to sham. Additionally, we could analyse the interaction of the stimulation with the current trial congruency (Stroop effect) and with the current and preceding trial congruencies (Gratton effect). These factorial predictors were contrast-coded to extract their main effects and their interactions on the grand means of reaction time and accuracy. We report the
DMC assumes that the total response time is the sum of the duration of the decision process (
Model fitting was done on individual participants per session (and individual “original” datasets in the recovery study) as described in [ Plausible starting values from the pilot study were drawn for all parameters from a uniform distribution Minimization of The first two steps were repeated 30 times. Computations were done in parallel with the Göttingen Campus High-Performance Computing Centre as each repetition had a run time of around 30 h
We further analysed the parameters which best fit the data as indicated by the
Arousal levels in the Stroop task correlate with better performance in congruent trials and worse performance in incongruent trials [
Overall accuracy was 94.9% (SD 2.3%), and mean RTs were 624.3 ms (SD 54 ms). Within sham stimulation, accuracy was lower and mean RTs prolonged for incongruent trials (94.6%, SD 2.4%; 652.3 ms, SD 59.1 ms) compared to congruent trials (95.9%, SD 2.8%; 604.2 ms, SD 50.6 ms). Equally, in the active stimulation condition, incongruent trials (93.7%, 2.7%; 638.8 ms, 61.9 ms) were more erroneous and slower than congruent ones (95.3%, SD 2.9%; 602.2 ms, 59.1 ms) (see Table
Descriptive statistics of both experiments. For both experiments, the difference in behavior between congruent and incongruent trials is broken down per stimulation condition. Mean values are reported with their respective standard deviation.
Accuracy (%) | Response times (ms) | |
---|---|---|
Sham | ||
Congruent | ||
Incongruent | ||
6 Hz | ||
Congruent | ||
Incongruent | ||
Sham | ||
Congruent | ||
Incongruent | ||
6 Hz tACS | ||
Congruent | ||
Incongruent | ||
9.7 Hz tACS | ||
Congruent | ||
Incongruent |
To assess the effect of the stimulation condition, we were interested in the main effect of the stimulation, its interaction with the congruency of the current trial and its effect on the Gratton effect (i.e., the interaction between congruency of the current and the previous trials). Additionally, we expected an interaction between congruency of the current trial and the stimulation conditions when the preceding trial was either congruent or incongruent. Two generalized linear mixed models were conducted: one for error rates including all trials and the other for the nontransformed response times excluding all error and posterror trials (10.4% of all trials; see Table
Statistical analysis of the first experiment. The results of the GLMMs are shown for both accuracy and response time data of the first experiment. Additionally, the response times were divided according to the congruency of the previous trial in additional model runs. For every factor, the mean values and standard errors of each factor level are reported. The results of the statistical testing of the difference between these mean values are also reported as
Estimate ( |
|||
---|---|---|---|
CCT | 2.801 | <0.01 | |
Congruent | |||
Incongruent | |||
CPT | 1.491 | 0.13 | |
Congruent | |||
Incongruent | |||
Stimulation | 1.875 | 0.06 | |
Sham | |||
6 Hz | |||
CCT (Stroop effect) | 4.37 | <0.001 | |
Congruent | |||
Incongruent | |||
CPT | 1.06 | 0.28 | |
Congruent | |||
Incongruent | |||
Stimulation | 0.49 | 0.61 | |
Sham | |||
6 Hz | |||
CCT x stimulation | 1.847 | 0.06 | |
Stroop effect (sham) | |||
Stroop effect (6 Hz) | |||
CCT x CPT | 3.026 | <0.01 | |
Stroop effect (CPT: congruent) | |||
Stroop effect (CPT: incongruent) | |||
CCT x CPT x stimulation | 1.828 | 0.06 | |
CCT x stimulation | 2.87 | <0.01 | |
Stroop effect (sham) | |||
Stroop effect (6 Hz) | |||
CCT x stimulation | 0.01 | 0.98 | |
Stroop effect (sham) | |||
Stroop effect (6 Hz) |
SE: standard error.
For accuracy, significant main effects exist for the congruency (congruent, incongruent) of the current trial (CCT;
The analysis of the response times revealed significant main effects for CCT (
Effect of stimulation on response time. The response times for congruent and incongruent trials and the time difference between these (Stroop effect) are plotted for each stimulation condition individually for all trials in the left panels. In the middle panels, only data of trials which were preceded by a congruent trial are displayed; in the right panels, only for trials preceded by an incongruent trial. Experiment 1 (first row): CCT (size of Stroop effect) and stimulation interact significantly in trials preceded by congruent trials. Experiment 2 (middle row): the interaction CCT x stimulation is significant for all trials but not for the data subsets differentiated by the preceding trial. The active control stimulation in the alpha range did not change the interaction between stimulation and CCT. Combined dataset of both experiments (last row): the interaction between stimulation and CCT is significant across all trials. The significant interaction for trials preceded by congruent trials underlies the effect across all trials. All data is plotted including the 95% confidence interval.
