General Study Design and Equipment
The present study is a sham-controlled study on individuals diagnosed with ADHD. The study is designed as a two × two mixed design with factors time (pre/post intervention) and group (stim/sham). Each patient attended one experimental session of approximately two and a half hours duration. The session was split into three blocks (pre, during, post conditions), the first and last of which were EEG-only blocks. In the second block, tACS stimulation or sham stimulation was applied. Measurements took place at the laboratories of the Experimental Psychology Lab in the Department of Psychology at the University of Oldenburg, Germany, in an electrically shielded room. A 32 channel EEG was recorded according to the international 10–10 System plus right Electrooculogram (EOG). Signals were amplified using a BrainAmp amplifier (Brain Products, Gilching, Germany), digitized at a sampling rate of 1000 Hz. The amplitude digitization range of the amplifier was ± 3.2768 mV with a resolution of 0.1 µV. tACS was delivered using a battery-operated stimulator system (DC-stimulator plus, Neuroconn, Ilmenau, Germany). The visual task was implemented using Presentation software (Version 18.01, Neurobehavioral Systems Inc., Albany, CA, USA). Data preprocessing and outcome variable extraction were conducted with Matlab (Version 9.2.0, The Mathworks Inc, Natick, MA, USA) and the interactive Matlab toolbox eeglab (Delorme and Makeig 2004). Statistical analyses were conducted with Matlab and the R software package (Version 3.3.0, R Foundation for Statistical Computing, Vienna Austria).
The study sample comprised 18 patients (7 females, age: M = 31.3, SD = 9.89 years, ranging from 19 to 57 years of age). Nine patients received tACS stimulation, the other nine received sham stimulation The reader is referred to Table S2 (Online Resource) for all sample characteristics specified by group. All patients were diagnosed with ADHD and were currently undergoing treatment at the specialized outpatient clinic for adult ADHD of the University Hospital for Psychiatry and Psychotherapy at the University Oldenburg, Germany. Diagnosis according to DSM-IV was based on psychiatric expert assessment and was validated using observer rating scales and self-rating scales including the Wender Utah Rating Scale (Retz-Junginger et al. 2002), the ADHD diagnostic checklist (Rösler et al. 2004), the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) and Structured Clinical Interview for DSM-IV Axis II Personality Disorders (SCID-II) (Fydrich et al. 1997; Wittchen and Fydrich 1997). Concurrent use of ADHD medication (reported by nine of the participating patients; agent: methylphenidate) was discontinued at least three days prior to the measurement under the supervision of the medicating doctor. Exclusion criteria comprised left-handedness, metal near brain or skull, epilepsy in medical record or direct family, comorbid neurologic conditions, severe affective disorders and schizophrenia, intake of psychopharmacological medication, substance addiction, autism, dermatosis and pregnancy. Participation was voluntary. Eligible patients, as decided by their medicating doctor, would be contacted by phone and invited to take part, if inclusion criteria were met. Patients received monetary compensation. Written informed consent was obtained from all participants. All investigations were in accordance with the Declaration of Helsinki and the GCP (good clinical practice). The study received ethics committee approval from the medical ethics review committee of the University of Oldenburg.
Procedure and Task
Eligible patients were randomly allocated to groups in a one-to-one ratio using a simple randomization script. The experiment consisted of three blocks (pre, during, post) during which the patients sat in a dark room in front of a computer screen performing a visual oddball task. The task consisted of frequent, irrelevant ‘O’-stimuli (0.75 probability of occurrence), further referred to as standards, and infrequent ‘X’-stimuli (0.25 probability of occurrence), further referred to as targets. Both letters were presented in white on black background (see Fig. 2 for schematic representation). In response to the targets, a button was to be pressed with the right-hand index finger. In between stimuli, the patient was instructed to fixate on a cross symbol displayed in the center of the screen. Stimuli were presented for 1000 ms, inter-stimulus interval (fixation cross displayed) jittered between 1000 and 2000 ms. The pre and post condition included 400 trials (approximately 300 standards and 100 targets) with an average duration of 16.6 min (400 × 2500 ms). The during condition had a fixed duration of 20 min, after which stimulator and visual presentation turned were off automatically. Stimulus size was 7 mm × 7 mm for fixation crosses, and 10 mm × 10 mm for targets and standards. Patients’ distance from the monitor was 1 m.
