Transcutaneous auricular vagus nerve stimulation modulates the processing of interoceptive prediction error signals and their role in allostatic regulation

Abstract

It has recently been suggested that predictive processing principles may apply to interoception, defined as the processing of hormonal, autonomic, visceral, and immunological signals. In the current study, we aimed at providing empirical evidence for the role of cardiac interoceptive prediction errors signals on allostatic adjustments, using transcutaneous auricular vagus nerve stimulation (taVNS) as a tool to modulate the processing of interoceptive afferents. In a within‐subject design, participants performed a cardiac‐related interoceptive task (heartbeat counting task) under taVNS and sham stimulation, spaced 1‐week apart. We observed that taVNS, in contrast to sham stimulation, facilitated the maintenance of interoceptive accuracy levels over time (from the initial, stimulation‐free, baseline block to subsequent stimulation blocks), suggesting that vagus nerve stimulation may have helped to maintain engagement to cardiac afferent signals. During the interoceptive task, taVNS compared to sham, produced higher heart‐evoked potentials (HEP) amplitudes, a potential readout measure of cardiac‐related prediction error processing. Further analyses revealed that the positive relation between interoceptive accuracy and allostatic adjustments—as measured by heart rate variability (HRV)—was mediated by HEP amplitudes. Providing initial support for predictive processing accounts of interoception, our results suggest that the stimulation of the vagus nerve may increase the precision with which interoceptive signals are processed, favoring their influence on allostatic adjustments.

We aimed at investigating the role of interoceptive prediction errors on allostatic adjustments, using transcutaneous auricular vagus nerve stimulation (taVNS) to increase precision. We found that taVNS facilitated the maintenance of interoceptive accuracy levels, increased heart‐evoked potentials amplitudes (correlates of cardiac‐related prediction error signals) and mediated their influence on allostatic adjustments (measured by heart rate variability).

1. INTRODUCTION

New perspectives of brain functioning suggest that the brain is not a passive entity that reacts to incoming information but actively and continuously predicts what the upcoming environmental and body states might be. These predictions can be understood in terms of Bayesian probabilities (Clark, 2013; Friston, 2010). Expectations (i.e., priors) created based on previous experiences serve as initial beliefs about probable causes of the incoming input (i.e., observations). Observations are then contrasted with the priors to create the posterior probabilities that serve as estimators of the causes of the inputs (e.g., perceptions). According to this predictive processing framework, neural representations of the predictions, conveying as top‐down signals that propagate hierarchically throughout the cortex (Bastos et al., 2012; Friston, 2010; Friston & Kiebel, 2009; Katsumi et al., 2023), are tested against incoming (e.g., sensory) information represented as bottom‐up neural activity. The outcome of this contrast depends on precision signals—the attention devoted to increasing the signal‐to‐noise ratio of observations in favor of specific predictions (Feldman & Friston, 2010)—that modulate the strength and durability of predictions and prediction error signals (Feldman & Friston, 2010; Kanai et al., 2015). If high levels of uncertainty are detected (Friston, 2010), resulting from a notable difference between bottom‐up and top‐down signals (i.e., prediction errors), the organism may act upon them by updating predictions based on prediction errors, reducing attention (i.e., precision) to prediction error signals and/or taking actions that promote the predicted sensations (Friston & Kiebel, 2009).

Predictive processing may serve as a general framework to explain the computational functioning of different cortical and subcortical brain regions (Barron et al., 2020; Chanes & Barrett, 2016; Katsumi et al., 2022; Kleckner et al., 2017; Rao & Ballard, 1999; Sales et al., 2019) and associated psychological processes such as visual (de Lange et al., 2018), auditory (Todorovic & de Lange, 2012), or pain perception (Büchel et al., 2014; Geuter et al., 2017; Strube et al., 2021). Beyond the sensory domain, interoception, which refers to the processing of hormonal, autonomic, visceral and immunological signals (Khalsa et al., 2018), has also been proposed to be governed by mechanisms oriented to reduce prediction error signals (Barrett & Simmons, 2015; Seth, 2013). To favor an optimal distribution of energetic resources, the organism needs to generate predictions about the physiological state of the body. These predictions are compared against interoceptive afferent signals (e.g., autonomic, visceral, muscular changes) and the resulting prediction errors are considered to ensure an adequate allostatic adjustments to maintain homeostasis. Predictions may arise from agranular visceromotor regions such as the anterior insula and anterior cingulate cortex and propagate to more granular regions such as the posterior insula. Bodily signals are transmitted through the lamina I pathway and the afferent branch of the vagus nerve, which projects to the nucleus of the solitary tract (NST). In turn, the NST via the ventromedial thalamic nucleus innervates the mid and posterior insula (Craig, 2002; Rajendran et al., 2023), regions known to be involved in the computation of bottom‐up, prediction errors signals. From the mid and posterior insula, prediction error signals are sent back to agranular regions (e.g., anterior insula) to update predictions (Barrett & Simmons, 2015; Seth, 2013; Seth & Friston, 2016).

