Direct current stimulation modulates prefrontal cell activity and behaviour without inducing seizure-like firing

Abstract

Transcranial direct current stimulation (tDCS) has garnered significant interest for its potential to enhance cognitive functions and as a therapeutic intervention in various cognitive disorders. However, the clinical application of tDCS has been hampered by significant variability in its cognitive outcomes. Furthermore, the widespread use of tDCS has raised concerns regarding its safety and efficacy, particularly in light of our limited understanding of its underlying neural mechanisms at the cellular level. We still do not know ‘where’, ‘when’ and ‘how’ tDCS modulates information encoding by neurons, in order to lead to the observed changes in cognitive functions. Without elucidating these fundamental unknowns, the root causes of its outcome variability and long-term safety remain elusive, challenging the effective application of tDCS in clinical settings.

Addressing this gap, our study investigates the effects of tDCS, applied over the dorsolateral prefrontal cortex, on cognitive abilities and individual neuron activity in macaque monkeys performing cognitive tasks. Like humans performing a delayed match-to-sample task, monkeys exhibited practice-related slowing in their responses (within-session behavioural adaptation). Concurrently, there were practice-related changes in simultaneously recorded activity of prefrontal neurons (within-session neuronal adaptation).

Anodal tDCS attenuated both these behavioural and neuronal adaptations when compared with sham stimulation. Furthermore, tDCS abolished the correlation between response time of monkeys and neuronal firing rate. At a single-cell level, we also found that following tDCS, neuronal firing rate was more likely to exhibit task-specific modulation than after sham stimulation. These tDCS-induced changes in both behaviour and neuronal activity persisted even after the end of tDCS stimulation. Importantly, multiple applications of tDCS did not alter burst-like firing rates of individual neurons when compared with sham stimulation. This suggests that tDCS modulates neural activity without enhancing susceptibility to epileptiform activity, confirming a potential for safe use in clinical settings.

Our research contributes unprecedented insights into the ‘where’, ‘when’ and ‘how’ of tDCS effects on neuronal activity and cognitive functions by showing that modulation of the behaviour of monkeys by the tDCS of the prefrontal cortex is accompanied by alterations in prefrontal cortical cell activity (‘where’) during distinct trial phases (‘when’). Importantly, tDCS led to task-specific and state-dependent alterations in prefrontal cell activities (‘how’). Our findings suggest a significant shift from the view that the effects of tDCS are merely attributable to polarity-specific shifts in cortical excitability and instead propose a more complex mechanism of action for tDCS that encompasses various aspects of cortical neuronal activity without increasing burst-like epileptiform susceptibility.

single-cell recording, working memory, delayed match-to-sample task, transcranial direct current stimulation, primate behaviour

See Hogeveen (https://doi.org/10.1093/brain/awae307) for a scientific commentary on this article.

Introduction

Transcranial direct current stimulation (tDCS) is a non-invasive brain stimulation technique that has shown potential in modulating cognitive functions.^1-5 This technique involves application of a low-intensity direct current to the scalp, which presumably passes through underlying tissues (skin, muscle and dura) to reach neurons to modulate their functions.⁶ It has been suggested that tDCS could enhance cognitive abilities and aid in the rehabilitation of individuals with neuropsychiatric disorders or brain injuries, where executive control is often impaired.^7-10 It is acknowledged that the behavioural effects of tDCS extend beyond the duration of stimulation,^11,12 suggesting its potential for inducing long-lasting neuroplastic changes in neural circuits.^6,8,9,12-16 However, the reported outcomes of tDCS are not consistent. Some studies have reported improvements^17-19 in cognitive tasks, whereas others have found no effect^20,21 or even a decline^22,23 in performance. The variability of these results and lack of mechanistic evidence for its short- and long-term effects on neuronal circuitries make it challenging to standardize and incorporate tDCS into clinical practice.⁹ Furthermore, tDCS has received regulatory approval for the treatment of major depressive disorders,²⁴ underscoring its therapeutic potential. Nonetheless, the safety of repeated tDCS sessions remains under scrutiny, raising concerns about its long-term use.²⁵ Additionally, although tDCS is under investigation for treating a range of conditions, including depression, bipolar disorder, schizophrenia and obsessive–compulsive disorder, the variability in treatment outcomes across these disorders further complicates its application in a clinical setting.²⁶

With the scarcity in studies directly examining its influence on neuronal activity during cognitive tasks, a major uncertainty lies in grasping if (and, if so, how) tDCS modulates neuronal activity.^6,8,27,28 Hence, at a fundamental level, we still do not understand the link between the application of tDCS and the modulation of information encoding by neurons, which can lead to the observed cognitive-behavioural effects. However, there remains a significant debate regarding whether the electrical current from tDCS is strong enough to reach and modulate neuronal activity,²⁹ whether such effects persist beyond the stimulation and whether such effects can be traced to specific cortical or subcortical regions.^30,31

During the completion of a cognitive task, learning and practice effects are commonly observed, which are presumably mediated by adaptive neural mechanisms that modify behaviour to optimize task performance.^16,18 Accompanying such behavioural modulation, functional imaging has revealed learning- and practice-related changes in activation patterns in various brain regions.^32,33 Brain stimulation effects might intersect mechanistically with learning processes, as both processes potentially induce plastic changes in cortical and subcortical brain structures¹⁶ mediated by alterations of N-methyl-D-aspartate³⁴ and brain-derived neurotrophic factor³⁵ to lead to their observable cognitive effects.

At a neuronal level, the cognitive effects of tDCS could stem from different mechanisms. Studies in anaesthetized rodents³⁶ and monkeys³⁷ performing non-cognitive tasks have suggested that tDCS (both anodal and cathodal) results in generalized modulation of the spontaneous firing rates and active numbers of units in the cerebral cortex.^36,37 Other studies, based on analyses of EEG and haemodynamic responses, have indicated that these activation changes could be specific to different task-related events.³⁸ Another possibility is that tDCS influences the variability in neuronal responses. This has not yet been explored in experiments involving tDCS but is suggested by work indicating that transcranial magnetic stimulation and intracerebral electrical micro-stimulation could modulate neuronal firing rate variability.^39,40 Neuronal firing rate variability is intrinsically linked to cognitive states,⁴¹ executive control fluctuations⁴² and task difficulty^41,43 and leads to correlated firing across neuronal ensembles;^41,44,45 therefore, modulation of such variability might provide a mechanistic basis for the cognitive-behavioural effects of tDCS.

