Conventional vs. High-de nition tDCS: A Comparison of Neurophysiological and Behavioural Effects

Fabio Masina (  fabio.masina@ospedalesancamillo.net ) IRCCS San Camillo Hospital, Venice Giorgio Arcara IRCCS San Camillo Hospital, Venice Eleonora Galletti Department of General Psychology. University of Padua Isabella Cinque Department of General Psychology. University of Padua Luciano Gamberini Human Inspired Technology Research Centre. University of Padua Daniela Mapelli Human Inspired Technology Research Centre. University of Padua


Introduction
Transcranial direct current stimulation (tDCS) is a non-invasive brain stimulation (NIBS) technique that uses a constant weak electric current (generally 1-2 mA) to modulate speci c brain areas over which it is applied [1][2][3] . The current passes through the scalp, skull, and cerebrospinal uid down to the cortex, bringing membrane potential of neurons closer to or farther from the action potential threshold 4

. In
Conventional tDCS, the current is delivered through two electrodes, an anode and a cathode, typically sized 25-35 cm 2 1 . One electrode is placed over the area of interest, while the other one is placed over another site, either cephalic or extracephalic 5 .
Despite the growing popularity of Conventional tDCS, especially as a promising tool for the treatment of clinical conditions, several criticisms have undermined the reliability of this technique: the inconsistency of tDCS effects and the di culty to replicate previous ndings 6 . The reasons behind the crisis of tDCS are related to several aspects, mostly methodological, which likely had a substantial impact in the replication of the results 7,8 .
A further critical aspect, partially neglected in the literature, is related to biophysical properties, and concerns the spatial distribution of the electric eld produced by Conventional tDCS. As a matter of fact, Conventional tDCS produces a wide-spread electric eld with low spatial speci city and peaks of intensity falling outside the active electrode 9,10 . Since the stimulation reaches brain areas that are structurally and functionally different, the electric eld distribution may be another issue explaining such variability and inconsistency of tDCS-induced effects at both behavioural and neurophysiological levels.
Recently, new tDCS montages and devices have been introduced to overcome the limitation of Conventional tDCS. High-de nition tDCS (HD-tDCS) montages use smaller electrodes than Conventional tDCS allowing the electric current to be delivered with increased density and focality. The most used HD-tDCS montage is the 4 × 1 ring con guration, which consists of one active electrode placed on the area of interest and four return electrodes on the surroundings. In this way, the delivered electric current is constrained and localised within the return electrodes 9 . Finite element method (FEM) models predict a different strength and distribution of electric eld induced by Conventional and HD-tDCS [10][11][12] . In particular, speci c electrode placements of HD-tDCS contribute to reduce the uncontrolled diffusion of tDCS-induced electric elds, thereby improving the spatial precision with which the electrical current can target speci c cortical regions 13 . Findings coming from physiological studies have also presented a higher precision of HD-tDCS in modulating neurophysiological components as compared to Conventional tDCS 9,14 .
Although different electric eld distributions have been found in computational studies concerning Conventional and HD-tDCS, no clear conclusion can be drawn on whether HD-tDCS can really produce a more focal and strong modulation than Conventional tDCS, both at behavioural and neurophysiological levels 15,16 . Indeed, most of the differences between the two techniques are based on mathematical simulations, and not on empirical effects of the stimulation.
