Abstract
Everyday tasks and goal-directed behavior involve the maintenance and continuous updating of information in working memory (WM). WM gating reflects switches between these two core states. Neurobiological considerations suggest that the catecholaminergic and the GABAergic are likely involved in these dynamics. Both of these neurotransmitter systems likely underlie the effects to auricular transcutaneous vagus nerve stimulation (atVNS). We examine the effects of atVNS on WM gating dynamics and their underlying neurophysiological and neurobiological processes in a randomized crossover study design in healthy humans of both sexes. We show that atVNS specifically modulates WM gate closing and thus specifically modulates neural mechanisms enabling the maintenance of information in WM. WM gate opening processes were not affected. atVNS modulates WM gate closing processes through the modulation of EEG alpha band activity. This was the case for clusters of activity in the EEG signal referring to stimulus information, motor response information, and fractions of information carrying stimulus–response mapping rules during WM gate closing. EEG-beamforming shows that modulations of activity in fronto-polar, orbital, and inferior parietal regions are associated with these effects. The data suggest that these effects are not because of modulations of the catecholaminergic (noradrenaline) system as indicated by lack of modulatory effects in pupil diameter dynamics, in the inter-relation of EEG and pupil diameter dynamics and saliva markers of noradrenaline activity. Considering other findings, it appears that a central effect of atVNS during cognitive processing refers to the stabilization of information in neural circuits, putatively mediated via the GABAergic system.
SIGNIFICANCE STATEMENT Goal-directed behavior depends on how well information in short-term memory can be flexibly updated but also on how well it can be shielded from distraction. These two functions were guarded by a working memory gate. We show how an increasingly popular brain stimulation techniques specifically enhances the ability to close the working memory gate to shield information from distraction. We show what physiological and anatomic aspects underlie these effects.
Introduction
Everyday tasks, such as driving a car or keeping in mind a person's address while being given directions, involve the maintenance and continuous updating of information. Working memory (WM) is a facet of cognitive control and executive functions enabling these processes (Diamond, 2013). The idea of “WM gating” has been coined to illustrate the processes of WM in switching between two core states: maintenance and updating (O'Reilly and Frank, 2006). Only when the gate is open, new information has access to WM and can update its content. When the gate is closed, information can no longer access WM and its content is maintained (Rac-Lubashevsky and Kessler, 2015; Yu et al., 2022b).
From a neuroanatomical and neurobiochemical perspective, the fronto-striatal pathways (Beste et al., 2018) and the dopaminergic system play a crucial role in WM functions (O'Reilly and Frank, 2006; Arnsten, 2011). Yet, also the GABA system in fronto-striatal pathways modulates WM functions, likely via the suppression of task-irrelevant information (Michels et al., 2012; Yoon et al., 2016). GABA enables WM maintenance (Enomoto et al., 2011; Lozano-Soldevilla et al., 2014; Bañuelos and Wołoszynowska-Fraser, 2017; Ragland et al., 2020) and thus a function depending on WM gate closing (Rac-Lubashevsky and Kessler, 2015; Yu et al., 2022b). The GABAergic system, however, is affected by the glutamatergic system and catecholamines, such as noradrenaline (NA) and dopamine (Redgrave et al., 2011). These systems also affect WM functions in the PFC (Robbins and Arnsten, 2009), and especially NA seems to be essential for WM updating (Bays, 2014; Ma et al., 2014). Consequently, WM gate opening functions were reflected by correlates of the noradrenergic system (Yu et al., 2022b). Together, an increased GABAergic and noradrenergic system activity likely plays a role in WM gating and here in WM gate closing and opening processes, respectively. A possible joint causal role of these transmitter systems in WM gating is nevertheless unknown.
One possible way to jointly increase GABAergic and noradrenergic system activity, and thereby probably WM gate opening and closing processes, is to apply auricular transcutaneous vagus nerve stimulation (atVNS). Compared with invasive VNS and cervical tVNS, known to stimulate both efferent and afferent fibers (Clancy et al., 2013; Colzato and Beste, 2020), atVNS is supposed to stimulate only afferent fibers from the auricular branch to the brain, known to be noradrenergic and GABAergic (Colzato and Beste, 2020). Thus, although not always (see Keute et al., 2021), it is suggested that atVNS may enhance GABA/noradrenergic system activity and impact GABA and noradrenergic-driven cognitive functions (Colzato and Beste, 2020). For a review of the main neurotransmitters modulated by atVNS, see Materials and Methods. Consistent with this idea, the administration of atVNS promotes several facets of cognitive control processes known to depend on WM functions (Beste et al., 2016; Fischer et al., 2018; Ridgewell et al., 2021; Konjusha et al., 2022; Wang et al., 2022).
We examine modulatory effects of atVNS on WM gate opening and closing, using a validated experimental approach that is able to differentiate between WM gate opening and closing in combination with EEG recordings (Rac-Lubashevsky and Kessler, 2015; Yu et al., 2022b). We focus on theta and alpha band dynamics. The theta band activity is increased during WM gating processes (Rempel et al., 2021) with modulations in that activity being related to measures of noradrenergic system activity (i.e., pupil diameter) (Yu et al., 2022b). Alpha band activity is increased during WM maintenance (Bonnefond and Jensen, 2012; Wianda and Ross, 2019), and this is related to the GABAergic system (Lozano-Soldevilla et al., 2014). Therefore, we examine modulations in the theta/α frequency bands and also examine which functional neuroanatomical structures reflect modulations in these frequencies using EEG-beamforming. Previous findings using the same paradigm showed that WM gating affects different aspects of information (i.e., stimulus information, information reflecting stimulus–response translation processes, and motor information) that are concomitantly coded in the EEG (Rempel et al., 2021; Yu et al., 2022b). To examine whether atVNS-induced modulations in theta/alpha band activity affect all these coding levels, we applied a similar analytical pipeline to the previous studies using residue iteration decomposition (RIDE) (Ouyang et al., 2015). To objectify possible effects on the noradrenergic system, we examine salivary α-amylase (sAA) and pupil diameter modulations during the task.
Materials and Methods
Participants
N = 43 healthy participants took part in the current study. N = 6 participants were excluded because of either poor performance (accuracy <50%) and/or bad signal in EEG (void electrode channels during the preprocessing step) and/or insufficient number of trials in EEG and/or eye tracking data (<10 valid trials detected) in at least one task condition on at least one of the appointments (for details, see below). Thus, N = 37 participants were included in the analysis. Participants were aged between 18 and 31 years (mean age 25.09 ± 3.60 years) and 19 of them were males. In determining the sample size for our study, we used a sample size comparable to that of previous studies (Mertens et al., 2020; Konjusha et al., 2022), thereby ensuring that the sample has sufficient statistical power and is comparable to previous studies. All participants had a normal or corrected-to-normal vision. The inclusion criteria involved the following: (1) age; (2) no history of neurologic or psychiatric disorders; (3) no history of substance abuse or dependence; (4) no history of brain surgery, tumor, or intracranial metal implantation; (5) no psychoactive medications; (6) no pregnancy; (7) no susceptibility to seizures or migraine; and (8) no pacemaker or other implanted devices. Participation was granted only after participants were informed about their participation rights and signed the consent form. Upon completing the experiment, they received a financial reimbursement of 50 euros. This research was conducted in accordance with the Ethics Committee of the Medical Faculty of the TU Dresden (EK 369072019) and was conducted following the Declaration of Helsinki.
Task
A modified version of the Reference Back Task (Rac-Lubashevsky and Kessler, 2016) was applied to assess the effects of atVNS on gating mechanisms of WM processes (Rempel et al., 2021; Yu et al., 2022b,c). This task enables the empirical differentiation between the WM subprocesses, more precisely gate opening, gate closing, and updating. A schematic representation of the reference task is shown in Fig. 1A).