Overall accuracy was 97.6% (SD 2.1%), and mean RTs were 578.1 ms (SD 57 ms).
In sham stimulation, accuracy was lower and mean RTs prolonged for incongruent trials (97.8%, SD 1.5%; 604.5 ms, SD 84.9 ms) compared to congruent trials (98.2%, SD 1.7%; 569.5 ms, SD 68.1 ms). Equally, in the 6 Hz condition, incongruent values are 96.9%, SD 3.4%; 583.4 ms, SD 68.6 ms and congruent values are 97.6%, SD 2.5%; 554.9 ms, SD 60.5 ms, and in the active control condition, incongruent values are 97.0%, SD 2.3%; 595.9 ms, SD 69.0 ms and congruent values are 98.1%, SD 1.9%; 560.3 ms, SD 49.2 ms (see Table
To assess the effect of the stimulation condition, we were interested in the main effects of the two active conditions (stimulation: 6 Hz; control: 9.7 Hz). The two interactions were individually compared to sham stimulation. We further investigated their interaction with the congruency of the current trial and their effect on the Gratton effect (i.e., the interaction between congruency of the current and the previous trials). Additionally, we expected a change in their interaction between congruency of the current trial and the stimulation conditions when the preceding trial was either congruent or incongruent. Two generalized linear mixed models were conducted: one for error rates including all trials and the other for the nontransformed response times excluding all error and posterror trials (4.7% of all trials; see Table
Statistical analysis of Experiment 2. The results of the GLMMs are shown for both accuracy and response time data of the second experiment. For every factor, the mean values and standard errors of each factor level are reported. The results of the statistical testing of the difference between these mean values are also reported as
Estimate ( |
|||
---|---|---|---|
CCT | 2.952 | <0.01 | |
Congruent | |||
Incongruent | |||
CPT | 0.441 | 0.65 | |
Congruent | |||
Incongruent | |||
Stimulation | |||
Sham | |||
6 Hz (vs. sham) | 0.579 | 0.56 | |
6 Hz | |||
9.7 Hz (vs. sham) | 0.43 | 0.66 | |
9.7 Hz | |||
CCT (Stroop effect) | 3.12 | 0.001 | |
Congruent | |||
Incongruent | |||
CPT | 2.28 | 0.02 | |
Congruent | |||
Incongruent | |||
Stimulation | |||
Sham | |||
6 Hz (vs. sham) | 0.78 | 0.43 | |
6 Hz | |||
9.7 Hz (vs. sham) | 0.06 | 0.95 | |
9.7 Hz | |||
CCT x stimulation | |||
Stroop effect (sham) | |||
Stroop effect (6 Hz vs. sham) | 2.11 | 0.03 | |
Stroop effect (6 Hz) | |||
Stroop effect (9.7 Hz vs. sham) | 1.44 | 0.14 | |
Stroop effect (9.7 Hz) | |||
CCT x CPT | 3.48 | <0.001 | |
Stroop effect (CPT: congruent) | |||
Stroop effect (CPT: incongruent) | |||
CCT x CPT x stimulation | |||
CCT x CPT (6 Hz vs. sham) | 0.35 | 0.72 | |
CCT x CPT (9.7 Hz vs. sham) | 0.87 | 0.38 |
SE: standard error.
For accuracy, significant main effects existed for CCT (
The analysis of the response times revealed significant main effects for CCT (
The size of the Stroop effect depends on whether participants were stimulated with 6 Hz tACS (Stroop effect:
The active control did not significantly interact with either CCT (
The response time datasets of Experiments 1 and 2 were combined post hoc and reanalysed to increase statistical power (see Table
Statistical analysis of combined dataset. The results of the GLMMs are shown for both accuracy and response time data of the combined dataset of both experiments. Additionally, the response times were divided according to the congruency of the previous trial in additional model runs. For every factor, the mean values and standard errors of each factor level are reported. The results of the statistical testing of the difference between these mean values are also reported as
Estimate ( |
|||
---|---|---|---|
CCT (Stroop effect) | 3.98 | <0.001 | |
Congruent | |||
Incongruent | |||
CPT | 1.17 | 0.23 | |
Congruent | |||
Incongruent | |||
Stimulation | 1.25 | 0.20 | |
Sham | |||
6 Hz | |||
CCT x stimulation | 2.37 | 0.01 | |
Stroop effect (sham) | |||
Stroop effect (6 Hz) | |||
CCT x CPT | 4.40 | <0.001 | |
Stroop effect (CPT: congruent) | |||
Stroop effect (CPT: incongruent) | |||
CCT x CPT x stimulation | |||
CCT x stimulation | 2.65 | <0.01 | |
Stroop effect (sham) | |||
Stroop effect (6 Hz) | |||
CCT x stimulation | 0.71 | 0.47 | |
Stroop effect (sham) | |||
Stroop effect (6 Hz) |
SE: standard error.