In order to ll this gap and to further elucidate the differential effects of Conventional and HD-tDCS, the main goal of this research is to compare these two methods, investigating in the same study, for the rst time, both behavioural and neurophysiological outcomes. Importantly, to evaluate tDCS effects, this study capitalises on some methodological advances as compared to common standards. The large majority of studies using tDCS investigate effects by comparing baseline and post-stimulation measurements 17,18 , or by using change scores, that is analysing effects obtained by subtracting baseline from post-stimulation measurements 19,20 . These widespread approaches assume several hidden statistical assumptions (often not satis ed) and ignore some well-known potential distortions. One of these is the phenomenon of regression to the mean 21 , according to which extreme values (very low or very high) tend to be closer to the mean in a repeated measure for pure statistical reasons. Moreover, baseline levels can constraint potential outcome of treatment also for other reasons. For example, if the behavioural scores are already at ceiling at baseline, no further improvement is possible and this cannot be explicitly taken into account using mean scores or change scores in the statistical analysis (as it happens with t-test or ANOVA).
To overcome these limitations and following gold standard for randomised control treatment design in biostatistics 21 , in the present study post-stimulation measurements were adjusted for baseline levels, considering these latter as covariates in statistical models. By using this approach, the question of the study changes from "Is HD-tDCS better than Conventional tDCS?" to "Are there speci c conditions under which HD-tDCS is better than Conventional tDCS?". This is a methodological shift towards addressing more proper questions, as the investigation of tDCS effects cannot ignore the potential of observing actual modi cations which highly depend on baseline values.
Remarkably, applying this rationale and method to neural data (e.g., EEG data)allows to properly control for the initial activation state of brain before the stimulation, hence taking into account the phenomenon of state-dependency 2,22 . Although several articles advocate the importance of state-dependency in explaining the variability of observed effects or the failure of observing signi cant effects 2,23 , yet few studies actually take into account initial state as a variable in the analysis, so explicitly investigating the actual in uence of statedependency effects.
In the present study, tDCS was used in combination with a motor task given the large body of evidence suggesting the technique would induce the strongest effects within the motor domain (for a review and meta-analysis, see 24 ). In particular, tDCS was delivered with the aim to modulate participant's dexterity, which was assessed with a computerised version of the Finger Tapping Task, administered before, during, and after the stimulation stage. Furthermore, the Purdue Pegboard Test 25 was used to assess dexterity in a non-computer based fashion. We expected that for the HD-tDCS condition we would observe a greater improvement in participants' dexterity relative to Conventional tDCS and sham. Furthermore, we expected greater dexterity improvement in Conventional tDCS relative to sham.
At a neurophysiological level, resting-state EEG was recorded before and after the stimulation stage. Power spectrum density was calculated selecting two EEG bands, respectively alpha (8-12 Hz) and beta band (18)(19)(20)(21)(22). The rationale of selecting alpha and beta bands is based on the functional role of these two bands. While the alpha frequency band has been linked with general inhibitory mechanisms 26 , thus liking the decrease of alpha power to motor facilitation, the beta power has been related to sensory-motor functions 27 .
A recent review 28 shows that anodal tDCS reduces alpha power while increasing beta power. For this reason, we expected that anodal HD-tDCS would produce a greater modulation of the neurophysiological signal, leading to a signi cantly greater decrease of alpha and increase of beta power as compared to the Conventional and sham conditions. Moreover, we expected that Conventional tDCS would cause a greater modulation relative to sham.