tACS Configuration and Parameters
In the stim condition 20 min of tACS stimulation were delivered at an intensity of 1 mA (peak-to-peak). 20 min stimulation length was chosen as it was the maximum stimulation length approved by the local ethics committee at that time. The current was ramped up and down over the first and last 10 s of stimulation in order to minimize discomfort. In the sham condition, the procedure and stimulation parameters (see below) were set and determined in the same way as for the stim condition to blind the patient from the experimental condition. In contrast to the stim condition, the stimulation in the sham condition faded in for 10 s and after reaching the amplitude of 1 mA, the signal immediately faded out for another 10 s. All stimulation parameters were manually entered to the device. Therefore, the experimenter could no longer be blinded after the pre condition ended. EEG electrodes were used for stimulation. Before running the stimulation, chosen EEG electrodes were connected to the stimulator (further details below). Patients were then familiarized with the stimulation. The waveform of the applied stimulation was sinusoidal without DC offset.
There are two fundamental aspects that we considered requirements to the successful manipulation of P300 via tACS stimulation: for one, the phase of the oscillatory stimulation current and the phase of the theoretical P300 oscillation should match. Otherwise, stimulation peaks could coincide with P300 oscillation troughs, which would result in cancelling out the P300 wave instead of amplifying it. Secondly, the electrode montage should be designed to direct the current towards the areas of the brain involved in P300 generation.
To meet these requirements, stimulation parameters were determined on a single patient basis. After the pre condition ended, data were briefly preprocessed (low-pass filter: 20 Hz, high-pass filter: 0.5, epoch length: − 3 to 4 s around stimulus onset, baseline removal relative to − 50 to 0 ms interval). In the following, mean P300 latency (in ms, relative to stimulus onset) was estimated. This was done by finding the maximum value of the averaged ERP wave at electrode Pz between 0 and 900 ms after stimulus onset. Visual oddball tasks have a typical P300 latency of 400 ms (Polich and Criado 2006). Accordingly, mean amplitude latency in a pilot study including healthy subjects was 440 ms (Popp et al. 2019). Figure 1 illustrates stimulation frequency estimation. A time–frequency decomposition was computed using a wavelet transform (epoched Pz data: − 3 to 4 s normalized to − 3 to 0 s baseline interval, frequency range: 1.5 to 20 Hz, 3 cycles in each analysis wavelet, frequency resolution: 0.5 Hz, time resolution: 0.024 s). Individual stimulation frequency was the frequency at the maximum event-related spectral perturbation (ERSP) within ± 150 ms time window around individual P300 latency (maximum in time of maxima in frequency). The employed rationale was that P300 ERO would be reflected by the strongest contributing frequency component within the P300 time window (see “Introduction” section). The time interval around P300 latency was introduced in order to account for temporal smearing of the employed time–frequency decomposition method. Resulting stimulation frequencies had a mean of 3.00 Hz and a standard deviation of 1.24 Hz (Mstim = 3.06 Hz, SDStim = 1.40 Hz).
In order to tailor the stimulation individually, estimated P300 latency and ERO frequency were used to manipulate the timing of stimulation. To ensure that P300 latency would coincide with a stimulation peak, the general presentation script was individually adapted as follows. At every zero crossing of the tACS stimulation (at the beginning of each oscillatory cycle) the stimulator sent a trigger to the presentation computer. Before a visual stimulus would be presented on the screen, a wait-trial (wait) was inserted that covered the time until P300 latency and the stimulation peak of tACS would co-occur (see Fig. 3). Since stimulation duration was set to 20 min, this procedure resulted in different numbers of stimulus trials in the during condition (more trials for higher stimulation frequency; resulting mean trial number of 418.7 ± 32.3). To monitor stimulus onset timing throughout the session, a diode channel was attached to the presentation screen. In this way, errors in the script that could have caused undesired time jitters were ruled out.