Owing to the challenges in directly recording single-neuron activity in humans, an animal model is necessary for exploring the concomitant neuronal and behavioural effects of tDCS. Non-human primate models, such as macaque monkeys, are suitable, given their capacity to perform hundreds of trials of well-established cognitive tasks, in addition to similarities to humans in terms of prefrontal cortex structure.^46,47 Here, we examined the influence of tDCS in the context of a delayed match-to-sample (DMS) task (Fig. 1A and C), commonly used to assess executive control processes, such as working memory. The prefrontal cortex plays a vital role in performing the DMS task, with damage in this region inducing performance impairments in humans and monkeys.^48,49 To bridge the gap between cellular-level modulations and cognitive-behavioural outcomes of tDCS, we investigated the extent to which tDCS applied over the dorsolateral prefrontal cortex (dlPFC) (Fig. 1B and D) influences the executive control of behaviour and concurrently modulates activity of individual prefrontal neurons (both firing rate and variability) in two awake macaques performing the DMS (Fig. 1). The aim of this study, therefore, was to understand the neuronal underpinnings of tDCS, providing an evidence-based mechanistic framework for its behavioural effects.

Figure 1

Experimental design and paradigm. (A and C) Task and experimental structure. (B and D) Recording and transcranial direct current stimulation (tDCS) apparatus. (A) Delayed match-to-sample (DMS) task. Trial sequence: initiated with a grey circle (requiring switch press and hold; 350 ms), followed by a white fixation point on a black screen (requiring 500 ms of fixation), then a sample item (1000 ms), a delay (1500 ms) and, finally, a selection from three test items within 4000 ms for a reward (∼1 ml water). Errors prompted a pink annulus and an additional inter-trial delay (2500 ms). Shape-DMS necessitated shape-matching, and Colour-DMS necessitated colour-matching. Shape- and Colour-DMS trials were randomly intermingled and not cued. Touching the matching item (within 4000 ms) was considered as a correct response, and a reward was given. (B) Within a surgically implanted circular chamber over the left dorsolateral prefrontal cortex (dlPFC), an insulated recording electrode and a tDCS active electrode were submerged in a conductive agarose solution to enable recording of neuronal activity and application of tDCS, respectively. The tDCS reference electrode (4 cm × 4 cm rubber electrode) was placed over the trapezius muscle. Created with BioRender.com. (C) The experimental protocol begins with cell isolation, a 45 min stabilization, followed by a practice block (90% success rate needed from 10 consecutive trials), then a 30 min prestimulation period (∼120 trials). tDCS was then applied to the dlPFC for 10 min, alternating daily between active or sham, ending with a post-stimulation period (≥30 min and 120 trials, contingent on task performance and sustained cell isolation). Comparisons are made between pre- and post-stimulation periods for behaviour and neuronal activity. (D) Coronal (anteroposterior +33) MRI of the monkey brain shows the principal sulcus (PS), corresponding to the dlPFC in monkeys. V indicates the vertical distance from the interaural line. The green line represents the electrode track that enabled recording from dlPFC.

Materials and methods

Participants

The study involved two adult female Japanese macaques (Macaca fuscata), weighing 6.7 and 7.3 kg, obtained from the National BioResource Project of the MEXT, Japan. Following daily sessions, they were housed in standard conditions with a regulated 12 h–12 h light–dark cycle. All experimental protocols were reviewed and approved by the Research Ethics Committee at RIKEN Center for Brain Science and were consistent with the ARRIVE guidelines.

Apparatus

Monkeys completed the DMS while seated in a primate chair, customized to allow access to a 38-cm touchscreen (ELO Entuitive) and a centrally placed response switch. The front panel of the chair was adapted specifically for this study to facilitate interaction with the screen and switch. Visual stimuli were displayed against a black backdrop on the touchscreen, and correct responses were reinforced with a water reward delivered through a computer-controlled solenoid valve system.

Behavioural task

Monkeys performed a computerized DMS task, matching a sample to one of three test items based on either colour (Colour-DMS) or shape (Shape-DMS) (Fig. 1A and Supplementary material). Data acquisition was at millisecond resolution, controlled by CORTEX (National Institute of Mental Health), and required eye and head fixation for single-cell recording.

Data were collected across 50 sessions per animal. Each session began with a practice block (9 correct responses from 10 trials) before proceeding to data collection. The first data block (prestimulation period) lasted 30 min (∼120 trials), followed by 10 min of tDCS applied to the dlPFC, then a second data block (post-stimulation period) of ≥30 min (∼120+ trials) (Fig. 1C). Both monkeys performed the task with their right hand.

Transcranial direct current stimulation

Single-cell recordings were conducted while the monkeys performed the DMS task. After the prestimulation period, tDCS was applied over the left dlPFC for 10 min while the animals continued the task, followed by the post-stimulation period. Anodal tDCS was administered through a rubber electrode in a conductive agarose solution within a chamber (18 mm in diameter) over the left dlPFC (Fig. 1B and D), with a large rubber electrode over the trapezius muscle serving as the reference. A direct current of 0.1 mA was applied for 10 min using PlexStim (Plexon), with a 30 s linear fade-in and fade-out period. The recording and stimulation electrodes were positioned at the start of each session and remained in place throughout. The sham condition used a transient current to replicate the sensation of stimulation. Active and sham stimulations alternated across sessions. The dlPFC location was determined by three-dimesional (3D) MRI, and the tDCS electrode was positioned within an agarose-filled chamber over the dlPFC. This method ensures precise and uniform current distribution across the dural surface. For additional detail, see the Supplementary material.