Results
The Finger Tapping Task The mean response times (RTs) and accuracy for each stimulation condition are summarised in Table 1.
With regard to RTs, in the model with the beta power entered as covariate, a main effect of stimulation condition was found [F(2,29.9) = 3.51, p = .043]. Post-hoc contrasts did not show signi cant differences between the conditions (lowest p = .113). In this model, a signi cant stimulation condition * beta power at baseline stage interaction was found [F(2,29.9) = 3.58, p = .04], (Fig. 1). Contrasts were performed at the level of the 1st, 2nd, and 3rd quartile of the beta power at the baseline stage, respectively − 14.6, -14.3, -14 power units. Post-hoc tests showed a tendency towards signi cance: for lower beta at baseline (i.e., -14.6 power units), HD-tDCS reduced RTs compared to sham (1514 ms vs. 1634 ms, p = .068). No effect was identi ed in the models that considered as covariate the alpha power, as well as all the models on accuracy (see supplementary material).

The Purdue Pegboard Test
The mean performance at the Purdue Pegboard Test for each stimulation condition is shown in Table 1. The models considered in the analysis did not show any signi cant effects (see supplementary material).

Discussion
The present study aims to provide new evidence characterising functional differences between Conventional and HD-tDCS. To properly evaluate effects of both the montages, we adopted a state-of-the-art approach for the statistical analysis, that overcomes some common suboptimal statistical methods and focalises on properly identifying the conditions associated with speci c treatment outcome 21 . This approach consisted in adjusting the post-stimulation measurements for baseline levels, considering these latter as covariates in the analysis. Of importance, this method allowed to properly account for state-dependency 2,29,30 , a phenomenon often evoked to explain variability and inconsistency of tDCS-induced effects 31 , but generally neglected in statistical analysis.
Results from the present study outline distinctive modulation of Conventional and HD-tDCS. This evidence, observed empirically, is possibly related to the different strength and distribution of the electric eld induced by Conventional and HD-tDCS, as predicted by FEM computational models [10][11][12] . Firstly, we found that following HD-tDCS, a reduction of EEG alpha power during resting state could be observed. This result is consistent with previous studies showing an inhibitory effect of anodal tDCS on alpha power 28 . The functional role of alpha is commonly related to cortical deactivation and inhibition 26,32,33 and animal models provide evidence that alpha-band oscillations have an inhibitory in uence on the generation of spikes 34 . Consequently, if an increase of alpha power re ects inhibition, a reduction in power should re ect release from inhibition, supporting evidence that anodal HD-tDCS on M1 induces cortical excitability 14 .
In line with our expectations, the state of neural activation before applying HD-tDCS played a role in the modulation of alpha. Speci cally, a reduction of alpha power was observed only in participants that had lower EEG alpha before HD-tDCS. A possible explanation of this result may rely on the cognitive role linked to alpha band, namely its involvement in attention 26 . This viewpoint would suggest that only participants with a proper attentional asset (i.e., lower alpha and, consequently, higher level of attention before the stimulation) bene ted from a release of inhibition following the administration of HD-tDCS. Remarkably, this result contributes to identify which physiological markers can predict tDCS effects, especially the effects of HD-tDCS that is relatively recent among NIBS techniques.
Contrary to our expectations, we found that Conventional tDCS induced a reduction of beta power. Speci cally, in our study the reduction of beta occurred for participants who already showed a higher level of beta power before the administration of Conventional tDCS. The inhibition of beta induced by anodal tDCS is not unusual in the literature 35,36 , despite previous evidence showing an opposite pattern, namely an increase of beta power 28 .
The inconsistency of these ndings may be accounted for as a consequence of the uncontrolled spread of the electric eld in Conventional tDCS 9 . As shown, in tDCS the current generally concentrates at the edge of the electrode 37 . Thus, the larger is the electrode size the lower is the probability to keep the edge of the electrode over the target area. With the unlikelihood of being able to control the electric current diffusion, a possible concern arises in the fact that tCDS may affect more than just the target region. Consequently, Conventional tDCS, which typically uses large electrodes sized 25-35 cm 2 , may lead to undesired or mixed effects due to the stimulation of nearby areas connected to the target area. Further studies are necessary to con rm this hypothesis, possibly by comparing functional outcomes of Conventional and HD-tDCS in the same study.
With respect to behavioural outcomes, HD-tDCS produced some evidence of motor improvement in participants who had lower beta power at the baseline stage. This nding would suggest that HD-tDCS can induce enhancement of unimanual dexterity, as shown by a previous study 38 . Functionally, beta power has been hypothesised to be linked to sensory-motor functions 27 . Generally, a reduction of beta power is seen in planning and execution of motor action, followed by an increase of power after the end of the movement 39,40 . However, the dynamic uctuations of beta power before, during, and after a voluntary movement have to re ect several underlying processes. While the desynchronisation of beta (i.e., power decrease) would be related to the asynchronous activation of the motor cortex during a movement, the synchronization (i.e., power increase) seems to re ect several mechanisms including motor control 41 and the maintenance of tonic activity at the cost of voluntary movements 42 . Interestingly, this latter hypothesis is coherent with a recent study showing that the increase of beta activity would induce a slowdown of movements 43 . Thus, the improvement in dexterity granted by HD-tDCS might prove to be higher if the brain state has an adequate level of disposition to the movement onset (i.e., lower beta power). Nevertheless, given the small effect found, further research is necessary to con rm the possibility to enhance dexterity by HD-tDCS, possibly adopting a different task or investigating clinical population, in which the risk of a ceiling effect is reduced and the potential to improve motor performance is higher 44 .
Of importance, from a statistical point of view, results of this study underline the importance of adjusting the post-stimulation measurements for baseline levels, including these latter as covariates in statistical models. This approach, in line with recent suggestions for randomised control treatment designs in biostatistics 21 , should be considered as a good practice in future NIBS studies, especially with the aim to investigate the effect of stimulation not only per se but also in relation to the initial state of brain, and to the actual potential to be in uenced by tDCS.
In summary, the present study contributes to characterise the differences between Conventional and HD-tDCS, both from a behavioural and neurophysiological perspective. HD-tDCS represents a recent advance in NIBS since it would overcome the low precision of Conventional tDCS. However, few studies have compared whether the increased focality of HD-tDCS could determine different modulation. Interestingly, our ndings support this theory and show how HD-tDCS can induce more predictable outcomes than Conventional tDCS. Signi cantly, the present study also highlights the importance of considering the initial state of brain activity before tDCS application, as it can be crucial to in uence the effects of tDCS.