A, Graphic illustration of the reference back task including the description of the stimulus. The trials, which include letters within a red frame, are reference trials; trials within a blue frame are comparison trials (for a more detailed description, see text). The response buttons indicating “different” or “same” are denoted with the left or right control buttons, respectively. When the letters are identical, the right “Ctrl” button should be pressed; if not, the left “Ctrl” button should be pressed with the corresponding index finger. Participants have to constantly store the previous red-framed letter in WM and update it when a new red-framed letter appears. B, A schematic outline of the calculation of WM gate opening and closing behavioral processes from the reference-back task. The titles “switch” and “no switch” are used to describe the repetition or the alternation of the trial type, more specifically comparison versus reference. Gate opening is calculated as the difference between reference trials that were preceded by comparison trials (switch trials) and reference trials preceded by reference trials (i.e., no switch trials). In contrast, gate closing is calculated as the difference between comparison trials that were preceded by reference trials (switch trials) and comparison trials that are preceded by comparison trials (no switch trials).
A sequence of capital letters either “X” or “O” is presented in the middle of the screen. In this task, there are two types of trials: comparison trials, which demand a matching decision; and reference trials, which demand both matching and updating. In contrast to the classical version of the n-back task, in the reference back task, the capital letter is framed with either a red or a blue square. The color of the frame indicates the type of trial: the letters framed with a red square are reference trials, while the letters framed with a blue square are comparison trials. The blue-framed letters served as comparison trials, meaning that they could only be used as a comparison with the last previously shown red-framed letter. Therefore, the stimuli presented in a red frame are references, to which the following trials are compared. Thus, these are defined as reference trials given that they require the update of WM. Although the stimuli presented in a blue frame, defined as comparison trials, demand a match to the preceding red frame reference, they do not require the update of WM. Yet, both trial types (reference and comparison) require a same/different response. It is noted that gate opening process only occurs in the first updating trial in a row (specifically, when switching from comparison trials), rather than as part of every updating operation.
It has been suggested that the control over WM content is achieved by a gating mechanism, which can be either open or closed (Rac-Lubashevsky and Kessler, 2016). When confronted with relevant and useful information, the gate opens and facilitates flexible updating of WM, whereas, when confronted with distracting or irrelevant information, gate closing serves as a maintenance of current information by preventing this information to enter WM. The reference back task is considered to be a valuable tool in the measurement of the gate opening and closing. Indeed, compared with the classical version of the n-back task, this task has “the baseline for comparison”: trials in which a comparison to preceding trials in WM is to be executed without WM updating (Rac-Lubashevsky and Kessler, 2018). Consequently, this allows the identification of the contribution of additional processes, including gating and comparison process. Studies have shown that shifting between updating and maintenance or vice versa, is associated with increased reaction times (RTs) (Kessler and Oberauer, 2013; Rac-Lubashevsky and Kessler, 2015, 2016). The trials were categorized into eight conditions according to three features: reference/comparison, switch/no-switch, and match/mismatch. According to previous work (Rac-Lubashevsky and Kessler, 2016; Yu et al., 2022b), gate opening and gate closing indices were calculated as follows:
Gate opening = (switch_match_reference + switch_mismatch_reference) − (no-switch_match_reference+ no-switch_mismatch_reference) (formula 1)
Gate closing = (switch_match_comparison + switch_mismatch_comparison) − (no-switch_match_comparison + no-switch_mismatch_comparison) (formula 2)
The participants completed the task on a 19-inch computer screen and a standard keyboard with a QWERTZ layout was used. They were instructed to place their left index finger on the left control key and their right index finger on the right control key. Each trial began with the presentation of a fixation cross for 800 ms; this was jittered between 600 and 1200 ms. Then the stimulus (“X” or “O” in either red or blue frame) appeared for 1400 ms or until a response was made. Afterward, the screen turned blank for 1000 ms. A thorough explanation and a 30-trial exercise were conducted before the actual experiment to familiarize participants with the reference back task. The task consisted of 75% trial type, which were similar between trials (no-switch) and 25% alternated (switch). The actual task had 16 blocks each consisting of 60 trials, equaling in total 960 trials and lasted ∼1 h, with 30% of them being switch trials. Regardless of the participants response (correct, incorrect, none), the stimulus presentation ended after a button press or in case of no response after 1400 ms. Each block began with a reference trial which does not require any response. A balanced distribution of the frame color and the necessary response was applied. Although the order in which the stimuli were presented was random, it was the same for every participant.
Study design
A sham-controlled crossover design was used in the present study. Participants were unaware of the kind of stimulation and the experimental procedure. The experiment took place twice, with ∼1 week or more between the appointments. The participants were divided in two halves: one half received active atVNS stimulation at the initial appointment and sham stimulation at the following appointment; the other half received sham stimulation at the initial appointment and active atVNS stimulation at the following appointment. Order of stimulation condition (atVNS-sham; sham-atVNS) was applied in a counterbalanced manner across all participants. The stimulation was applied ∼20 min before the start of the experiment following other studies (Beste et al., 2016; Konjusha et al., 2022). Participants received continuous stimulation throughout the experiment (see below). After each appointment, participants filled out an atVNS aversive effects questionnaire, which included a 5-point scale rating of various possible side effects, such as the following: headache, neck pain, nausea, muscle contraction in face and/or neck, stinging sensation under the electrodes, general uncomfortable feelings, and other unwanted sensations and/or adverse effects. Generally, none of the participants reported significant complaints or discomfort during or after atVNS stimulation (see Results). The nested study design implied different blocks (up to 16) in two sessions (Session 1 and Session 2). Previous studies have revealed that WM gating processes are prone to time on task effects (Yu et al., 2022b). Moreover, participants underwent two sessions which may have led to systematic difference in Session 2 because of learning effects. In an extra conducted data analysis (see Results), we therefore examined the session effect. It is shown that a session effect does not affect the pattern of results.
atVNS
The auricular branch of the vagus nerve was stimulated using the atVNS nextGen research device (tVNS Technologies). Based on previous research (Ellrich, 2011; Fischer et al., 2018), the stimulation intensity was adjusted individually for each participant; that is, it was adjusted above the detection threshold and below the pain threshold. In each appointment, participants received increasing and decreasing series of 10 s stimulation trials. For each trial, they had to rate the subjective sensation of the stimulation in a 10-point scale, starting from nothing (0), light tingling (3), strong tingling (6) to unpleasantly painful (10). The increasing series of trials started from 0.1 mA and increased gradually by 0.2 mA on a trial-by-trial basis. This was done up to the point to which the participants would report a tingling sensation (8). Next, the procedure was repeated similarly but with the decreasing series of trials starting from their sensation value of 8 until they felt a subjective sensation of 6 or lower. This whole procedure was done twice, and the final stimulation intensity applied for the experiment was calculated based on the average of the four intensity rates chosen by each participant: more specifically, two from the increasing and two from the decreasing series of trials. In our study, the only distinction between active atVNS stimulation and sham condition was the location of electrode placement. For the sham condition, the electrode was placed in the earlobe, which is not known to contain vagal afferent nerves, whereas for the active stimulation, the electrode was placed in the cymba conchae, which is considered the optimal stimulation point. For both conditions, the stimulator was activated to ensure that the participants could hardly distinguish the active atVNS stimulation from the sham condition. According to the recent consensus on atVNS stimulation parameters (Farmer et al., 2021), the stimulation was delivered continuously with a pulse width of 200-300 ms at 25 Hz, and the electrode was positioned only on the left ear for both active atVNS and sham condition. The application of continuous stimulation was based on previous research that used similar stimulation parameters, presuming that the systematic activation of the noradrenergic system might be conducted via continuous atVNS stimulation (Fischer et al., 2018; Ventura-Bort et al., 2018). Several lines of evidence from anatomic and clinical studies have revealed that atVNS is safer when placed in the left and not right ear to prevent possible cardiac side effects (Nemeroff et al., 2006; Kreuzer et al., 2012). The average stimulation intensity used for the active atVNS stimulation was 2.4 mA and 3.0 mA for the sham stimulation. The stimulation intensity differed between the conditions (t(36) =−3.41, p = 0.002; d = 1.0). Importantly, however, since the blinding of participants was successful and there was no difference in side effects perceived between the conditions (see Results), differences in stimulation intensities do not distort the interpretation of the findings. Studies indicate that the earlobe is free from vagal afferent fibers (Peuker and Filler, 2002; Kraus et al., 2013; Colzato and Beste, 2020), and possibly not producing any activation in the cortex or brain stem (Frangos et al., 2015). In the present study, atVNS was initiated 20 min before the commencement of the experiment, while the participants were being prepared for EEG recording. Participants then completed an additional task, unrelated to WM, for ∼20 min, more specifically the Stop Change Task. Regarding the duration of atVNS aftereffects, there has been limited research conducted to date. An fMRI study found that the activation of the nucleus of the solitary tract (NTS) persisted for 11 min after atVNS stimulation and the neural activation gradually returned to baseline (Frangos et al., 2015).