Rationale for stimulation parameters

The choice of a 0.1 mA current for 10 min was based on safety and efficacy. Unlike typical human tDCS studies using higher intensities (1–2 mA for ≥20 min), our direct application of the electrode to the dural surface reduces electrical shunting across the skin, scalp and skull, ensuring that more current reaches the cortical surface. Past studies support the efficacy of lower intensities (0.3–0.6 mA) in producing significant neurophysiological and behavioural effects.^50-52 We validated our stimulation intensity using a finite element method simulation, following extensive prior work.^53,54 The simulation demonstrated that the peak electrical field was 0.09 V/m (Fig. 2), which is well within the range applied in past tDCS literature.^50-52 For additional detail on the finite element method approach, see the Supplementary material.

Figure 2

Electric field distribution of stimulation. Top view (A), left lateral view (B), right lateral view (C) and posterior view (D) of the electric field (EF) distribution induced by anodal transcranial direct current stimulation (tDCS) applied over the dorsolateral prefrontal cortex. The colour scale represents the EF intensity (in volts per metre), ranging from 0 (blue) to 0.09 V/m (red), highlighting regions of higher and lower stimulation intensity. The 0.1 mA stimulation in our study had a peak electric field of 0.09 V/m.

Recording

After 15 months of training for the final DMS task, including head and eye fixation, both animals underwent surgical implantation of a cylindrical recording chamber over the left dlPFC (Fig. 1B and D). The chamber placement was guided by the principal and arcuate sulci, as determined by structural MRI in a stereotaxic frame (Fig. 1D). Recordings were made from the lateral surface of the dlPFC, both dorsal and ventral to the principal sulcus. Unit isolation used Plexon software without prescreening for specific neuronal properties (such as firing rate or waveform shape). Recording boundaries were set within 3 mm ventrodorsally from the sulcus centre and up to 3 mm anterior to its posterior end. Insulated tungsten electrodes (Alpha Omega Engineering) were advanced through the dura by an oil hydraulic pulse motor microdrive micromanipulator (MO-951; Narishige) to record extracellular action potentials in each daily session. A stabilization period of 45 min followed each cell isolation to ensure the stability of neuronal activity and mitigate any transient effects resulting from electrode positioning. Action potentials from each single cell were isolated online using template matching and recorded using a Plexon system (Plexon Neural Recording Data Acquisition System; Plexon). The recorded data were then processed in Offline Sorter, with spike sorting confirmed by the 3D Valley Seeking method.

Control study

To validate the relevance of our findings in monkeys to human behaviour, a control study was conducted to investigate whether humans had comparable performance in the DMS (see ‘Control Study’ in the Supplementary material).

Data analyses

Behavioural

Response time (RT) was measured from test item presentation to switch release and normalized within each session following convention.^18,48,55-57 Accuracy was calculated as the percentage of correct responses within a 4000 ms window. RT outliers, defined as values 3 standard deviations (SD) from the mean, were excluded. The influence of tDCS on RT and accuracy was analysed using repeated-measures ANOVAs, with factors including Stimulation (active/sham), PrePost (prestimulation/post-stimulation), Dimension (Colour-DMS/Shape-DMS) and Monkey (Monkey 1/Monkey 2). Sphericity was assessed with Mauchly’s test, and Greenhouse–Geisser corrections were applied when necessary. Pairwise comparisons were conducted using two-tailed t-tests with Bonferroni corrections, and the partial eta-squared (η_p²) statistic quantified the variance attributable to significant effects.

Behavioural variation: coefficient of variation

The coefficient of RT variation (CoV), defined as the standard deviation of RT divided by the mean RT for each condition in correct trials, assessed variability in performance.^58,59 Repeated-measures ANOVA (Stimulation × PrePost × Dimension × Monkey) analysed the influence of tDCS and Dimension on CoV.

Neuronal activity

Neuronal activity was analysed in three epochs per trial (Supplementary Fig. 1): ‘Sample’, ‘Decision’ and ‘Pre-Reward’. The inter-trial interval (ITI) epoch was used to examine baseline activity. No systematic differences in neuronal activity were observed based on recording location, hence all recorded cells were pooled.

Differences in activity between active and sham transcranial direct current stimulation

Neuronal activity was recorded continuously during prestimulation, 10 min of active or sham stimulation, and post-stimulation periods. The percentage difference in neuronal firing rate (ΔF)³⁷ between pre- and post-stimulation periods was calculated to assess tDCS effects:

$$\mathrm{\Delta }F=100{\displaystyle \frac{F_{\mathrm{Post}}-F_{\mathrm{Pre}}}{\max (F_{\mathrm{Pre}},F_{\mathrm{Post}})}}$$

(1)

A ΔF of zero indicates no change in firing rate, positive ΔF indicates an increase and negative ΔF a reduction.

Neuronal firing rate variability

Neuronal firing rate variability was characterized using the Fano factor, defined as variance divided by the mean for each epoch.⁴¹ Repeated-measures ANOVA (Stimulation × PrePost × Dimension × Monkey) assessed the influence of tDCS on variability during the Decision and Pre-Reward epochs, and (Stimulation × PrePost × Monkey) for the Sample epoch.

Burst-like firing analyses

Kolmogorov–Smirnov tests compared spike rate distributions pre- and post-stimulation for active and sham conditions. Paired t-tests assessed within-group changes, and independent t-tests examined between-group differences following stimulation.

For additional details on data analyses, see the Supplementary material.

Results

Behaviour

Humans and monkeys exhibited comparable behaviour in the DMS

In parallel studies (Supplementary material), we examined the behaviour of humans performing a similar DMS task (Supplementary Fig. 2 and Supplementary material), in which they performed the DMS in pre- and post-testing periods (separated by a rest period). Multifactorial ANOVAs revealed that humans exhibited within-session RT slowing [F(1,32) = 9.09; P < 0.01; η_p² = 0.22; Supplementary Fig. 2A], which appeared as slower responses in the post-testing period. Humans showed significant dimensional bias (behavioural advantage) in RT [F(1,32) = 97.65; P < 0.001; η_p² = 0.75; Supplementary Fig. 2B], in accuracy [F(1,32) = 48.35; P < 0.001; η_p² = 0.60] and in behavioural variability [indexed by the CoV, calculated as SD divided by mean RT; F(1,32) = 15.38; P < 0.001; η_p² = 0.33] towards colour-matching. In humans, there was no observable within-session change in accuracy [F(1,32) = 1.53; P = 0.23; η_p² = 0.05] or behavioural variability [CoV; F(1,32) = 2.63; P = 0.12; η_p² = 0.08].