Recent computational modeling research has provided initial evidence for the role of interoceptive predictions and prediction errors in allostatic regulation, understood as the process by means of which the brain anticipates the physiological, immunological, and metabolic needs of the body and aims at meeting those needs before they arise (Sennesh et al., 2022; Sterling & Eyer, 1988). By simulating psychophysiological signals, Allen et al. (2022) observed that a profile with precise viscerosensory sensations was associated with an increase of heart rate variability (HRV), a valuable indicator of adaptation and self‐regulation capacity (Laborde et al., 2017; Laborde et al., 2023), and a readout measure of allostatic regulation of demands by interoceptive and exteroceptive perturbations (Grossman & Taylor, 2007; Thayer et al., 2012; Thayer & Sternberg, 2006). In the current study, we aimed at extending these findings to empirical human data, by investigating how the processing of cardiac afferent signals and the modulation thereof—by means of transcutaneous auricular vagus nerve stimulation (taVNS) (Farmer et al., 2021; Paciorek & Skora, 2020)—affects allostatic regulation, as indexed by HRV.

It has been suggested that objective interoceptive tasks (e.g., heartbeat counting task [HCT] (Schandry, 1981); heartbeat detection task, (Kleckner et al., 2015)) that assess how well participants can detect their own bodily signals (objective interoceptive accuracy [IAcc]) capture the extent of precision with which bodily changes are processed (Ainley et al., 2016; Owens et al., 2018). In this regard, accuracy in cardiac‐related interoceptive tasks has been associated with stronger activation in regions involved in the computing of prediction errors and prediction update, such as the anterior insula and anterior cingulate cortex (Craig, 2009; Critchley et al., 2004; Kuehn et al., 2016; Zaki et al., 2012). Furthermore, higher IAcc has been associated with reduced functional connectivity between the posterior and anterior insula (Kuehn et al., 2016), suggesting that in participants who could accurately detect their cardiac activity, the precision of such signals increased (higher signal‐to‐noise ratio) at the expense of lower precision for other bodily changes not relevant to the task at hand (Kuehn et al., 2016).

If the ability to accurately perceive bodily changes may reflect the extent to which bodily prediction errors are attended, the associated brain activity may inform about the weight with which these interoceptive prediction error signals are processed (Ainley et al., 2016; Petzschner et al., 2019). Generally, event‐locked electrophysiological activity has been proposed as a potential indicator of prediction error signal processing (Arnal & Giraud, 2012; Bastos et al., 2012; Feldman & Friston, 2010; Friston & Kiebel, 2009), and within interoception, the heartbeat‐evoked potential (HEP, i.e., averaged electrophysiological signal time‐locked to the cardiac cycle), as a suitable proxy of prediction error processing of cardiac activity (Ainley et al., 2016). This view was supported by a recent study in which the role of attention (and thus precision) to interoceptive signals on the HEP amplitudes was investigated (Petzschner et al., 2019). The authors found that HEP amplitudes, as correlates of interoceptive prediction errors, were increased when participants focused on interoceptive versus exteroceptive signals, suggesting that interoceptive information—when prioritized (i.e., increased precision to cardiac interoceptive signals)—elicits stronger prediction error signals (Petzschner et al., 2021). These results were partly replicated by Zaccaro and colleagues (Zaccaro et al., 2022, 2023) who observed increased HEP amplitudes during an interoceptive cardiac task compared to respiratory or exteroceptive tasks, particularly during exhaling. Importantly, if HEP amplitudes may reflect how well interoceptive prediction error signals are processed, their enhancement during interoceptive cardiac tasks should be particularly pronounced in participants with higher interoceptive abilities (Ainley et al., 2016). In line with this hypothesis, it has been shown that HEP amplitudes during cardiac‐related interoceptive tasks are specially increased in participants showing high IAcc scores (Pollatos & Schandry, 2004; Yuan et al., 2007; Zaccaro et al., 2022), indicating that individuals with higher abilities to perceive their cardiac activity, show increased processing of cardiac prediction error signals when such information becomes relevant.

