Targeting The Frontoparietal Network Using Bifocal Transcranial Alternating Current Stimulation During A Motor Sequence Learning Task in Healthy Older Adults Part 2

2.3. Motor learning task

Participants executed two different versions of the SFTT basedon the SFTT task used in earlier studies [3,47]. They were asked to perform a 9-item sequence with their non-dominant left hand. Thenon-dominant hand was used to allow for a larger range ofimprovement [46]. They were orally instructed to continuously tapthe same sequence as fast and as accurately as possible on a four-button keyboard (Current Designs, Philadelphia, PA, USA). 

Secondly, mental health also has a great impact on improving memory. Good sleep habits, positive thoughts and emotions, and reducing stress can all help improve our memory. Research shows that emotionally stable and positive people tend to have better memory performance. At the same time, people should overcome negative emotions such as low self-esteem, anxiety, and depression, so as not to interfere with their memory processes and performance.

The sequences consisted of 4 digits from 2 to 5, which corresponded to thefour fingers from the index (2) to the little finger (5) of the left hand. Acursor underneath the displayed sequence moved in response toevery finger tap to identify the target digit, regardless of whether ornot the button was pressed correctly. Different sequences wereused for the baseline and the training measurements. 

All sequenceswere matched in complexity verified with the Kolmogorovcomplexity test [48]. The baseline measurement consisted of oneblock of 90 s, and the training measurement consisted of seven blocks of90 s, with 90 s breaks after every block which lasted 20 min in total.The task was implemented in Presentation software (Neurobehavioral Systems, Berkeley, CA, USA). 

Participants performed alow and a high WM load SFTT version. The low WM load versiondisplayed the sequence on the screen asking participants to executethe sequence without prior familiarization. This version is referredto as the "non-memorized" version. For the high WM load version,participants had to memorize the sequence before the task started. 

They received the sequence on paper and were asked to learn thesequence by heart, without practicing it on the button keyboard.With the use of a distractor task, during which they had to spellrandom words in a reversed order sufficient memorization of thesequence was verified. More precisely, participants had to spellbackward 3 words in a row and recall the sequence out loud afterward. After 3 times correct, the sequence was deemed sufficiently learned [49]. 

During the memorized version of the task, theparticipants could not see the sequence. Displayed on the screen was a sequence of 9 "X's" with the moving cursor underneath toidentify the target digit. Participants performed both versionsdivided over day 1 and day 2 in a randomized order. This order ofversions was reversed after cross-over, see Fig. 1C.

2.4. Cognitive task

Participants were asked to perform the N-back task to verifywhether this stimulation paradigm enhanced WM performance.The task was implemented in Matlab (The MathWorks Inc., Natick,Massachusetts, USA) and was based on the single N-back task usedby Jaeggi and colleagues [50]. The script was adapted from Quent,A.J. [51]. in terms of language (French & English), length, and difficulty level. 

The task consisted of a sequence of visual stimuli thatwere shown on a computer screen. The participants had to respondby clicking the right "Control" button on a computer keyboardwhen the stimulus was the same as the stimulus presented N positions back. Participants should not respond when a differentstimulus is presented. The visual stimuli consisted of 10 randomshapes, eight 8-point shapes (numbers 14, 15, 17, 18, 20, 22, 23, and27), and two 12-point shapes (numbers 20 and 24) taken fromVanderplas and Garvin [52]. 

The stimuli were presented for 500 mseach with a 2500 ms interstimulus interval. The participants wererequired to respond within the response window that starts at theonset of the stimulus until the end of the interstimulus interval(3000 ms). The task consisted of 1 to 3 back levels, in that order.The task was divided into a baseline and training session, with thebaseline session consisting of 1 block per n-back level (3 blocks intotal) and the training session of 3 blocks per level (9 blocks intotal). 

Every block consisted of 20 þ n trials, with 6 targets and14 þ n non-targets. The reaction times, hits, misses, false alarms, and correct rejections were measured. Please see Fig. 1 B for aschematic illustration of the task. We had to exclude N ¼ 9 beforecross-over N-back task data sets due to an error in the responserecording. A total of N ¼ 31 N-back data sets were considered.

2.5. Transcranial alternating current stimulation

Multifocal tACS was applied to the right FPN using two neuroConn DC plus stimulators to enable bifocal stimulation (neuroConnGmbH, Ilmenau, Germany). Participants received both real (30 min)or sham (30 s) stimulation in randomized order, before or aftercross-over [24,25,53]. The stimulation protocol consisted of thefollowing parameters: in-phase (0⁰ phase lag), intensity 2 mA(peak-to-peak) was gradually ramped up/down with an interval of8 s. The in-phase stimulation between the two stimulators wasassured by a repeated trigger from stimulator A to stimulator B afterevery completed cycle to signal the start of a new cycle [54]. 