We examined whether the RT of the monkeys was modulated in the DMS, independent of active stimulation (when sham stimulation was applied), by applying a multifactorial ANOVA [PrePost (prestimulation/post-stimulation) × Dimension (colour/shape) × Monkey (Monkey 1/Monkey 2)] to the normalized RT data in correct trials in the sham stimulation condition. A significant main effect of PrePost [F(1,21) = 5.14; P = 0.03; η_p² = 0.20] was found, indicating within-session response slowing, as was seen in humans (Supplementary Fig. 2). A significant main effect of Dimension factor [F(1,21) = 70.19; P < 0.001; η_p² = 0.77] confirmed a shorter RT in Shape-DMS than in Colour-DMS trials, indicating a dimensional bias towards shape-matching in monkeys. Although the dimensional bias in monkeys was in the opposite direction to humans, this aligns with our previously reported dimensional bias in humans and monkeys performing other cognitive tasks.^42,55,57

Response time and coefficient of variation were not correlated

We assessed whether RT and CoV were correlated aspects of the task. To examine the relationship between RT and CoV in the sham condition, we used Pearson correlation coefficients. The analysis revealed that the correlation between RT and CoV in the sham condition was not statistically significant [r(40) = 0.28, P = 0.07]. This indicates that although RT and CoV followed a similar pattern of modulation, they were uncorrelated aspects of the task.

Anodal stimulation attenuated within-session changes in monkey response time

To examine how tDCS over the dlPFC influences behaviour and single-neuron activity in the DMS, multifactorial repeated-measures ANOVAs were applied. In these ANOVAs, a significant main effect of Stimulation (active or sham) indicates overall performance differences between the two stimulation conditions but does not necessarily indicate a significant modulation by active tDCS. Instead, an interaction between Stimulation and PrePost (pre- versus post-stimulation periods) for a given measure indicates modulation by the active tDCS, showing that changes from pre- to post-stimulation were dependent on the stimulation type. In all analyses, unless explicitly stated, there were no main effects or interactions of the Monkey factor, indicating that any significant observation occurred uniformly in both monkeys.

To assess whether tDCS application influenced the RT of monkeys in the DMS, we applied a multifactorial ANOVA [Stimulation (active/sham) × PrePost (prestimulation/post-stimulation) × Dimension (colour/shape) × Monkey (Monkey 1/Monkey 2)]. The main effects of PrePost [F(1,19) = 4.73; P = 0.04; η_p² = 0.20; Supplementary Fig. 2C] and Dimension [F(1,19) = 130.6; P < 0.001; η_p² = 0.87; Supplementary Fig. 2D] remained significant. Importantly, there was a significant interaction between Stimulation and PrePost factors [F(1,19) = 5.4; P = 0.03; η_p²= 0.22]; the magnitude of within-session RT slowing was attenuated by anodal stimulation (Fig. 3A and C). A planned comparison of RT (paired t-test) in the prestimulation period between active and sham stimulation confirmed that RT was not significantly different between stimulation conditions before stimulation [t(21) = 0.025, P = 0.98]. Post hoc analysis (paired t-tests) confirmed that following stimulation in the sham stimulation condition, RT increased significantly [t(21) = 2.85, P < 0.01], whereas in the active condition, it did not change significantly [t(19) = 1.00, P = 0.32].

Figure 3

Transcranial direct current stimulation (tDCS) modulated within-session slowing and response time (RT) variability. (A and C) tDCS attenuated within-session RT slowing. (B and D) tDCS and dimension interactively influenced behavioural variability. (A) Comparing prestimulation (Pre) and post-stimulation (Post) normalized RT revealed that tDCS attenuated within-session slowing. (B) RT variability is measured by the coefficient of RT variation (CoV) and is shown for Colour and Shape in Pre and Post for each stimulation condition (sham and active). (C) The bar graph (top left) depicts the magnitude of within-session change in RT for sham and active conditions; the scatterplot (below) displays the distribution of Pre and Post RT with individual data points (session means). Colour-coded mean and standard error are included for sham (grey circles) and active (yellow triangles) conditions. The histogram (top right), scaled to the scatterplot, shows the magnitude of difference between Pre and Post for sham and active conditions. Active tDCS stimulation attenuated the magnitude of within-session slowing. (D) Similar graphical conventions to B, but here showing within-session CoV changes, with active tDCS differentially reducing RT variability in Colour-DMS but increasing RT variability in Shape-DMS. In all panels, statistically significant differences are indicated with asterisks (P < 0.05). All bar graphs show the mean ± SEM.

In contrast, accuracy was not modulated by anodal stimulation, indicated by a non-significant interaction between Stimulation and PrePost factors [F(1,19) = 0.06; P = 0.81]. However, like RT, the main effect of PrePost for accuracy was significant [F(1,19) = 5.49; P = 0.03; η_p² = 0.22]; accuracy decreased in the post-stimulation period (Supplementary Fig. 3A). The main effect of Dimension was also significant [F(1,21) = 33.48; P < 0.001; η_p² = 0.62], with higher accuracy in Shape-DMS than in Colour-DMS trials (Supplementary Fig. 3B and Supplementary material). There were also corresponding differences in neural firing rate between Colour- and Shape-DMS trials (Supplementary Fig. 4).

Examining the modulation of response time variability by anodal stimulation

To characterize the behavioural variability in the DMS, independent of stimulation, we applied a multifactorial ANOVA (PrePost × Dimension × Monkey) to the CoV in the sham stimulation condition. There was a significant main effect of Dimension [F(1,21) = 20.73; P < 0.001; η_p² = 0.50]; the CoV was lower in Shape-DMS trials than in Colour-DMS trials, indicating that behavioural variability was attenuated in trials that required matching in the preferred (shape) dimension. The main effect of PrePost was not significant [F(1,21) = 0.19; P = 0.67], indicating that behavioural variability was consistent between the prestimulation and post-stimulation periods.