Neurotransmitter systems likely affected by atVNS
Transcutaneous vagus nerve stimulation (atVNS) has been proposed as a means of influencing the activity of several neurotransmitters, with the most studied among them being GABA and NA. Although, to date, the mechanisms of action of atVNS are not fully understood, it is argued that one of the mechanisms of action involved via atVNS is the increase of GABA transmission in the brain, which is thought to occur because of the activation of the NTS (Colzato and Beste, 2020). Previous research has indicated that invasive VNS in epilepsy patients seems to lead to an increase in the levels of extrasynaptic GABA concentration and its receptor density (Ben-Menachem et al., 1995; Marrosu et al., 2003). Recent studies using fMRI seem to support the idea that atVNS can enhance the release of GABA in the brain by activating the NTS via the stimulation of afferent fibers (Dietrich et al., 2008; Hein et al., 2013; Frangos et al., 2015; Yakunina et al., 2017). Additionally, atVNS has been found to significantly enhance the short-interval intracortical inhibition, a measure of GABAA activity, in the right motor cortex (Capone et al., 2015). Supporting this, Keute et al. (2018) demonstrated behavioral and electrophysiological effects of atVNS, which were explained with the modulation of GABA transmission in the motor cortex. Contrary to these findings, one study did not find any increase of GABA concentration; and as such, no GABA-associated MEG modulations in response to atVNS (Keute et al., 2021). Moreover, the aforementioned studies (Capone et al., 2015; Keute et al., 2018) show opposite GABAergic mechanisms, where in the first one the results are more compatible with a GABA increase and in the second one with a GABA decrease, which leads to more questions. Research investigating the effects of atVNS on GABA remains sparse, and the results from previous studies need to be considered cautiously, particularly the findings from epilepsy patients since the usual inhibitory effect of GABA transmission in epileptic brains is considered to be pathologic because of an imbalance in the levels of GABA and other neurotransmitters (Sarlo and Holton, 2021). Despite the somewhat inconsistent findings, further research investigating the effect of atVNS on GABA is needed. The main challenge seems to be the difficulty in measuring GABA. To date, there are no reliable “online” indirect markers that can show the levels of GABA in the brain during tVNS (Colzato and Beste, 2020). However, by using methods, such as MRS and PET, future studies might be able to indirectly measure the modulation of GABA because of atVNS and lead to more conclusive findings.
Much of the current literature pays more attention to the atVNS modulation of NA, which is another neurotransmitter involved in the atVNS effects through the activation of the locus coeruleus-NA system (LC-NA), which is considered to be the main noradrenergic source of the brain (Aston-Jones and Waterhouse, 2016). The underlying mechanisms involve vagus nerve sending information about the adrenergic release from the adrenal gland to the brain, and from there it projects to the NTS (Wan et al., 2008), followed by the NTS sending excitatory projections to nucleus paragigantocellularis, which is linked to the LC and from there projects to many other brain regions modulating behavior through phasic and tonic firing (Schwarz and Luo, 2015; Butt et al., 2020). The involvement of vagal influence on the LC-NA system is implicated in various studies. Animal studies have robustly shown the effects of VNS on firing rates of LC neurons compared with sham (Chen and Williams, 2012; Hulsey et al., 2017). Additionally, reduced LC firing rates have been evident after vagotomy (Svensson and Thorén, 1979). fMRI studies in humans have also shown an increased activation in the brainstem region, including the LC after vagus nerve stimulation (Dietrich et al., 2008; Frangos et al., 2015; Yakunina et al., 2017). Nevertheless, other studies using noninvasive markers of NA, such as pupil size and P3b magnitude, have not revealed a reliable increase in these markers using different intensities of atVNS (Keute et al., 2019b; Warren et al., 2019; Burger et al., 2020b). On the other hand, evidence from epileptic patients has shown increased resting pupil diameter (Desbeaumes Jodoin et al., 2015) and an increase in P3 amplitude (De Taeye et al., 2014) after invasive VNS. Other research conducted by Sharon et al. (2021), which used an individually adjusted atVNS stimulation, implemented a 3 s stimulation on and 30 s stimulation off pattern; and it was determined that atVNS significantly enhanced phasic pupil dilation compared with the sham stimulation. To date, the evidence for effects of atVNS on pupil size remains thus rather inconclusive. Nevertheless, other indirect markers of central NA activation, such as sAA, in some studies was increased after active atVNS stimulation compared with sham condition (Fischer et al., 2018; Ventura-Bort et al., 2018; Warren et al., 2019). This has also been supported from a pooled mega-analysis reporting that vagal activation via atVNS increases sAA levels compared with sham stimulation (Giraudier et al., 2022), indicating that atVNS triggers NA release. Nevertheless, in other studies, no such increase has been reported (Giraudier et al., 2020; D'Agostini et al., 2021). It should be noted that the use of stimulated saliva for the assessment of sAA as a noradrenergic marker has been questioned because of the increase of noise and the interindividual differences in the mastication process, which in turn might affect the involvement of the parotid glands and the sAA level determination (Bosch et al., 2011). Therefore, sAA might be suitable to be considered only as a somewhat reliable marker of NA (Burger et al., 2020a). In summary, the existing research on the topic has produced relatively inconsistent results. This discrepancy might be attributed to a number of factors, including the methodological differences, the diversity of stimulation parameters, voltage, durations, pulse widths, and control sites used across studies. Although an attempt has been made (see Farmer et al., 2021), there is currently a lack of consensus in the field regarding the optimal stimulation parameters, which makes it challenging to compare results and achieve conclusive remarks. As such, the findings should be approached with caution, and additional research is necessary to gain a more comprehensive understanding of the relationship between atVNS and its associated neurotransmitters.
EEG recording and processing
High-density EEG data were recorded via 60 equally spaced Ag/AgCL electrodes on a cap, where the ground electrode coordinates were theta = 58, ϕ = 78 and reference electrode coordinates were at theta = 90, ϕ = 90. The electrode impedances were <5 kΩ. In order to record the EEG data during the task, we used the BrainVision Recorder program (Brain Products) with a sampling rate of 500 Hz. After recording, the EEG signal was downsampled offline to 256 Hz. The preprocessing of the EEG data was done by using the “automagic” pipeline (Pedroni et al., 2019) and EEGLAB (Delorme and Makeig, 2004) on MATLAB 2019a (The MathWorks). The first step involved flat channel removal and the rereferencing of the EEG data to an average reference. Next, the PREP preprocessing pipeline was run (Bigdely-Shamlo et al., 2015) followed by the EEGLAB clean_rawdata() pipeline. The PREP preprocessing pipeline involves the removal of line noise at 50 Hz via a multitaper algorithm, and it applies a robust average reference (after the removal of contaminations by bad channels). The EEGLAB clean_rawdata() pipeline initially detrends the EEG data using an FIR high-pass filter of 0.5 Hz (stop band attenuation 80 dB, order 1286, transition band 0.25-0.75 Hz). Afterward, the flat line, noisy, and/or outlier channels are identified and removed. Too many bad channels cannot reliably be interpolated. Therefore, we chose the more conservative way and excluded participants with too many bad channels to have a high-quality dataset. Not only the bad quality EEG data, but also the behavioral data were discarded from further analysis. The reason is simple: EEG/neurophysiological evidence was extracted from EEG data to interpret behavioral performance on a neural level. This makes only sense when the same set of trials is used for the behavioral and neurophysiological data analysis. The reconstruction of epochs with extraordinary strong power (i.e., >15 SDs relative to calibration data) was done using the Artifact Subspace Reconstruction (burst criterion: 15) (Mullen et al., 2013). Time windows that could not be reconstructed were discarded from further analyses steps. Then a low pass filter of 40 Hz was applied (Sinc FIR filter:order:86) (Widmann et al., 2015). The subtraction method was used to remove the electro-oculographic artifacts (Parra et al., 2005). The independent component analysis using the Multiple Artifact Rejection Algorithm (Winkler et al., 2011, 2014) was used to remove other artifacts, such as muscle and/or remaining eye artifacts. The components that contained cardiac artifacts, such as pulse, were also identified and eliminated (Pion-Tonachini et al., 2019). Afterward, the missing or eliminated channels were interpolated using the spherical method. Next, a manual visual inspection was conducted as a quality check and to remove possible residual epochs with bad data. The EEG data were then epoched into 960 trials and locked into stimulus. Only trials with correct responses within a response time window (1400 ms) were included in the analysis. Each epoch had a length of 4000 ms, starting 1000 ms before the stimulus until 3000 ms after stimulus onset. The total number of trials in each session with correct responses and without artifacts were categorized into the eight conditions described in the task section above. To ensure a reasonable signal-noise ratio for further analysis, participants with <10 trials in any of the eight conditions were excluded from all data types.