In Experiment 1, Fisher-Pitman permutation tests investigated statistical differences in the DMC parameter (
In the first experiment, the Wilcoxon signed rank test indicated no significant differences between stimulation conditions in mean arousal (
Response conflict increases midfrontal theta dynamics between dACC and the DLPFC [
This is in line with a reduction of congruency effect in the Simon task by theta tACS targeted towards the dACC [
EEG recordings in healthy subjects and intracranial recording in epilepsy patients suggest a causal role for the neural oscillatory connection between dACC and DLPFC [
We chose the DLPFC as a target since its location at the brain surface allows a more reliable stimulation. Stimulation of the dACC, in which due to its deep location is a more difficult target, has been done however [
It has to be noted that in the first experiment the reduced congruency effect was clearly driven by trials which were preceded by congruent trials. It fits very well in our second hypothesis that stronger cognitive control is exerted when tACS increases the normally low DLPFC activity. However, the Stroop effect was reduced for all trials in the second experiment, not only those preceded by congruent trials. Therefore, the first hypothesis that theta-range tACS reduces the Stroop effect is fulfilled. While 6 Hz tACS reduced the Stroop effect in both experiments, it is a partial replication as different subsets of data are affected. In the combined dataset of the studies, both effects survive the joint analysis, showing a general effect of DLPFC on Stroop effect across all participants. Both experiments were designed equally except for the active control condition. Participants acted as their own control by participating in all sessions of an experiment, which cancels out possible difference in performance between the experiments. Therefore, the pooling of the data of both experiments is statistically valid and allows the interpretation of trends underlying both datasets. Inconsistent effects of tACS have been reported before in internal replications [
We confirmed the validity of the used DMC by recovering simulated data (see Supplementary Material), replicating an earlier study [
The stimulation frequency of 6 Hz chosen as oscillatory power in narrow-band theta (6 Hz–7 Hz) in the left-frontal region correlates with reaction time in conflict adaptation [
Our choice of active control frequency in the second experiment fell on a nonharmonic frequency in the alpha range. In previous studies, alpha power decreased after conflict trials as it marks higher arousal [
Alpha tACS showed a trend towards a reduced congruency effect during the Simon task in an earlier study [
For future studies, the stimulation of the DLPFC in a broad gamma range would be a promising target as DLPFC gamma power after response predicted response times in subsequent trials. Also, theta-gamma cross-frequency stimulation paradigms promise stronger abolishment of the Stroop effect as they effectively change functionality of distant brain regions which exhibited this type of cross-frequency behavior [
It is of note that the theta stimulation to the DLPFC could be equally effective if limited to the time after response. Therefore, the effect of stimulation on adaptation could be isolated while not interfering with conflict detection and resolution. A transfer of the stimulation paradigm to different conflict tasks could show causally if the cognitive control network’s physiology is equal in all these tasks.
The Stroop task is a frequently applied neurophysiological test to study neural mechanisms of inhibitory control and its dysfunction [
This is the first study stimulating the DLPFC by theta tACS. We demonstrate that the cognitive control network can also be influenced by stimulation targeting the DLPFC. We were able to reduce the Stroop effect in a subset of trials over both experiments. The equalization of response times in congruent and incongruent trials suggests that postconflict adaptation was changed. We propose the hypothesis that theta stimulation of the DLPFC is effective in changing preparatory mechanisms after conflict resolution. The key questions to be clarified are whether (a) gamma tACS leads to more reduction of the Stroop effect, (b) theta stimulation applied only after a conflict resolution is equally effective, and (c) the results are generalizable to other conflict tasks.
The behavioral data used to support the findings of this study are available from the corresponding author upon request.
The funders had no role in study design, collection and analysis of data, decision to publish, or drafting the manuscript.
The authors declare that there is no conflict of interest regarding the publication of this paper.
We acknowledge support by the German Research Foundation and the Open Access Publication Funds of the Göttingen University. The study was funded by the University Medical Center Göttingen, Göttingen, Germany.
The supplementary material is a recovery study for one of the used models, which shows that the model works and can be applied to the data. Additionally, it includes one table specifying the age and gender of the participants and one table showing the quality of sleep and the arousal parameters.