Next, to assess whether application of tDCS influenced RT variability in the DMS, we applied the same ANOVA with the addition of the Stimulation factor. There was a significant main effect of Dimension [F(1,19) = 43.13; P < 0.001; η_p² = 0.69]. Moreover, there was a significant interaction between Dimension and PrePost factors [F(1,19) = 11.73; P < 0.01; η_p² = 0.38]; the CoV was overall lower in Shape-DMS than Colour-DMS trials (Supplementary Fig. 5A); however, during a single testing session, Colour-DMS variability decreased, whereas Shape-DMS variability increased (Supplementary Fig. 5B and C). Post hoc analysis (paired t-test) confirmed the significant difference in CoV change between colour and shape dimensions [t(21) = −2.56, P = 0.02; Supplementary Fig. 5C]. The main effect of PrePost was not significant [F(1,19) = 0.54; P = 0.47]. There was a significant interaction between Stimulation, PrePost and Dimension [F(1,19) = 5.08; P = 0.03; η_p² = 0.21]. However, as demonstrated in Fig. 3B, the significant interaction might have arisen owing to the high CoV in colour in the prestimulation period. Consequently, we cannot necessarily conclude that this change was induced by tDCS.

Neuronal activity

We recorded 167 neurons (107 from Monkey 1 and 60 from Monkey 2) in the dlPFC across sessions with a minimum of 120 trials in each pre- and post-stimulation period. Neuronal activity was analysed in three epochs in each trial (Supplementary Fig. 1): ‘Sample’, ‘Decision’ and ‘Pre-Reward’. The ITI epoch was also used in some analyses to examine baseline activity.

Anodal stimulation attenuated within-session changes in population-level neuronal activity

The cognitive-behavioural effects of tDCS application might be mediated by the modulation of the baseline firing rate of neurons.^36,37 To assess this possibility, we analysed stimulation sessions, alternating daily between Active and Sham, for baseline neuronal activity changes. This was done by contrasting normalized firing rates in the ITI (the time between trials), before and after stimulation. In the Sham stimulation condition, there was a significant [t(178) = −6.13, P < 0.001] practice-related increase in firing rates from the pre- to post-stimulation period during the ITI epoch. However, there was no significant [t(152) = 0.59, P = 0.56] change in firing rate from the pre- to the post-stimulation period in the active stimulation condition, indicating that anodal stimulation attenuated within-session changes in baseline neuronal activity. A direct comparison between sham and active conditions indicated that anodal stimulation significantly [t(165) = 3.60, P < 0.001] reduced the within-session neuronal activity changes relative to the sham condition.

To evaluate the effects of tDCS on neuronal activity, we calculated the ΔF, which represents the percentage change in firing rate from pre- to post-stimulation periods [Equation (1)³⁷]. A ΔF of zero indicates no change in neuronal activity between pre- and post-stimulation periods. Mann–Whitney U-tests, corrected for multiple comparisons, were applied to ΔF for each epoch, including all recorded cells. There was a significant difference in the change in neuronal firing rate (indexed by ΔF) between Active and Sham stimulation conditions in both the Sample (U = 2766, z = −2.24, P = 0.025, r = −0.17; Fig. 4A and C) and Decision (U = 2843, z = −2.00, P = 0.046, r = −0.15; Fig. 4B and D) epochs. In the active tDCS condition, neuronal activity was stable during the Sample and Decision epochs (ΔF close to zero), whereas in the sham condition, there was an increase in activity post-stimulation (ΔF shifted from zero) (Fig. 4). This indicates that anodal stimulation attenuated the within-session changes in neuronal activity compared with sham. In the Pre-Reward epoch, there was no significant difference in firing rate changes between active and sham conditions (U = 2940, z = −1.69, P = 0.09), indicating that the influence of active stimulation on neuronal activity was specific to the Sample and Decision epochs.

Figure 4

Active transcranial direct current stimulation (tDCS) maintained consistent neuronal activity. ΔF represents the percentage change in neuronal firing rate from before (Pre) to after (Post) stimulation, where zero indicates no change in firing rate. (A and C) Sample epoch. (B and D) Decision epoch. (A) The distribution of ΔF is shown for each stimulation condition (sham and active) during the Sample epoch. (B) The distribution of ΔF is shown for each stimulation condition (sham and active) during the Decision epoch. (C) The conventions for the scatterplot and histogram are the same as those in Fig. 3C, but here contrasting firing rates before (Pre) and after (Post) stimulation for both stimulation conditions in the Sample epoch. Means and standard errors are shown for sham (grey circles) and active (yellow triangles) conditions. Active tDCS maintains ΔF around zero, differing significantly from the dispersion observed with sham stimulation, indicating more consistent neuronal activity during the Sample epoch with active tDCS. (D) The conventions for the scatterplot and histogram are the same as those in Fig. 3C, but here contrasting firing rates before (Pre) and after (Post) stimulation for both stimulation conditions in the Decision epoch. Means and standard errors are shown for sham (grey circles) and active (yellow triangles) conditions. Active tDCS maintains ΔF around zero, differing significantly from the dispersion observed with sham stimulation, indicating more consistent neuronal activity during the Decision phase with active tDCS.

Transcranial direct current stimulation attenuated within-session changes in neuronal firing rate variability

Cortical neurons exhibit pronounced variability in their firing rates, even when representing identical events or movements.^41,60 Notably, this variability is correlated across neuronal ensembles^41,44,45 and varies with task difficulty.^41,43 Considering that other neuromodulatory interventions, such as transcranial magnetic stimulation and direct electrical micro-stimulation, can modulate this variability,^39,40 it is plausible that tDCS might also alter this variability, thereby providing a neural basis for its observed cognitive-behavioural effects.