Residue iteration decomposition (RIDE)
RIDE allows one to identify distinct aspects of information concomitantly processes in the neurophysiological signal in a conceptually relevant manner (Mückschel et al., 2017a; Opitz et al., 2020; Pscherer et al., 2020; Yu et al., 2023). Previous findings have shown that, during WM gating, different aspects of concomitantly coded information are affected by WM gate opening and closing processes. For that reason, and to increases comparability across studies using the same paradigm to disentangle WM gating processes in EEG data (e.g., Rempel et al., 2021; Yu et al., 2022b), we used RIDE decomposition in the current study.
RIDE was used to process the segmented single-trial EEG data by means of the RIDE toolbox in MATLAB (Ouyang et al., 2011). Based on the idea that distinct components with variable or static latencies can be differentiated within ERPs, RIDE seeks to decompose single-trial ERPs into various components with variable or static latencies. These component clusters can be associated with specific stages of information processing depending on the timing and variability. RIDE is especially sensitive to the channel-specific latency variability information since it is applied for each electrode individually. RIDE derives three clusters: S-cluster, R-cluster, and C-cluster. S = (“stimulus”) and R = (“response”) clusters are determined based on the latency information associated with stimulus and response onsets, respectively. The C = (“central”) cluster's latency information is calculated in every single trial, and it is iteratively improved using L1-norm minimization algorithm. Consequently, C-cluster's latency is initially estimated in every single trial while reflecting some global waveform (Verleger et al., 2014). For the estimation of the S-cluster RIDE subtracts C- and R-cluster from every trial and aligns the residual of all trials to the latency information of S-cluster. Therefore, the outcome is the median waveform for each time point in the S-cluster interval. The C- and R-clusters are derived in a similar way. For a thorough explanation of the RIDE procedure, see Ouyang et al. (2015). To use RIDE, it is necessary to specify the time frames to extract the waveform of each component. These time frames should encompass the range within which each component is supposed to occur. The current study used the following time frames: for S-cluster, it was from −200 ms before the target and 900 ms after the target; for the C-cluster, it was 200-900 ms after the target; and for the R-cluster, it was ±300 ms around the response trigger. These parameter settings were used in previous studies with the same experimental paradigm (Rempel et al., 2021; Yu et al., 2022b).
Time-frequency decomposition
The time-frequency analysis was used to examine atVNS stimulation effects on theta band (4-7 Hz) and alpha band (8-12 Hz) activity for WM gating processes. To compute the time-frequency powers for each WM gating process for each participant, we first decomposed the time-frequency powers in the single-trial level for each RIDE cluster (S, C, R) and each condition and session (atVNS and sham). The time-frequency decomposition was performed in a frequency range of 4-12 Hz (i.e., spanning from theta to α bands) with a step of 0.5 Hz. Morlet wavelets with a width of 5 cycles were used. For each RIDE cluster and condition, the decomposed time-frequency representation was averaged across trials and then normalized through decibel (dB) conversion using baseline activities between 200 and 0 ms relative to the stimulus onset. After that, the time-frequency representations of WM gating processes were computed for each participant using Formula 1 and Formula 2 (described above). For further respective statistics for α and theta bands, we averaged the time-frequency decompositions across respective frequency bands (i.e., 4-7 Hz for theta band and 8-12 Hz for alpha band). Two steps of cluster-based permutation tests were conducted to examine differences between active atVNS stimulation and sham stimulation in gate opening and gate closing for each of theta and α bands and for each RIDE clusters using above-processed time-frequency representations. The first cluster-based permutation tests aimed to extract the time window showing significant difference between active atVNS stimulation and sham stimulation. Based on the extracted time window, we selected and averaged corresponding α or theta band activities across this time window, which was later used for the second step of the cluster-based permutation tests aiming to extract the electrodes showing significant difference between sham and atVNS. The reference distribution of the permutation test relied on dependent t test results for (each time point for the first step and) electrode (for the first and second step) and estimated using the Monte–Carlo method with 1000 random draws. The minimum number of electrodes forming a cluster was 1. Cluster-based permutation tests were run via the FieldTrip toolbox (Oostenveld et al., 2003) (α = 0.05).
Several previous studies corroborate the soundness of combining with time-frequency decomposition methods (and beamforming) with RIDE decomposition (Mückschel et al., 2017a; Pscherer et al., 2020; Yu et al., 2023), including the same experimental paradigm that is used in this study (Rempel et al., 2021; Yu et al., 2022b), also for atVNS effects on cognitive control processes (Konjusha et al., 2022). To validate the RIDE approach in the context of this study, we analyzed EEG data without RIDE by following the same pipeline and parameter setting (see Results).
Beamforming analysis
In order to reconstruct the sources and identify the neuroanatomical areas related to the differences between atVNS and sham, a dynamical imaging of coherent sources was performed (Gross et al., 2001). This was only implemented for WM gate closing conditions and for the alpha band frequency (8-12 Hz) according to the cluster-based permutation tests on time-frequency analysis (see Results). Beamforming analysis was done for each of RIDE clusters separately following these steps: first, α powers and the cross-spectral density matrix of gate closing-related conditions from 1300 to 1800 ms, as well as −500 to 0 ms relative to the stimulus onset as a baseline time window were extracted for each participant using a single Hanning taper frequency transformation. The time window of 500 ms for the analysis was selected based on the cluster-based permutation test from above that the significant difference between atVNS and sham was centered at ∼1550 ms relative to the stimulus onset for all RIDE clusters. A 500 ms interval involves at least four cycles for the alpha band to be sufficient to extract alpha power. After that, dynamical imaging of coherent sources constructed a spatially adaptive common filter using the above-mentioned cross-spectral densities of baseline and activity of interest from all gate closing conditions for each participant. This common filter (regularization filter: 5%) was then applied to the baseline and activities of interest in individual conditions to estimate the source. A 10 mm resolution 3D grid representing the brain is created. Additionally, the source power for each grid point is calculated using the “colin27” MNI brain template and the “standard_bem “forward model by Oostenveld et al. (2003). After that, the source power estimation of each condition was first decibel-normalized with their respective baseline interval from −500 to 0 ms relative to the stimulus onset. The normalized source power estimation was computed for each condition according to the following formula (toi = time period of interest):
Relying on the decibel power of each condition, the source power of gate closing for each session was calculated with Formula 1 (see above). Afterward, the grand-average gate closing α powers were computed by averaging across participants for atVNS and sham separately. Based on the cluster-based permutation test for sensor level data that revealed the α activities of atVNS as higher than sham condition for WM gate closing. Therefore, we selected the 1% voxels showing highest positive contrast between atVNS and sham in cortical regions to identify the anatomic regions of atVNS effect on gate closing using the “DBSCAN” algorithm (Ester et al., 1996), according to previous studies (Adelhöfer et al., 2020). After that, a linearly constrained minimum variance (LCMV) beamformer (Van Veen et al., 1997) was run to reconstruct the time series of activities from the delineated functional neuroanatomical regions for corresponding RIDE clusters. Based on the DBSCAN results (see Results), anatomic sources of each RIDE cluster could be roughly divided into two large clusters formed by voxels in the frontal cortex and the temporal cortex, respectively. Thus, LCMV was performed on the two regions separately for each gate closing condition, RIDE cluster, and session at the single-subject level. First, a covariance matrix was calculated from corresponding RIDE-decomposed single trials to create a spatial filter, which was then applied on the RIDE-decomposed trials to reconstruct the time series of each voxel in the corresponding anatomic regions of the current RIDE cluster. The average time series across voxels was computed as the reconstructed time series and further decomposed to time-frequency representations using Morlet wavelets. The process of time-frequency decomposition was the same as at the sensor level. We calculated the reconstructed task-related alpha band activity time course of gate closing for each session and RIDE clusters using the same methods as previously. We applied the same beamforming analysis on EEG data without RIDE (see Results).