Thestimulation frequency was adjusted to the personal theta peakfrequency, which was recorded during an EEG recording whileperforming a pre-baseline N-back test of 1 block per level. Rubberconcentric electrodes were used: center electrode size diameter: ca.20 mm, area: ca. 3 cm2 and ring electrode size diameter: out 100mm/in 70 mm, area: ca. 40 cm2. 

Electrode location was definedwith the use of a standard 64-channel, EEG actiCAP with a 10/20system (Brain Products GmbH), targeting F4 corresponding to thedorsolateral prefrontal cortex (DLPFC) and P4 corresponding to theposterior parietal cortex (PPC). The paste used for conductivity withadequately low impedance was SAC2 electrode cream (Spes Medical Srl, Genova, Italy). This paste was adhesive which ensuredstable electrode placements. The electrode placement and theelectric field distribution were visualized with the use of a standardtemplate in SimNIBS (Version 3.2) [55]. The script to implementbifocal stimulation with ring electrodes was adapted from the open-access Matlab script (© G. Saturnino, 2018). 

A template head modelwas used to simulate the electrode placement and electric fielddistribution. For the electrode placement and electric field distribution, please see Fig. 2 A & B. At the end of the last stimulationsession, we investigated whether the stimulation was well tolerated and if there was a significant difference in experienced sensations between the real and sham conditions. 

Moreover, we askedthe participants to indicate whether they thought they hadreceived real or sham stimulation during the before and after crossover sessions. The stimulation sensations were described with theuse of a structured interview [56]. We checked for the following sensations: itching, pain, burning, metallic/iron taste in the mouth,warmth, fatigue, and others. With the possibility to respond: "none","mild", "moderate", or "strong".

2.6. EEG

All EEG recordings were done in a shielded Faraday cage. Acustomized electrode set-up with 9 electrodes was used, Frontal(Fp1, Fp2, F3, Fz, F4), parietal (Cz, P3, Pz, P4), please see Fig. 2C.Using a 64-channel ANT Neuro EEG cap with eegotmmylab software(ANT Neuro, Netherlands). EEG was recorded during the performance of the N-back task. With the use of markers, the beginningand the end of every separate N-back level were defined. Recordings were done during 3 N-back blocks resulting in approximately 3 min of recording time. 

The peak frequency in the thetarange (4e8 Hz) was calculated using a custom Matlab script (TheMathWorks Inc., USA) adapted from the script used by SalamancaGiron and colleagues [54] and made suitable for theta frequencyanalysis during N-back task performance. Theta frequencies fortACS were personalized similar to previous work [54,57]. However,we did not intend to compare the efficacy of tACS with personalizedfrequencies to tACS with standard (non-personalized) frequencies. 

Therefore, this study does not aim to demonstrate the beneficialphysiological effects of tACS with personalized over tACS withstandard (non-personalized) frequencies. The target electrodes F4& P4, which are the same as the stimulation locations show a smallvariance in recorded theta frequency. The average theta frequencyfor the F4 electrode was 4.71 (range 4.12e7.77) and for the P4electrode 4.97 (range 4.11e6.84).

2.7. Data analyses

The normality of the data was visually checked with histograms andQ-Q plots of residual values and confirmed by verification ofskewness ranging between 1 and -1 [58]. P-values of < .05 indicatestatistical significance. Pre-processing of the behavioral data ofthe SFTT was done with an in-house script implemented in Matlab. The main output measures were: correct sequences, total completedsequences, and correct sequences/completed sequences. Preprocessing of the individual N-back data was done with RStudio(version 1.4.1717, 2021) [59]. 

The baseline blocks were compared in R using paired-sample t-tests. The equality of the baseline blocks was verified usingBayesian statistics by computing a Bayesian paired-samples t-testwith the use of JASP software (version 0.16.0.0). All other analyses ofthe SFTT and the N-back data were done in Rstudio (version1.4.1717). Data were analyzed with the use of Linear mixed-effectsmodels that were fitted with the "largest" package. Output was a type III ANOVA table with p-values for F-tests [60]. 

The effect size wasdetermined using partial eta squared with the "effect size" package.Post-hoc analysis was done by pairwise comparisons, using theestimated marginal means and Tukey correction. Analysis of thetACS stimulation sensations and blinding responses were analyzedwith JASP (version 0.8.5.1) [61]. Responses to the real vs. shamstimulation estimations were analyzed using a binomial test. Thestimulation sensations were analyzed using contingency tableswith chi-squared analysis to control for differences between thereal and sham stimulation conditions.

3. Results

3.1. Sequential finger tapping task

The two SFTTs have been analyzed separately as they differ in the amount of WM load (high and low WM load). Before the mainanalysis, the baseline performance between active and shamstimulation was compared and was not significantly different forboth the memorized condition t(19) ¼ 0.72, p ¼ .48, d ¼ 0.16, andthe non-memorized condition t(19) ¼ 0, p ¼ 1, d ¼ 0. To furtheranalyze the null result and to confirm the equality of the groups activevs sham groups were compared in both conditions using Bayesianstatistics. The analysis indicated for the memorized conditionBF01 ¼ 3.41, meaning it is 3.4 times more likely that the baselineresults are equal than different. The non-memorized conditionindicated BF01 ¼ 4.3, therefore is 4.3 times more likely that thebaseline groups are equal.