Initially, to characterize the neuronal variability in the DMS independent of stimulation, we applied a multifactorial ANOVA (PrePost × Monkey) to the Fano factor (calculated as the variance of firing rate divided by the mean firing rate⁴¹) in the sham stimulation condition. There was a significant practice-related change in Fano factor; the Fano factor increased significantly post-stimulation compared with prestimulation in the Sample epoch [F(1,88) = 4.27; P = 0.04; η_p² = 0.05], but this effect was not present in the Decision [F(1,88) = 1.55; P = 0.22] or Pre-Reward [F(1,88) = 1.35; P = 0.25] epochs.

Next, to assess whether application of tDCS influenced the variability of neuronal responses in the DMS, we applied the same ANOVA with the addition of the Stimulation factor. The Dimension factor was also included in the Decision and Pre-Reward epochs, where dimensional information was relevant. In both the Sample [F(1,76) = 4.44; P = 0.04; η_p² = 0.06; Fig. 5A and C] and Decision [F(1,76) = 5.15; P = 0.03; η_p² = 0.06; Fig. 5B and D] epochs, there was a significant interaction between Stimulation and PrePost factors. In both epochs, although there was an increase in the neuronal firing rate variability from pre- to post-stimulation with sham stimulation, the variability was decreased from pre- to post-stimulation with active stimulation. This indicates that active stimulation led to a decrease in neuronal firing rate variability. In both epochs, main effects for Dimension, Stimulation and PrePost, in addition to all interactions between factors, were not significant (all P > 0.22).

Figure 5

Active transcranial direct current stimulation (tDCS) decreased the variability of neuronal activities. The Fano factor represents the variability of neuronal responses. (A and C) Sample epoch. (B and D) Decision epoch. (A) The Fano factor is shown for pre- and post-stimulation for sham and active conditions during the Sample epoch. (B) The Fano factor is shown for pre- and post-stimulation for sham and active conditions during the Decision epoch. (C) The conventions for the bar graph, scatterplot and histogram are the same as those in Fig. 3C, but here contrasting the Fano factor before (Pre) and after (Post) stimulation for both sham and active stimulation conditions in the Sample epoch. Means and standard errors are shown for sham (grey circles) and active (yellow triangles) conditions. Active tDCS decreased the variability of neuronal responses compared with sham stimulation during the Sample epoch. (D) The conventions for the bar graph, scatterplot and histogram are the same as those in Fig. 3C, but here contrasting the Fano factor before (Pre) and after (Post) stimulation for both sham and active stimulation conditions in the Decision epoch. Means and standard errors are shown for sham (grey circles) and active (yellow triangles) conditions. Active tDCS decreased the variability of neuronal responses compared with sham stimulation during the Decision epoch.

In contrast, in the Pre-Reward epoch, there was no significant interaction between Stimulation and PrePost factors [F(1,76) = 2.52; P = 0.12], indicating that the variability of the neuronal responses in the Pre-Reward epoch was not modulated by the stimulation condition. The main effects of Dimension, Stimulation and PrePost and all interactions between factors were not significant (all P > 0.12).

Transcranial direct current stimulation increased task-specific neuronal activities

Having observed that active stimulation attenuated within-session learning and practice-related changes in population-level neuronal firing rate (Fig. 4) and neuronal activity variability (Fig. 5), we examined whether tDCS modulated the task-related activity of individual prefrontal neurons. Previous research using functional imaging and EEG has demonstrated that the application of tDCS can induce task-related behavioural optimizations¹⁶ and modulations of their associated electrophysiological correlates.³⁸

Modulation of cellular activity was defined as firing rate changes [a significant (P < 0.05) main effect of PrePost in a one-way ANOVA] from the pre- to post-stimulation period. Next, we defined task-specific modulation as cells that exhibited significant modulation during task-related epochs (Sample, Decision or Pre-Reward) and not during the ITI epoch. A χ² test revealed a significant difference in the proportion of cells showing task-specific modulation between Active and Sham conditions. Cells showing significant pre- to post-stimulation changes in task-related epochs (Sample, Decision or Pre-Reward) but not in the ITI epoch were more prevalent in the active condition (24 of 81 cells) than in the sham condition (13 of 86 cells) [χ²(1,167) = 5.10; P = 0.024]. These results suggest that active tDCS increased the likelihood of task-specific neuronal firing modulation compared with sham. This is exemplified by the activity patterns of two dlPFC cells depicted in Fig. 6 for both pre- and post-stimulation in active (Fig. 6A) and sham conditions (Fig. 6B). These figures also highlight that the tDCS-induced activity modulation was specific to a particular trial epoch and, presumably, to particular involved cognitive processes.

Figure 6

Task-specific modulation of single dorsolateral prefrontal cortex (dlPFC) cell activity by active transcranial direct current stimulation (tDCS). Neuronal firing is presented for both prestimulation (left) and post-stimulation (right) for active and sham conditions for all correct trials. Rastergrams (top) display neuronal activity (aligned to fixation onset, green line) in correct trials. Each row corresponds to a trial (first trial at bottom), and each point represents an action potential, while the peristimulus time histograms (bottom) show firing rates in 100 ms bins, also aligned to fixation. The yellow highlight indicates the Decision period. The P-value indicates the significance of the difference in activity between pre- and post-stimulation periods. (A) A dlPFC cell that exhibited significant task-specific modulation in the Decision epoch induced by active tDCS application (P = 0.001). (B) A dlPFC cell that did not exhibit significant task-specific modulation in the Decision epoch following sham tDCS application (P = 0.47).

Transcranial direct current stimulation abolished the correlation between neuronal activity and response time

Previous studies have shown a correlation between dlPFC activity and RT while humans and monkeys perform executive control tasks.^42,61 Furthermore, temporary disruption of dlPFC activity by transcranial magnetic stimulation leads to RT slowing.⁶² Therefore, to investigate how tDCS affects the firing rate–behaviour correlation, we assessed: (i) whether a correlation existed between firing rate and RT; and (ii) whether application of tDCS (either sham or active) influenced this correlation. We analysed the correlation between neuronal firing rate and RT in pre- and post-stimulation periods, for both sham and active conditions in each epoch (pooled RT and neuronal firing rate for all trials from all cells). In the sham condition, a positive correlation between RT and neuronal firing rate was observed in the Decision epoch both before [prestimulation; r(10 885) = 0.040, P < 0.001] and after [post-stimulation; r(12 170) = 0.030, P < 0.001] sham stimulation. In the active stimulation condition, the correlation between RT and neuronal firing rate was also observed in the Decision epoch before active stimulation [prestimulation; r(9305) = 0.03, P < 0.001]. However, this correlation was abolished after active stimulation [post-stimulation; r(12 796) = 0.007, P = 0.44]. There was no significant correlation between the RT of monkeys and neuronal firing rate in the Sample epoch for either stimulation condition (all P > 0.06).