sAA measurements
Noradrenergic activity can be measured using sAA (Chatterton et al., 1996). Increased sAA levels in the active atVNS stimulation condition compared with the sham condition have been revealed (Fischer et al., 2018; Ventura-Bort et al., 2018; Warren et al., 2019). Nevertheless, there are other studies that show no increase of sAA levels after stimulation (D'Agostini et al., 2021, 2022), but a recent mega-analysis revealed that sAA might be a promising indirect marker of LC-NA system (Giraudier et al., 2022). We collected sAA with saliva samples at four time points using regular Cortisol Salivette (Sarstedt) sampling probes with the swab. sAA was collected before and after both active atVNS and sham stimulation conditions: before active atVNS stimulation (pre), after active atVNS stimulation (post), before sham stimulation (pre-sham), and after sham stimulation (post-sham). The participants were instructed to refrain from drinking any beverages and/or eating at least 15 min before sample collection. Furthermore, they were asked to lightly chew the swab to stimulate saliva flow for ∼30-60 s. After the samples were collected, they were stored at −20°C freezers to ensure their preservation. The concentration of AA in saliva was indexed following a published protocol (Rohleder et al., 2006) using an enzyme kinetic method, and saliva was analyzed on a Genesis RSP8/150 liquid handling system (Tecan). Because of technical problems, samples from 2 participants were excluded from the analyses of the sAA measures.
Pupil diameter analysis
For the purpose of recording and processing pupil diameter, an eye tracking device was used (RED 500 eye tracker). The eye tracker device was positioned underneath the computer monitor at an ∼60 cm distance from the participants. The program iView X (SensoMotoric Instruments) was used to record pupil diameter data. The sampling rate of 256 Hz was used. Before starting with the task, the eye tracker device was calibrated via a 9-point calibration procedure. Afterward, when the task began, eye movements were recorded with the EEG data at the same time. Using the EYE-EEG extension of the EEGLAB, raw pupil diameter measurements were synced with the raw EEG data after the recording using the same start and end markers in both datasets. The syncing process enabled the alignment of the pupil diameter measurements with all the markers in the EEG data. In the same manner as the EEG data, segmentation and baseline correction were also conducted. Baseline-corrected pupil diameter data were correlated (Pearson correlation) with the beamformed source-level RIDE data (i.e., the time course of alpha band activity in each source) at the single-subject level for each condition and RIDE cluster. The pupil diameter time series were chosen from 0 to 2 s. Previous studies have shown that the used sample size in the current study is sufficient to obtained reliable effects in these analyses, and the sample size is comparable to previous studies by our group (Chmielewski et al., 2017; Mückschel et al., 2017b; Yu et al., 2022a). The correlation with pupil diameter was also conducted for EEG data without RIDE (see Results).
Behavioral and sAA data statistical analysis
The IBM SPSS Statistics version 28.0.0.0 was used for the behavioral and physiological data statistical analyses. In order to test the effects of stimulation on salivary levels, a repeated-measures ANOVA with within-subject factors stimulation (active atVNS stimulation vs sham stimulation) and time (pre vs post) was conducted. The statistical analysis of the behavioral data, including RT and accuracy, was calculated for each session, condition, and participant. Only trials with accurate responses were used for the analysis. Separate ANOVAs were conducted with RTs and accuracy as dependent variables. ANOVAs were run using the following within subject factors: “stimulation” (active stimulation vs sham stimulation) and “gating” (opening vs closing). For all repeated-measures ANOVAs, Greenhouse–Geisser and Bonferroni correction were applied when necessary. For all descriptive statistics, the mean and the SEM are reported. Additionally, we conducted Bayesian analyses to substantiate whether the null hypothesis was more likely than the alternative hypothesis for sAA effects. Simply put, this method returns the likelihood of the H0 and H1, given the obtained data. It is argued that values between 1 and 3 are regarded as anecdotal evidence for the null hypothesis, values 3-10 are regarded as substantial evidence, values 10-30 as strong evidence, values 30-100 as very strong evidence, and values >100 as extreme evidence for H0 (Wagenmakers et al., 2011). To investigate the association between sAA and behavioral performance and particularly whether it was affected by atVNS, we conducted a Person's correlation analysis on sAA and behavioral parameters. Because of the two missing sAA samples, we removed the corresponding participants from the behavioral datasets. Afterward, we correlated each sAA level (i.e., pre-stimulation, post-stimulation, and the difference between pre- and post-stimulation) and the behavioral parameters (i.e., accuracy and RT) for each WM gating condition (i.e., gate opening and closing) for each stimulation condition (i.e., atVNS and sham). A total of 24 correlation coefficients were obtained. Additionally, to reduce the potential interference of the skewness of sAA, we log-transformed the sAA data as suggested by the previous study (Giraudier et al., 2022) and implemented the same correlation analysis subsequently.
Data and code availability statement
The data of the current study are available from OSF: https://osf.io/c2s4b/.
Results
Subjective reports on atVNS effects
The subjective reported side effects from the participants were analyzed with paired-samples t tests for each side effect separately. Mean subjective ratings for the stimulation side effects in the active atVNS stimulation condition and the sham stimulation condition are presented in Table 1. Statistical analyses of subjective ratings revealed no significant difference between the active atVNS stimulation condition and the sham stimulation condition in none of the symptoms examined (p > 0.05), indicating that the stimulation side effects were minimal comparable for both conditions. Participants were also asked to report their insight in which appointment they believed they were under active atVNS stimulation. It is revealed that participants' guesses did not differ from chance level (χ2(1) = 0.243, p = 0.622), thereby showing successful blinding in the study.
Mean (SE) subjective ratings for the stimulation side effects in the active atVNS stimulation condition and the sham stimulation conditiona
Behavioral and neurophysiological task-related data
The behavioral data are shown in Figure 2. The repeated-measures ANOVA for RTs revealed a significant main effect of “gating” (F(1,36) = 9.639; p = 0.004; ηp2 = 0.211; BF01 = 1.000), as participants revealed faster RTs in gate opening (77.827 ± 9.08 ms) than in gate closing (123.847 ± 11.58 ms). No other main effects or interaction effects were revealed (all F ≤ 0.499; p ≥ 0.485; ηp2 ≤ 0.014). The repeated-measures ANOVA for the accuracy revealed a significant main effect of “gating” (F(1,36) = 18.535; p < 0.001; ηp2 = 0.340; BF01 = 7.084), as participants revealed higher accuracy rates in gate opening (−6.4 ± 1.1%) than in gate closing (0.9 ± 1.1%). Furthermore, there was a significant interaction effect of stimulation × gating (F(1,36) = 10.529; p = 0.001; ηp2 = 0.226; BF01 = 29.282). Therefore, we conducted post hoc t tests to investigate the interaction. The post hoc paired t tests revealed that there was a significant difference in accuracy for gate closing in the active atVNS stimulation session (2.6 ± 1.1%) and the sham stimulation session (−0.78 ± 1.3%) (t(36) = 2.47; p = 0.009; d = 0.4; BF01 = 0.399). In contrast, the post hoc paired t tests revealed no significant difference in accuracy for gate opening in the active atVNS stimulation session (−7.5 ± 1.2%) and the sham stimulation (−5.29 ± 1.3%) (t(36) = −1.61; p = 0.057; d = −0.2, BF01 = 0.579). We conducted subsequent t tests to determine whether there was a difference between gate closing (−0.0078) and gate opening (−0.0529) in the sham condition (t(36) = −2.64; p = 0.012; d = 0.1; BF01 = 3.603). Additionally, there was a difference between gate closing (0.0260) and gate opening (−0.0750) in the active atVNS condition (t(36) = −4.84; p < 0.001; d = 0.1; BF01 = 908.496). Thus, our results showing a higher accuracy in the gate closing than in the gate opening in both conditions is confirmed, in line with many previous studies (Rac-Lubashevsky et al., 2017; Rempel et al., 2021; Yu et al., 2022a).