Stimulation was delivered through a multi-electrode configuration using the Ag/AgCl-ring electrodes that were already attached to the patient’s head. Using EEG electrodes yields a more focal stimulation and allowed us to customize spatial distribution of stimulation (Minhas et al. 2010). Stimulation electrodes were selected based on task-specific P300 topography (Fig. 4a, b) and finite element model (FEM) simulations of current flow (Fig. 4c, d, figure adapted from Popp et al. 2019). The presented FEM simulation was performed on an MNI standard brain and results in an electrical field strength of ~ 0.1 V/m in parietal and temporal cortices, which matches with the pattern of P300 generators (Bledowski et al. 2004). The main goal of the simulation was to ensure that the stimulated brain areas coincide with the targeted areas. The reader is referred to Popp et al. (2019) for the detailed description of the electrode selection procedure. Stimulation electrodes at EEG locations C3, C4, CP3, CP4, P3, and P4 were connected to the anode of our stimulator. Stimulation electrodes at EEG locations T7, T8, TP7, TP8, P7, and P8 were connected to the cathode of our stimulator. In order to achieve a relatively equal current distribution across the six electrodes per output, we assured that the electrode impedance was below 10 kOhm.
The first part of the analysis was carried out during the experiment, in between the pre and during condition. It was performed in order to determine the individual stimulation parameters, namely the stimulation frequency and the peak latency of the stimulation function (see previous paragraph). A high amount of noise impeded the data collected in this ADHD sample. In order to increase the precision of parameter estimation, the time window for P300 latency estimation was changed from [0–900 ms] to [300–600 ms] and a single trial rejection routine was implemented in the online analysis (only), after patient 5. It included the visual inspection of target trials and the rejection of those that showed eye blinks (large deflection, uniform on all channels) in the time interval between 0 to 1 s after stimulus onset. This resulted in a mean trial number of 86.39 ± 5.51 used for parameter estimation. However, ERP waveforms still displayed some noise.
Due to the noise, the following additional preprocessing steps were employed for offline analysis after the completion of the experiment: 8 Hz low pass filter, re-referencing to average reference and independent component analysis (ICA). For ICA, the experimenter visually inspected and removed all components identifiable as artifacts, e.g., cardiac rhythm related, muscular artifacts related to movement or components showing a large amount of high frequency noise (mean number of rejected components for pre: 2.89 ± 1.57 and post: 3.11 ± 2.00). In order to avoid the rejection of components that contribute to P300 generation, components were only excluded if the time course of the component did not appear to have P300 typical activity in it (activity time-locked to stimulus onset peaking within 300 to 600 ms time window). Additionally, after the end of preprocessing an ERP analysis of the rejected ICA components was conducted to ensure that no P300 related activity was excluded from statistical analysis. Furthermore, resolution of the time frequency analysis for investigation of event-related spectral perturbation was increased (frequency range: 0.8 to 10 Hz, frequency resolution: 0.25 Hz, time resolution: 0.014 s).
From the preprocessed data, the following outcome variables were extracted for primary analyses:
P300 amplitude and latency: P300 amplitude was defined as the maximum value [µV] of the averaged ERP waveform at Pz electrode in 300 to 600 ms time window after stimulus onset per condition and patient. P300 latency was its corresponding time point relative to stimulus onset [ms].
Reaction time mean (RT-M) and reaction time variability (RT-V): reaction time (RT) was calculated based on EEG triggers, by subtracting the latencies of button presses and respective target stimulus onset. Means and standard deviations over trials were computed, serving as measures of RT-M and RT-V, respectively. Trials in which RT was longer than 1000 ms or less than 200 ms were judged invalid and were excluded from all analyses.
Errors: two types of errors were investigated. All targets that were not followed by any button press were considered omission errors. Omission error trials were naturally excluded from reaction time analysis (mean number of trials: pre: 88.39 ± 7.37, during: 106.72 ± 17.75, post: 101.00 ± 6.49), while remaining in the EEG data analyses including P300 amplitude and latency estimation (mean number of trials: pre: 91.50 ± 0.71, post: 103.94 ± 0.24; see discussion in the Online Resource). A button press in response to a standard stimulus was classified as false alarm or commission error.
Testing Hypothesis 1: P300 Amplitude Modulation
The main research question of this study was whether P300 amplitude could be increased by the application of tACS in an ADHD patient group. As this hypothesis was directed, one-sided testing was employed. In order to compensate for variance across subjects, we computed the change of P300 amplitude from the pre condition to the post condition relative to the individual pre condition in %, i.e. (amplitude post − amplitude pre)/amplitude pre × 100. This amplitude change was then compared between the stimulation and the sham group. Due to the small sample size, Mann–Whitney test was employed (Shapiro–Wilk test, p > 0.05; see Table S1). All testing was performed at an alpha-level of 0.05.