Following convention,⁶³ we examined whether the correlation between neuronal firing rate and behaviour in the Decision epoch was significantly decreased from the pre- to the post-stimulation period for each stimulation condition. Fisher’s exact test confirmed a significant decrease in this correlation in the active condition (P = 0.04), while the correlation in the sham condition was unchanged (P = 0.22), confirming that tDCS over the dlPFC abolished the correlation between neuronal activity and behaviour of the monkeys.

Transcranial direct current stimulation did not induce epileptiform burst-like firing

In exploring the therapeutic potential of tDCS for epilepsy, in which cathodal tDCS has shown efficacy in reducing severe seizures in animal models,⁶⁴ there remains a concern regarding the risk of seizure induction by anodal tDCS, highlighted by a documented case of a seizure following its application.⁶⁵ Furthermore, repeated tDCS sessions might be necessary to address chronic aspects of symptoms in neuropsychological disorders (such as depression).⁶⁶ Therefore, we examined whether repeated anodal tDCS influenced the distribution of burst-like firing. Previous studies have shown burst-like neuronal firing frequencies of 200–500 Hz during epileptic episodes⁶⁷; however, owing to the significant variability and complexity in firing rates at prefrontal and cortical epileptic loci,⁶⁸ we contrasted the frequency distribution of spike rates from pre- and post-stimulation periods for both active and sham stimulation conditions to examine whether tDCS induced epileptiform burst-like firing.

To evaluate the effects of tDCS on burst-like firing rates, we calculated spike frequencies as the number of spikes within a 50 ms sliding window (refer to the ‘Materials and methods’ section). Using Kolmogorov–Smirnov tests, we found no significant differences in the distribution of spike frequencies pre- and post-stimulation in both the active (D = 0.23, P = 0.898) and sham (D = 0.31, P = 0.588) stimulation conditions. Likewise, paired t-tests assessing within-group changes showed no significant alterations in the active (t = −9.36 × 10⁻¹⁵, P = 1.00) and sham (t = −2.33 × 10⁻¹⁴, P = 1.00) stimulation conditions. Additionally, independent t-tests comparing between-group changes post-stimulation also indicated no significant differences (t = −1.77 × 10⁻¹⁴, P = 1.00).

Discussion

Despite numerous studies in humans, the neuronal mechanisms underlying the effects of tDCS on cognitive function remain unknown.^27,28 To address this, we investigated the effect of anodal direct current stimulation over the dlPFC of awake macaque monkeys on their behaviour and prefrontal single-neuron activity during the performance of a cognitive task. Our study provided evidence towards answering questions such as where, when and how tDCS influences neuronal activity to lead to the modulation of cognition. We found that tDCS over the dlPFC exerted a significant influence on task-relevant and context-dependent activities of neurons and, concomitantly, modulated the behaviour of the monkeys. These modulations appeared as a reduction in practice-related alterations in neuronal activity at cell-population level (Figs 4 and 5), which was also accompanied by a concurrent attenuation in practice-related behavioural modulation (RT slowing) (Fig. 3). Furthermore, the correlation between neuronal activity and response time was disrupted, and the variability of prefrontal cell activities was significantly attenuated (Fig. 5), whereas task-related activities of individual prefrontal cortical neurons were enhanced (exemplified in Fig. 6).

‘Where’ and ‘how’ does transcranial direct current stimulation modulate neuronal activity?

The debate regarding the capacity of tDCS to reach brain tissue to the extent that it could alter behaviour and neuronal plasticity has been persistent, with varying results from in vitro,^27,69ex vivo,³¹in vivo^70,71 and cadaver studies.²⁹ Targeting the prefrontal cortex, specifically the dlPFC in the context of working memory tasks, was purposeful owing to its central role in supporting executive control processes.⁷² Our findings provide direct evidence for the modulation of neural activities at single-neuron and cell-population levels by anodal stimulation delivered over the dlPFC. This critically addresses a long-lasting question regarding whether the effects of tDCS are exerted directly through the modulation of neurons in the underlying cortex.

Our findings also demonstrate how tDCS influences neuronal activity to facilitate behavioural modifications. We found that the effects of anodal stimulation on neuronal activity were not simply attributable to general changes, either as a decrease or an increase in baseline neuronal firing rate. Instead, active tDCS modified how neurons encoded information, manifested as five distinct effects:

• Stimulation-induced attenuation of neuronal adaptation. Active stimulation attenuated the practice-related increases in neuronal activity, observed in the sham condition and, consequently, maintained more consistent neuronal activity across pre- and post-stimulation periods (Fig. 4).

• Stimulation-induced reduction in firing rate variability during specific epochs. Active stimulation moderated the practice-related increase in variability of neuronal firing rates (Fig. 5), leading to a more uniform neuronal activity in the Sample and Decision epochs. Intriguingly, this was correlated with more consistent behaviour (Fig. 3). Moreover, this reduction in variability was not evident in the Pre-Reward epoch, suggesting that specific cognitive processes, which were recruited at specific trial stages, were influenced by active stimulation.

• Abolishing the correlation between neuronal activity and behaviour. Active stimulation decoupled the correlation between neuronal firing rate and RT. In the Decision epoch, there was a correlation between firing rate and RT both before and after sham stimulation. This correlation was also observed before active stimulation; however, after active stimulation, this correlation was eliminated.

• Enhancing task-specific activity of individual prefrontal neurons. A significantly higher proportion of neurons exhibited changes in task-related activity following active stimulation, compared with sham (exemplified in Fig. 6). These modulations were epoch specific, further emphasizing that the effects of active stimulation were linked selectively to the cognitive processes engaged during particular trial phases.