Behavioral performance for the reference back task. A, The gating effects (closing vs opening) for accuracy for each stimulation condition. B, The gating effects (closing vs opening) for RTs for each stimulation condition. In both plots, bar plots represent the mean value of the corresponding condition. Error bars indicate SE. Points represent individual performance.
Possible session effects and considerations about time on task
We ran additional behavioral analyses adding “order of stimulation” as a between-subject factor to examine the effects of stimulation order whether it interacts with the other factors. The repeated-measures ANOVA for the accuracy rates did not reveal a significant interaction effect between stimulation × gating × order of stimulation (F(1,35) = 1.230, p = 0.275; ηp2 = 0.034; BF01 = 92.698), whereas the ANOVA for the RTs did indeed reveal a significant interaction effect between stimulation × gating × order of stimulation (F(1,35) = 4.564, p = 0.040; ηp2 = 0.115; BF01 = 46.851), indicating that order of stimulation had an effect only on RTs. A possible learning effect is suggested, however, since in the main analyses the atVNS effect is confined to the accuracy, and we do not see the “order effect” there; therefore, it is not possible that the order effect has had an impact. It may have had an influence in the effect if we were to show that atVNS had affected the RTs, which was not the case. Thus, there is an order effect, but this order effect is not affecting the behavioral parameter that shows the atVNS modulation (accuracy) and therefore is not critical to our study. Our previous study on fatigue effects (Yu et al., 2022b) used 3600 trials divided across four consecutive session of 900 trials each. The experiment presented in this study on atVNS effects thus equals one session of the study on fatigue effects. Importantly, robust fatigue effects in the previous study (Yu et al., 2022b), especially in the EEG, were only evident when comparing the first 900 trials (Session 1) with the last 900 trials (Session 4). Moreover, and most important, no effect of fatigue was observed for the gate closing condition, which reveals the atVNS effects in the current data. This is makes it very unlikely that fatigue may have contaminated/affected the obtained pattern of results. The neurophysiological data are shown in Figure 3.
Task-related α and theta activities at the sensor level. A, The electrodes and the time points of the difference of gating effects between atVNS active stimulation and sham for S-, C-, and R-RIDE clusters for alpha band. B, The electrodes and the time points of the difference of gating effects between atVNS active stimulation and sham for S-, C-, and R-RIDE clusters for theta band. Only the data points revealing significant differences (p ≤ 0.05) are shown. Color bar represents the t values. *p ≤ 0.01. Warm color of the topography plot represents higher alpha activity in atVNS condition. Colder color represents lower alpha activity in sham condition.
The analysis was confined to the significant stimulation effect observed for gate closing in the behavioral data. The cluster-based permutation tests did not show any significant differences in theta band activity in gate closing between active atVNS and sham stimulation, regardless of the information coding levels, that is, in all RIDE S-, C-, and R- clusters. Opposed to this, for the alpha power on all RIDE clusters (S-, C-, and R-) showed significant differences (p ≤ 0.05) between active atVNS stimulation and sham stimulation for gate closing. It is shown that alpha band activity was larger during active atVNS than sham atVNS. For all RIDE clusters, the time windows revealing significant differences of task related gate closing α powers were centered at ∼1000-2000 ms after target stimulus presentation. More specifically, for the R-cluster, it was from 1210 to 1910 ms with the mean time 1550 ms; for the S-cluster, it was from 1290 to 1920 ms with the mean time 1590 ms; and for the C-cluster, it was from 1300 to 1880 ms with the mean time 1600 ms. Thereby, the time window of 1300 and 1800 ms was chosen as basis for following analyses. For the time window from 1300 until 1800 ms after stimulus onset, a significant difference between active atVNS and sham stimulation was indicated mainly at frontal and posterior electrode sites for the three RIDE S-, C-, and R- clusters. Apparently, atVNS affected WM closing processes related to stimulus, response, and the transition between stimulus and response, in a similar way. However, this did not infer a failure of RIDE. RIDE does not lead to completely independent S-, C-, and R-clusters/components (Ouyang et al., 2013), which is also one reason why RIDE can be combined with independent component analysis (Gholamipourbarogh et al., 2023). Therefore, it may occur that the patterns of RIDE clusters are similar (but not identical). This has been also observed in previous studies (Prochnow et al., 2022; Yu et al., 2022b) and applies to the following beamforming results. The dynamical imaging of coherent sources beamformer source reconstruction revealed the anatomic regions of the highest positive atVNS and sham difference of task related gate closing alpha power, which are shown in Figure 4. For the S-cluster, the C-cluster, and the R-cluster, the modulated functional neuroanatomical regions overlapped and activity modulations were observed at the fronto-polar region (BA10, BA11) as well as in the orbital parts of the middle and inferior frontal gyrus (BA10, BA11, BA47). Moreover, activity modulations were observed in the dorsal parts of the middle and superior frontal gyrus (BA8, BA9). In addition to the frontal neural activity modulations, modulations in left inferior parietal cortices (BA40) and the temporo-parietal junction encompassing the superior temporal gyrus (BA42) were evident. For all contrasts (i.e., for all RIDE-clusters), the contrast was positive, indicating that alpha band activity was higher in these regions during active atVNS stimulation than sham stimulation.
Anatomical sources of gate closing effects. A–C, Plots represent the voxels of the top 1% (identified by DBSCAN) of atVNS – sham difference (atVNS > sham atVNS) of task-related gate closing α activities using RIDE clusters S-, C-, and R- respectively. Color bar represents the source- power estimate ratios between the contrasted conditions (atVNS-sham).
Figure 5A–C shows atVNS effects on EEG data without RIDE at the sensor and anatomic levels. The results from EEG data resembled RIDE-decomposed data, showing that the RIDE decomposition did not bias the results (see below for details). The finding of consistent effects, independent of the analysis procedure taken, also underlines the robustness of the obtained findings and the interpretation (see Discussion) that atVNS affects all coding levels important for WM gate closing processes, which have also been described previously (Rempel et al., 2021).
The atVNS effect on WM gating processes from the neurophysiological data. A, The electrodes and the time points of the difference of gating effects between atVNS active stimulation and sham conditions for alpha band of EEG signals. B, The electrodes and the time points of the difference of gating effects between atVNS active stimulation and sham conditions for theta band of EEG signals. A, B, Only the data points revealing significant differences (p ≤ 0.05) are shown. Color bar represents the t values. *p ≤ 0.01. The warm color of the topography plot represents higher alpha activity in atVNS condition. Colder color represents lower alpha activity in sham condition. C, The voxels of the top 1% of atVNS – sham difference (atVNS > sham atVNS) of task-related gate closing α activities using EEG data. Color bar represents the source-power estimate ratios between the contrasted conditions (atVNS-sham). D, The correlations between source-level gate closing task-related α activities from EEG signals and pupil activities. The index of the anatomic sources is indicated in Results. Red boundaries represent only significant correlation clusters formed by >1000 data points.