In this study, online learning is defined as a significantimprovement in behavior within the training session. A significanteffect of stimulation on learning is defined by a change inimprovement dynamics during the training. With the use of a linearmixed effects model, the analysis of the number of correct sequences of the memorized version of the SFTT showed a significanteffect for blocks F(6, 247) ¼ 18.57, p < .001, hp2 ¼ 0.31 indicating alarge effect size, as well as a significant effect for stimulation F(1,247) ¼ 18.83, p < .001, hp2 ¼ 0.07 with medium effect size, but noblocks stimulation interaction F(6, 247) ¼ 0.77, p ¼ .59, hp2 ¼ 0.02. 

To further define the effect of stimulation on learning, we determined the difference between the conditions at the end of thetraining, which showed a strong trend for a significant differencet(19) ¼ 2.07, p ¼ .052, d ¼ 0.46. The results of the nonmemorized version show a significant effect for blocks F(6,247) ¼ 16.00, p < .{{20}}01, hp2 ¼ 0.28 (large effect), but no stimulationF(1, 247) ¼ 0.46, p ¼ .499, hp2 ¼ 0.002 or interaction effect F(6,247) ¼ 0.36, p ¼ .901, hp2 ¼ 0.009. Indicating that in both conditions,participants learned significantly, but only in the memorized condition there was a significant effect of tACS stimulation on performance, see Fig. 3. 

The lack of an interaction effect does not allow toconclude a significant effect of stimulation on motor learningalthough the trend for a difference in performance on block 7 indicates a potential for stimulation effect. To further investigate theresults of the SFTT and appreciate the variance in performance theindividual trajectories of the participants are indicated in the supplementary material, Supplementary Fig. 1. Analyses with nonnormalized data revealed comparable findings, for details pleasesee the supplemental online material (SOM) and SOM Fig. 3.

3.2. Speed and accuracy

To further investigate the results of the memorized condition,the total amount of completed sequences was analyzed as ameasure of speed. The results showed a significant block effect F(6,247) ¼ 28.21, p < .001, hp2 ¼ 0.41 (large effect) and a significantstimulation effect F(1, 247) ¼ 15.92, p < .001, hp2 ¼ 0.06 (small effect), but no interaction effect F(6, 247) ¼ 0.23, p ¼ .968, hp2 ¼ 0.006,see Fig. 4A. Although the active stimulation group is fastercompared to the sham group, the similar pattern of improvementpoints towards a performance rather than a learning effect.

To see whether the increased amount of correct sequences was driven by faster sequence execution or by a simultaneous increase in accuracy, we analyzed the ratio between the totalamount of sequences and the correct sequences as an accuracymeasure. Upon inspection, the real stimulation group showsdifferent dynamics in accuracy than the sham group. The realstimulation group demonstrates a steep significant increase in accuracy between the first and the second training block while thesham group's increase is more gradual t(19) ¼ 2.68, p ¼ .015. Theaccuracy between the groups during the 1st training block was notsignificantly different T(19) ¼ 0.85, p ¼ .404. Therefore, to visualizethe difference in dynamics we measured the difference in accuracyabout block 1. 

Results showed a significant block effect F(6,247) ¼ 3.47, p ¼ .003, hp2 ¼ 0.08 (medium effect), and a significantstimulation effect F(1, 247) ¼ 18.31, p < .001, hp2 ¼ 0.07 (mediumeffect), but no block stimulation interaction F(6, 247) ¼ 0.85,p ¼ .529, hp2 ¼ 0.02, see Fig. 4B. In an additional analysis the comparison of behavior on block 7 shows a significant difference between verum and sham t(19) ¼ 2.31, p ¼ .032, d ¼ 0.51.Therefore, although the lack of an interaction effect does not signifysignificant motor learning effects, the results do indicate that accuracy significantly improved with stimulation in the early stage oftraining, which remained significantly different in the last block.Analyses with non-normalized data revealed comparable findings,for details please see SOM and SOM Fig. 4.

3.3. N-back task

The N-back task performance was analyzed by the followingoutcomes: hits, false alarms, accuracy (hits e false alarms), andreaction time for hits. All parameters were analyzed separatelyusing linear mixed-effects models. The stimulation conditions (realvs. sham) and the three N-back difficulty levels were included asindependent variables in the model. Two separate analyses wereperformed, one model included difficulty levels 1 and 2 to mimicthe conditions comparable to the study of Violante et al. (2017),additionally, we added difficulty level 3 to the model to test for astimulation effect on the task with higher cognitive demand [24].

For more information:1950477648nn@gmail.com