• Maintaining consistent burst-like firing rates. Active anodal tDCS did not alter burst-like firing rates when compared with sham stimulation, with frequencies remaining consistent between stimulation conditions. This stability suggests that although tDCS modulates neural activity, it does so without enhancing susceptibility to epileptiform activity, suggesting the potential for safe use in clinical settings where seizure risk must be impacted minimally.

Mechanistic basis of transcranial direct current stimulation effects

The task- and context-related effects of tDCS align with past findings that task performance modulates neural firing rates and reduces variability, observed across various sensory and cognitive tasks.^41,73-75 In particular, learning in a working memory task decreases neuronal variability in the prefrontal cortex, enhancing neural stability and consistent task execution.⁴¹ Similar reductions in variability and improvements in performance accuracy have been observed in auditory and motor tasks.^73-75 In our study, sham stimulation showed practice-related increases in neuronal firing rates and variability (Figs 4 and 5), consistent with these observations. However, anodal tDCS over the dlPFC significantly attenuated these practice-related changes, reducing increases in firing rates and variability, disrupting the correlation between neuronal activity and response time and enhancing task-related activities of individual neurons (Figs 5 and 6). This suggests that tDCS modulates neural encoding, leading to more stable and consistent firing patterns. The behavioural manifestation of changes in neuronal activity induced by active stimulation might have arisen owing to a direct influence on practice-related behavioural changes or via modulation of attentional control and vigilance. For an extended discussion, see the Supplementary material.

‘When’ does tDCS modulate neuronal activity?

The temporal dynamics of tDCS effects are crucial for a comprehensive understanding of its impact on the involved cognitive processes and underlying neural mechanisms. Our results elucidate the temporal dynamics of stimulation effects on neuronal activity by demonstrating that modulation of prefrontal cell activities by direct current stimulation persisted beyond the stimulation period (i.e. effects extended to the post-stimulation period). Notably, stimulation-induced behavioural changes also followed a similar pattern, highlighting the sustained impact of direct current stimulation on both neuronal activity and behaviour beyond the stimulation period. Furthermore, our findings indicate that the influence of tDCS on neuronal activity is state specific, in that tDCS selectively modulated practice-related neuronal alterations without changing the overall activity level (Fig. 4). Additionally, the effect of tDCS on the activity of individual cells was temporally specific (exemplified in Fig. 6), primarily influencing the Sample and Decision epochs, a finding that emphasizes that the effects of active stimulation over the dlPFC are linked to specific cognitive processes.

Such characterization of the temporal effects of tDCS aligns with and builds upon established observations from prior transcranial electrical stimulation research. In this context, previous research has shown that behavioural changes induced by transcranial alternating current stimulation might arise from neuronal ‘entrainment’ to the temporal pattern of the delivered current.⁷⁶ However, although some studies have shown the capacity of transcranial alternating current stimulation to entrain neural activity,^76-78 others have not.⁷⁹ Our observations demonstrating that anodal tDCS induces persistent electrophysiological changes beyond the stimulation period, which are not dependent solely on ‘entrainment’ to the temporal pattern of the applied current (Figs 4 and 5), contribute a nuanced layer to the broader understanding of transcranial electrical stimulation modalities and their interactions with neuronal activity and behaviour.

Time-dependent neuronal adaptation^76,79 and alterations in firing variability^39,40 have been linked with information encoding and correlated activity across various brain regions. Our findings (Figs 4 and 5) indicate that time-dependent firing rate variability can be modulated by tDCS over the dlPFC. Other transcranial electrical stimulation techniques, such as transcranial alternating current stimulation, have also been shown to attenuate neuronal adaptations in the medial temporal⁷⁹ and motor⁷⁶ cortices.

Our findings contribute to a mechanistic understanding of the neuronal underpinnings of the effects of transcranial electrical stimulation, resonating with and expanding upon past theoretical models.^80-83 The observed task-specific modulation (Figs 4 and 5) of neuronal firing rates in the post-stimulation period is consistent with theoretical models that emphasize the dependence of stimulation outcomes on the pre-existing state of neural activity,^80-82 which might also explain some aspects of variabilities in the behavioural outcome of tDCS in past studies. For an extended discussion, see the Supplementary material. Furthermore, our results highlight the importance of considering the broader network dynamics in which these neurons operate, highlighting the capacity of tDCS to influence neural network states and ongoing neuroplasticity.^80,81,83

Conclusion

Our findings showing concomitant modulation of prefrontal cell activity and behaviour provide mechanistic evidence to address ‘where’, ‘when’ and ‘how’ tDCS might exert its cognitive-behavioural effects. The present findings indicate that the impact of tDCS is not generalized, but instead involves intricate and specific modulations in how neurons encode task-related information, dependent on the underlying state of the neural networks. Moreover, multiple tDCS sessions did not increase burst-like epileptiform susceptibility. This study, therefore, serves as a significant step forwards in our understanding of the mechanistic underpinnings of tDCS, providing an evidence-based framework for its effects that bridges the gap between cellular-level modulations and cognitive-behavioural outcomes, paving the way for a safer and more effective application of brain stimulation techniques in research and clinical settings.

Data availability

Data are available upon reasonable request to the corresponding author.

Acknowledgements

We thank Abhishek Datta and Soterix (NJ, USA) for conducting the finite element method simulation for our study. We would also like to thank the RIKEN Support Unit for Functional Magnetic Resonance Imaging for conducting the structural MRI.

Funding

F.A.M.’s contribution to this project was supported by the Australian Research Council (ARC) . Contributions by D.J.F., M.G.P.R. and F.A.M. were also supported by the ARC Centre of Excellence for Integrative Brain Function. T.Y.’s contribution to this project was supported by Program for Brain Mapping by Integrated Neurotechnologies for Disease Studies, Japan (Brain/MINDS; JP15dm0207001 to T.Y.).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

Author notes

Marcello G P Rosa and Farshad A Mansouri contributed equally to this work.