Methods And Materials
Participants Thirty participants were recruited from the University of Padua. They were matched for gender ( Participants with a history of neurological or psychiatric diseases were excluded from the study. They were all checked for tDCS exclusion criteria 46 . All safety procedures were in line with tDCS guidelines 46 . Before the experiment, participants gave their written informed consent. The study was approved by the ethics committee of the Human Inspired Technology (HIT) Research Centre in Padua (nr. 2019_39) and was compliant with the ethical principles of the 1964 Declaration of Helsinki.
Transcranial direct current stimulation (tDCS) The tDCS and EEG recordings were carried out through a multi-focal tDCS-EEG device (StarStim, Barcelona) with 20 channels. The system was remotely controlled via the Neuroelectrics Instrument Controller (NIC; v2.0.11.4). Participants were involved in three experimental sessions, during which a different tDCS montage was applied, namely Conventional tDCS or HD-tDCS.
In Conventional tDCS, two circular saline-soaked surface sponge electrodes (surface = 25 cm 2 ; current density: 0.06 mA/cm 2 ) were used. The anode (active electrode) was placed on C4 (International 10-20 EEG System), while the cathode (return electrode) was placed over the contralateral (left) shoulder of participants.
Along with the real tDCS, a sham condition was included where the montages were counterbalanced. Hence, 50% of participants received sham with the Conventional tDCS montage and the other 50% received sham with the HD-tDCS montage.
At the end of each session, participants completed a questionnaire of tDCS-related sensations 49

Tasks
The description of the behavioural tasks performed by participants (i.e., the Finger Tapping Task and the Purdue Pegboard Test) is available in supplementary material. The Fig. 3 shows a representation the Finger Tapping Task.

Procedure
Participants were involved in three experimental sessions (Conventional tDCS, HD-tDCS, and sham), carried out on separate days and separated by a washout period lasting between 6 and 16 days. Importantly, the stimulation conditions were counterbalanced within the three experimental sessions. Each experimental session was divided into six steps.
Firstly, the EEG headcap was placed on the scalp. All impedances were kept below 5 KΩ. The session started with 5 minutes of restingstate EEG. During the EEG recording, participants were asked to stare a xation point kept at 60 cm distance. Successively, participants performed the Finger Tapping Task (baseline stage) without any EEG recording. After this stage, they were invited to perform the Finger Tapping Task for 20 minutes (stimulation stage) while being delivered the stimulation (Conventional tDCS, HD-tDCS, or sham). The stimulation condition lasted for 20 minutes, meaning that the real stimulation (i.e., Conventional tDCS or HD-tDCS) or sham was delivered for the entire duration of the Finger Tapping Task. Regardless of the stimulation condition, the current strength was 1.5 mA with a ramp up/ramp down time of 30 seconds. In the sham condition, the current linearly increased for the rst 30 seconds up to a 1.5 mA and then decreased to 0 mA in the next 30 seconds. After the stimulation stage, 5 minutes of resting-state EEG were recorded, following the aforementioned procedure. At the end of the EEG recording, participants performed the Finger Tapping Task for a third time (poststimulation stage). Finally, the Purdue Pegboard Test was administered. Figure 3 shows a representation of the procedure.