EEG data analysis without RIDE decomposition
To validate the RIDE results, we analyzed EEG data without RIDE decomposition by following the same pipeline and parameter setting. The results are shown in Figure 5. Specifically, we first conducted the time-frequency decomposition on the preprocessed EEG data to extract the theta and alpha band activities. Cluster-based permutation tests were further applied on α and theta band activities to identify the electrodes and time points showing significant atVNS effect (stimulation – sham) (Fig. 5A,B). Again, the significant atVNS effect was only revealed from the alpha band activities in the gate closing condition at ∼1.3-1.8 s after stimulus onset for most electrodes located in the frontal, central, and posterior area, resembling the RIDE-decomposed data (Fig. 5A). Subsequently, dynamical imaging of coherent sources beamforming and DBSCAN were applied on alpha band activities of the gate closing condition to identify the anatomic regions showing the largest atVNS effect, which were left-lateralized to the inferior, middle, and superior frontal cortex, as well as the superior and superior temporal cortex and the angular and precentral gyrus (Fig. 5C). LCMV beamforming was later applied on these regions (categorized to two parts: the frontal part and the temporal part) to reconstruct the time series of alpha band activities, which were further correlated with the pupil sizes (Fig. 5D). No significant correlation clusters were observed in the time of interest (i.e., 1.3-1.8 s for reconstructed alpha band activities and 1-2 s for pupil diameter data). All the above results from EEG data resembled the RIDE-decomposed data, corroborating the validity of the data and that RIDE does not distort time-frequency decomposition.
sAA data and task-related pupil diameter data
For sAA levels, there was neither a significant main effect of stimulation nor time and no interaction of these factors (all F ≤ 0.1982; p ≥ 0.167; ηp2 ≤ 0.047) (Fig. 6). We ran additional Bayesian analyses to substantiate whether the null hypothesis was more likely than the alternative hypothesis. For the interaction effect of the sAA factor with time and stimulation, the results indicated positive to very strong evidence for the null hypothesis (BF01 = 15.146). However, since sAA levels usually reveal a skewed distribution (Giraudier et al., 2022), we also did the same analysis for log-transformed sAA data. For sAA log-transformed levels, there was neither a significant main effect of stimulation nor time and no interaction of these factors (all F ≤ 1.417; p ≥ 0.241; ηp2 ≤ 0.034). Furthermore, we ran Bayesian analyses to substantiate whether the null hypothesis was more likely than the alternative hypothesis. For the interaction effect of the log-transformed sAA factor with time and stimulation, the results revealed strong evidence for the null hypothesis (BF01 = 10.844).
sAA levels at each assessment time point of active atVNS and sham stimulation. A, Bars represent the mean value of the corresponding condition. Error bars indicate SE. Points represent individual sAA levels. B, Log-transformed sAA data.
Regarding the correlation between sAA and behavior, Table 2 shows that sAA was not related to the behavioral data for all conditions, whereas Table 3 reveals only one significant negative correlation between the RT and the sAA changes (post – pre atVNS stimulation) for the WM gate closing condition. However, since our main finding on atVNS effects on behavioral performance was related to the accuracy instead of the RT, this significant correlation provides little meaning. Based on the above results, no further investigation on correlations between sAA and the frequency measures was conducted. There are two reasons: (1) lack of correlation at the behavioral level is hardly explained by an association between sAA and frequency measures (if it exists); and (2) the association between sAA and NA observed in previous studies referred to the tonic mode of NA (Fischer et al., 2018; Ventura-Bort et al., 2018; Warren et al., 2019) instead of the phasic mode. Thus, sAA did not necessarily reflect event-related processes or the phasic mode of NA.
Correlations between sAA and behavioral performance
Correlations between log-transformed sAA and behavioral performance
The pupil diameter data are shown in Figure 7. More specifically, what can be seen in Figure 7 is the baseline-corrected pupil diameter of gate opening and closing. No significant difference between atVNS stimulation condition and sham condition was observed. The only significant difference is revealed in gate opening, nevertheless, only before the stimulus onset. The pupil dilation was increased after ∼1000 ms for both gate closing and gate opening, reaching the peak at ∼1500 ms for gate closing and slightly later for gate opening. The results for the correlation analysis of the pupil diameter data and the LCMV beamforming data (i.e., the time course of activity in the reconstructed source) are given in Figure 8.
The task-related pupil sizes of gating processes. Thick lines slightly above the x axis indicate the time points showing a significant difference in pupil sizes between the atVNS and the sham sessions (p < 0.05).
Correlations between source-level gate closing task-related α activities and pupil activities for S-, C-, and R- clusters (A–L). The respective title denotes the RIDE cluster and the anatomic cluster of the source α activities. The index of the anatomic sources is indicated in Results. Red boundaries represent only significant correlation clusters formed by >1000 data points.
In general, Figure 8 shows that no evident cluster of significant correlations (p < 0.01) between the source alpha activity and the pupil dilation was observed in any RIDE cluster, anatomic regions, and sessions, except for a few small clusters, such as the frontal α activities of the RIDE R-cluster under the atVNS (Fig. 7). However, these clusters of significant correlation were found out of the time of interest of the source alpha activity (from 1300 to 1800 ms after stimulus presentation) or the pupil dilation (∼1500 ms), hence irrelevant to the focus of the current study. The present analysis suggests that, despite the small correlations displayed in Figure 8, there may exist a potential relationship between NA-mediated pupil dilation and α effects, thereby implying an influence of NA on the α modulations. However, it is pertinent to note that our primary focus lies in the designated time windows, and any sources beyond these windows cannot be traced. Specifically, despite the possible existence of a correlation between pupil diameter and alpha band activity, we lack certainty regarding the origin of this activity in other time frames, as we possess solely the information concerning alpha activity during the chosen time points. When using FDR correction, the obtained spurious correlations were not evident anymore. The results therefore suggest that the increase of α activities triggered by the atVNS was not related to the NA dynamics indicated by pupil dilation. Likewise, no significant correlation within time of interest between alpha band activities of EEG data and pupil diameters was detected (Fig. 4D), suggesting that the absence of interaction between pupil dilation and neurophysiological activities was robust.
Discussion
In the current study, we examined the effects atVNS on WM gating processes. For that, we used a validated experimental approach able to distinguish between WM gate opening and closing processes. The behavioral data on the general task effects are comparable with previous findings (Rac-Lubashevsky and Kessler, 2016; Rempel et al., 2021; Yu et al., 2022b). There were no behavioral effects of atVNS stimulation (i.e., differences between sham and active stimulation) on WM gate opening, but on WM gate closing processes. The direction of effects showed that active atVNS increases behavioral accuracy, compared with sham atVNS stimulation. This increase in behavioral accuracy indicates that atVNS enhanced WM gate closing processes and thus the stability of information during WM maintenance. On a neurophysiological level, the dissociation of atVNS effects on WM gate opening and closing processes was also evident. There were no significant differences in theta band activity, but in alpha band activity. Alpha band activity during WM gate closing was stronger during active atVNS, and this was case for all aspects of information coded in the EEG signal (i.e., for the S-cluster, the C-cluster, and the R-cluster). Thus, atVNS modulates all relevant codes (i.e., stimulus processing, stimulus–response translation, and motor response codes) during WM gate closing processes. Also, previous findings suggest that all aspects of information coded in the neurophysiological signal are in parallel affected by WM gating processes (Rempel et al., 2021; Yu et al., 2022b). The beamforming analysis revealed that, for all aspects of information coded in alpha band activity (reflected in the S-cluster, C-cluster, and R-cluster), the atVNS effects were associated with the fronto-polar region (BA10, BA11) as well as in the orbital parts of the middle and inferior frontal gyrus (BA10, BA11, BA47). Information theoretical approaches to PFC functioning conceptualize frontopolar regions important for so-called “branching” of control processes (Koechlin and Summerfield, 2007; Gohil et al., 2016; Mansouri et al., 2017), reflecting the ability to maintain information and use this to inform processes in other PFC regions to enable action selection (Koechlin and Summerfield, 2007). Of note, a recent study using a different EEG analysis strategy in the same experimental paradigm provides converging evidence for a role frontopolar and orbitofrontal regions for WM gate closing processes (Yu et al., 2022c) and thus the conceptual validity of the finding and the above interpretation. In light of the behavioral data, the atVNS-induced increase in alpha band activity in these regions during WM gate closing likely reflects the increased stabilization of WM content (i.e., maintenance). However, in addition to frontopolar and orbitofrontal regions, also left inferior parietal and superior temporal cortex regions (BA40; BA42) reflected modulations of alpha band activity as induced by atVNS effects. Especially left inferior parietal cortices are likely essential for the updating of internal representations of the environmental context using incoming information (Geng and Vossel, 2013). Exactly such processes are prevented through WM gate closing. Since parietal regions are involved in processing stimulus information, stimulus–response translation processes, and motor response programming (Andersen and Buneo, 2002; Gottlieb, 2007; Andersen and Cui, 2009), it is reasonable that also here, despite having a different time window, all aspects of information coded in alpha band activity are affected. Crucially, increased alpha band activity very likely reflects increased inhibitory gating processes assumed to be essential to decrease the effects of possibly distracting information (Klimesch, 2012). This role is central for WM maintenance (Bonnefond and Jensen, 2012; Lozano-Soldevilla et al., 2014; Wianda and Ross, 2019), which depends on WM gate closing (O'Reilly and Frank, 2006; Rac-Lubashevsky and Kessler, 2015; Yu et al., 2022b). When the gate is closed, distracting information can no longer access WM and its content is maintained. Increased atVNS-induced alpha band activity in the observed functional neuroanatomical regions likely reflects specifically this role. Interestingly, other findings corroborate that atVNS reduces the impact of distracting information (Fischer et al., 2018) through alpha band activity (Konjusha et al., 2022), which suggest that alpha band activity is an essential mechanistic element in atVNS effects on cognitive functions.