Testing Hypothesis 2: Behavioral Effects
The secondary research question referred to potential behavioral changes in line with an amplitude modulation of the P300 component due to tACS. To this end, error rates, RT-M and RT-V were tested between experimental groups. The behavioral measures comprised three time points namely pre, during, and post intervention, as the experimental task was also performed during the application of tACS. We hypothesized a decrease in all the mentioned behavioral measures from pre intervention to during intervention and post intervention, which led to left-sided testing. Testing was performed on difference values (e.g., post condition − during condition) in % relative to the respective baseline condition (e.g., during for during-to-post comparison), except for errors, where this would have resulted in dividing by zero in some cases. As hypotheses were not directed for comparisons of during-to-post measures, two-sided tests were employed for the comparison of relative change during-to-post. As normality assumption was not met for at least one of each behavioral variable sets (Shapiro–Wilk test with p < 0.05, e.g. for pre condition RT-V, see Online Resource Table S1) Mann–Whitney tests were employed. Furthermore, due to the factor time comprising 3 time points (pre, during, and post), p-values were corrected for multiple comparisons. For that, FDR correction method by Benjamini and Hochberg (1995) was used, as implemented in the function p.adjust of the R software package.
Testing Underlying Assumptions
To further evaluate our results and test our assumptions several secondary analyses were conducted. The first subset of analyses served to validate the assumptions underlying our hypotheses. These were:
There is a relationship between P300 amplitude and behavioral measures relevant for ADHD symptomatology (subsections: P300 amplitude and behavior).
P300 amplitude is increased via the enhancement of its related ERO in the (stimulated) delta/theta frequency band as represented in the ERSP (subsections: delta/theta activity).
The second subset included additional analyses like of tACS side effects and the comparability of the experimental groups. All analyses employed Mann–Whitney tests, as used for the primary analyses. For relational assessments Spearman correlation was used.
P300 Amplitude and Behavior
To investigate the link between changes in P300 amplitude and behavioral measures as assumed by our hypotheses, a correlational matrix comparing relative changes of the outcome variables from pre-to-post was computed.
In order to reveal potential event-related power changes in the stimulated frequency bands that we attribute to P300 activity, the relative change of event-related spectral perturbation (ERSP) was compared between groups. To this end, ERSP was analysed in the ± 3 Hz frequency and ± 150 ms time window around individual P300 latency and frequency used for stimulation. Subsequently, pre-to-post change, relative to the pre condition, of each patient’s maximum ERSP in the respective time- frequency window was determined, and groups were compared in a Mann–Whitney test comparison.
Parameter Estimation Error
Due to noise corruption and lower frequency resolution, there was an error in parameter estimation in the online analysis as evaluated based on the offline analysis. To consider its potential impact on the results, we investigated its relationship with all outcome variables in this study, for all patients as well as for the experimental groups separately. Here, the phase of the stimulation sinusoid at P300 Latency would be calculated for online parameters and offline parameters. Online parameters comprise P300 latency and frequency as estimated during measurement, whereas offline parameters comprise P300 latency and frequency after noise correction and using higher frequency resolution in time–frequency decomposition. The subtraction of the two phases resulted in the relative phase miss of stimulation which then served as quantification of parameter estimation error. Higher values correspond to a higher phase miss.
Comparability of the experimental groups was tested for all baseline outcome measures as well as for the following demographic characteristics: age, gender, education [in years], medication [yes/no] and ADHD symptom severity as assessed via ADHS-SB questionnaire.
To administer tACS side-effects in this study, patients filled out a questionnaire after the measurement ended (as proposed by Brunoni et al. 2011). The questionnaire further included an item asking patients whether they believed to have been part of the stimulation or sham group (tick item including both groups). This procedure is used to foreclose possible confounds caused by the experimenter or different protocols and is in conformity with the current transcranial electrical stimulation protocol (Nitsche et al. 2008; Nitsche and Paulus 2007). Based on the questionnaire, a qualitative assessment was conducted of the subjective experience of tACS side-effects as well as the identifiability of group membership.
For reasons of completeness, mean P300 latency was analyzed including its relation to reaction times in this sample. However, no tACS effect for P300 mean latency was expected under the given study design. Correlation was calculated for mean data across pre and post conditions.