Statistical analysis
All data were analysed using RStudio software 50 (version 1.2) and packages lme4 51 , lmerTest 52 , car 53 , and emmeans 54 . Linear mixed effect models (LMMs) and generalised linear mixed effect models (GLMMs) were used. Signi cance of the xed effects terms were assessed by means of F-test using Satterthwaite approximation 55 . Post-hoc pairwise contrasts were corrected with Tukey's multiple comparison test. For signi cant interactions between a continuous variable and a factor, estimated marginal means contrasts were performed at the level of the 1st, 2nd, and 3rd quartile of the continuous variable. All relevant data and R scripts are available at https://osf.io/j4acs/, while all models performed in the analysis are available in supplementary material.
In the models, post-stimulation measurements, both behavioural and neurophysiological, were covariate-adjusted for baseline levels 21,56 .

The Finger Tapping Task
Participant's performance at the Finger Tapping Task was evaluated in terms of RTs and accuracy. RTs were measured as the time interval between string onset and the typing of the fth digit (Fig. 3). RTs below 100 ms were removed from the analyses, as well as the strings not correctly typed.
Accuracy was computed as the ratio between the number of correct strings and the total number of strings. Error and correct strings were dichotomously coded, respectively as 0 and 1. As a consequence, accuracy was analysed by using GLMMs for binomial data, using a logit link function 57 .
The models were t to investigate the effects of tDCS on post-stimulation performance, considering baseline alpha or beta power as covariates. In the models, stimulation condition (Conventional tDCS, HD-tDCS, sham) and the covariates alpha/beta power at the baseline stage (see EEG analysis) were considered as xed-effect factors, and participant, stimulation condition, and string repetition as random intercepts (Table A.1, supplementary material).

The Purdue Pegboard Test
Participants' performance at the Purdue Pegboard Test was scored as the mean number of pins, collars, and washers placed in the board, accordingly to the task instructions.
Since the anode was placed over C4 (right hemisphere), we restricted the analysis only to the mean score at the subtest performed with the left hand. Differences among the stimulation conditions (Conventional tDCS, HD-tDCS, sham) were investigated by tting LMMs. Stimulation condition (Conventional tDCS, HD-tDCS, sham) and the covariates alpha/beta power at the baseline stage were considered as the xed effect factors, and participant was included as the random-effect (

EEG analysis
The EEG data were pre-processed o ine with Brainstorm 58 for Matlab R2017b (The Mathworks Natic, MA, USA). First, continuous EEG was band-pass ltered with a cut-off frequency of 0.1-47 Hz. Then, the continuous EEG signal was visually inspected and channels with noise signal were removed. Independent component analysis 59 was performed to correct the remaining artifacts (muscle activity and eye blinks). All independent components were visually inspected in terms of scalp distribution, frequency, timing and amplitude 60 . The mean number of removed independent components was 1.93 (SD = .87). Afterwards, the EEG was segmented into 150 non-overlapping epochs of 2000 ms.
Baseline correction was performed by subtracting the mean voltage of the whole epoch. The EEG signal was re-referenced to the mean of all channels. Epochs containing data points exceeding the amplitude of -100 mV/+100 mV were excluded from the analysis. An average of 8.47% epochs were excluded. Successively, power spectrum density (Welch's method) was conducted to extract power [signal units/sqrt(Hz)*10 − 5 ] relative to the two bands of interest: alpha band (8-12 Hz) and beta band (18-22 Hz). Data were averaged within each band and were log-transformed to reduce skewness. Only the power extracted from the electrodes close to the stimulation site (i.e., F4, Cz, T8, P4), and from C4 was considered in the analysis.
To investigate the effects of the stimulation conditions (Conventional tDCS, HD-tDCS, sham) on alpha/beta power at the post-stimulation stage, two LMMs were conducted. Stimulation condition (Conventional tDCS, HD-tDCS, sham) and the covariate alpha/beta power at the baseline stage were considered as xed-effect factors. Random structure of the models consisted of participant and stimulation condition (Table A.1, supplementary material).

DATA AVAILABILITY
All relevant data and R scripts are available at https://osf.io/j4acs/, while the description of tasks (i.e., the Finger Tapping Task and the Purdue Pegboard Test) as well as all the models performed in the analysis are available in supplementary material.