Importantly, the data from the current study, put in context with results from previous studies, also provide insights into the neurobiological basis of the observed effects. As mentioned, atVNS is likely to modulate the catecholaminergic and the GABAergic system (Colzato and Beste, 2020; Farmer et al., 2021; Ludwig et al., 2021; but see Keute et al., 2021), raising the question which of these system may underlie the observed behavioral and neurophysiological effects. Research on the role of alpha band activity in WM processes has already outlined the importance of the GABAergic system (Lozano-Soldevilla et al., 2014). Moreover, the conception that alpha band activity reflects inhibitory gating processes draws on the role of inhibitory (GABAergic) neural transmission (Klimesch et al., 2007; Klimesch, 2012). This, together with the previous findings reporting a direct activation of the nucleus tractus solitarius by atVNS (Borgmann et al., 2021), known to be rich in GABAergic neurons (Dietrich et al., 1982), as well as electrophysiological evidence on the involvement of the GABAergic system in atVNS (Keute et al., 2018, 2019a), makes it likely that especially the GABAergic system is likely to underlie the observed effects. Further evidence for this interpretation comes from the obtained data on sAA levels and pupil diameter recordings in this study. The sAA levels, known to reflect modulations of the catecholaminergic system (including NA) (Chatterton et al., 1996), were not modulated by atVNS; and the Bayesian analysis revealed substantial evidence for a lack of modulation. For the pupil diameter data, main effects of task conditions were obtained that were also observed previously (Yu et al., 2022b); however, no reliable modulations of the correlation of EEG data and pupil diameter data by atVNS were obtained in the current study. Also, the pupil diameter was not differentially modulated between sham and atVNS conditions. It may be argued that atVNS effects on these estimates of catecholaminergic/noradrenergic system activity are not very reliable because conflicting findings have been obtained for sAA levels (Fischer et al., 2018; Ventura-Bort et al., 2018; Warren et al., 2019; D'Agostini et al., 2022) as well as pupil dilatation measures (Keute et al., 2019b). Yet, using the same task as used in the current study, there is evidence that the noradrenergic system (as assessed using pupil diameter data) is of relevance (Yu et al., 2022b). Of note, this was only the case when effort was explicitly required and executed (i.e., in time-on-tasks when participants were informed of the long task duration in advance); and it came to WM gate opening processes, but not WM gate closing processes (Yu et al., 2022b). Moreover, in the EEG, concomitant modulations of WM gate opening processes were reflected in the theta and not the alpha band (Yu et al., 2022b). This also applied to correlations between pupil diameter and EEG theta band dynamics for the early stage of the time-on task when effort remained in a relatively high level (Yu et al., 2022b). Nevertheless, the missing correlations between EEG theta band dynamics and pupil diameter in the current study are incomparable with the previous study since the two studies largely differ in goals, procedures, instructions, etc., despite using the same paradigm. Thus, and considering the role of alpha band activity in WM maintenance as well as the role of GABAergic signaling for WM modulations of alpha band activity (see above), it is likely that especially the GABAergic system's effect of atVNS is relevant for the obtained findings.
One central implication of the results is that atVNS is possibly more potent to modulate the GABAergic system than the catecholaminergic system. The overarching pattern of previous research findings suggest that on the one hand catecholaminergic activity and theta band effects may be related to atVNS; on the other hand, GABAergic activity and alpha band effects. Future studies should elaborate whether GABAergic effects of atVNS are mechanistically relevant for cognitive functions strongly drawing on alpha band activity, while catecholaminergic effects of atVNS are mechanistically relevant for cognitive functions strongly drawing on theta band activity. So far, a consideration of the neurobiological effects and the evaluation of their correlates in studies on atVNS is missing and limited by the lack of a reliable “online measurement” of GABAergic activity during atVNS stimulation, a limitation that also applies to the current study. The current study, as well as a previous study by our group (Konjusha et al., 2022), provides evidence that atVNS has the ability to modulate cognitive processes and improve WM information maintenance. The modulatory effects of atVNS appear to be associated with the modulation of alpha band activity (diminished in the previous study) in different brain regions, rather than theta band activity. However, the modulation of atVNS is contingent on stimulation parameters and the specific cognitive task. Notably, the previous study used a 30 s on and 30 s off stimulation, while the current study used continuous stimulation. This distinction may explain the somewhat different results; however, it should be noted that, although WM and conflict monitoring are both executive functions, they have distinct functions and neural correlates. It is possible that the 30 s on and 30 s off stimulation induces phasic rather than tonic NA. It is difficult to predict how continuous stimulation with atVNS might alter tonic activity and, in turn, phasic firing. It is argued that it is important to consider both activity modes of the LC-NA. To address the issue of the best stimulation type to activate the LC-NA system, Villani et al. (2022) suggested an event-related atVNS approach, where short bouts of atVNS are synchronized with the presentation of stimuli. This approach has been shown to modulate both the tonic and phasic modes of the LC-NA system as well as cognitive processes. Future studies might consider applying this method in aims of understanding better the effects of different stimulation parameters of atVNS on the LC-NA system. To date, the existing literature has yielded contradictory findings regarding the mechanisms underlying atVNS. However, there is also a preponderance of evidence suggesting that atVNS may be modulated via noradrenergic pathways. Notably, the present study did not reveal any significant effects on proxies of NA, namely, sAA and pupillary dilation. This outcome may be attributed to the limited sample size or the utilization of continuous stimulation. Nonetheless, the current findings shed light on the potential impact of continuous atVNS, and further investigation is necessary to establish definitive conclusions.
In conclusion, the current study provides detailed insights into the effects of atVNS on WM gating dynamics. atVNS specifically modulates WM gate closing and thus modulates neural mechanisms enhancing the maintenance of information in WM. atVNS does so through the modulation of alpha band activity in fronto-polar, orbital, and inferior parietal regions. The data suggest that these effects may not occur because of modulations of the catecholaminergic (NA) system as indicated by lack of modulatory effects in pupil diameter dynamics, in the inter-relation of EEG and pupil diameter dynamics and saliva markers of NA activity. Considering these findings, it appears that a central effect of atVNS during cognitive processing refers to the stabilization of information in neural circuits, putatively mediated via the GABAergic system.
Footnotes
This work was supported by Deutsche Forschungsgemeinschaft Grants BE4045/36-1 and BE4045/43-1. We thank Clemens Kirschbaum for support in analyzing the sAA probes.
The authors declare no competing financial interests.
- Correspondence should be addressed to Christian Beste at Christian.beste{at}ukdd.de