The vagus nerve as one of the main afferent pathways conveying visceral information may play a pivotal role in the processing of interoceptive prediction error signals (Barrett & Simmons, 2015; Khalsa et al., 2018; Paciorek & Skora, 2020; Rajendran et al., 2023). Accordingly, invasively or transcutaneously—vagus nerve stimulation, by exerting its influence in the NST (Frangos et al., 2015; Müller et al., 2022; Teckentrup et al., 2021), may facilitate the processing of interoceptive prediction error signals (Paciorek & Skora, 2020). Supporting this idea, taVNS compared to sham stimulation has been shown to increase (Villani et al., 2019) or maintain IAcc levels across trials (pre vs. post stimulation) (Richter et al., 2021). Similarly, HEP amplitudes both at rest and during the performance of an interoceptive (and also an exteroceptive) task are modulated by taVNS (Poppa et al., 2022; Richter et al., 2021), and such modulation is associated (among other regions) with activity in the insula (Poppa et al., 2022).

The goal of the current study was twofold: replicating the enhancing effects of taVNS on cardiac IAcc and HEP amplitudes and investigating the relationship between the processing of interoceptive prediction error signals and allostatic regulation, as measured by HRV. To this end, participants in the current within‐subject crossover study underwent a cardiac interoceptive task (and a control task; HCT and the second counting task [SCT]) before and during taVNS or sham stimulation. We expected that taVNS would enhance the detection of cardiac signals as indicated by higher IAcc scores and HEP amplitudes, replicating previous findings (Poppa et al., 2022; Richter et al., 2021; Villani et al., 2019). Expanding prior work, we further expected that such enhancement in IAcc and processing of interoceptive prediction errors (i.e., HEP amplitudes) would favor more allostatic adjustments, as indexed by HRV.

2. METHODS

2.1. Participants

The sample size was determined using G*Power (Faul et al., 2007) based on the effect size of previous taVNS studies (Ventura‐Bort, Wirkner, et al., 2021) (η ² = 0.17). The estimated sample size to detect a significant effect (α = .05) with 80% power was 42. We collected data of a few more participants in case recording problems emerged or the quality of the data was compromised. A total of N = 53 (41 females, 6 males, 6 diverse; M _age = 24.46, SD_age = 5.18) students from the University of Potsdam underwent the study in exchange for course credits. All participants had normal or corrected‐to‐normal vision. All participants provided written informed consent for a protocol approved by the ethics committee of the University of Potsdam. Prior to the first session, participants were screened and invited to participate if they did not meet any of the following exclusion criteria: neurological or mental disorders, brain surgery, chronic or acute medication use, history of migraine and/or epilepsy (see S1 in the Supplementary Material for summary of data quality for each participant).

2.2. Material design and procedure

Participants performed the HCT (Schandry, 1981) and a control time estimation task (SCT (e.g., Murphy et al., 2018)). In the HCT, participants were instructed to silently count their heartbeats over varying periods of time without trying to guess them and without actively touching any part of the body (e.g., neck or wrist) where the heartbeats could be felt (Murphy et al., 2018). Similarly, in the SCT, participants had to count the seconds passed for varying periods. An acoustic signal indicated the beginning and end of each trial. Participants were instructed to close their eyes while performing the tasks. After each trial, participants indicated the number of heartbeats they had felt or how many seconds they had counted, and how confident they were in their answers. Both tasks were performed after each other in three different blocks. Within each block, the order of the tasks (HCT or SCT) was counterbalanced across sessions. Each task consisted of four trials of different lengths (short trials: 25, 35, 45, 100 s or long trials: 28, 38, 48, 103 s). Trials were presented randomly within blocks and trial length was counterbalanced across blocks, resulting in a total of 24 trials (12 in each task), the first 8 trials took place without stimulation and the remaining 16 trials were performed while stimulation was switched on (see Figure 1a).

FIGURE 1.

Design figure and behavioral results from the heartbeat counting task (HCT). (a) The design used in both transcutaneous auricular vagus nerve stimulation (taVNS) and sham sessions. Before receiving stimulation, participants performed a four‐trial block of the HCT and second counting task (SCT) (order counterbalanced across sessions). Thereafter, tonic blood pressure (BP) and heart rate (HR) were measured and the stimulation electrode was applied. The stimulation was then switched on and participants underwent a 6‐min resting phase before performing two more four‐trial blocks of the HCT and SCT. At the end of the session, BP and HR as well as side effects ratings were obtained. (b) The interoceptive accuracy (Acc) scores during the HCT. Bar plots and dark dots represent the average scores across participants. Black lines represent standard errors. Results are split by block: block without stimulation (NO) and averaged blocks in which the stimulation was switched on (ON). Results are further split by session (Session 1, S1, and Session 2, S2). taVNS condition is colored pink and sham condition is colored gray. Pink and gray dots represent individual scores. (c) The interoceptive Acc increase (difference between NO Stimulation and Stimulation ON) as a function of stimulation and block. Bar plots and black dots represent the average scores across participants. Black lines represent the standard errors. Participants are split between those receiving taVNS in session 1 (purple), and those who received taVNS in session 2 (yellow). Purple and yellow dots represent individual scores.

We used a randomized, single‐blinded, within‐subject, crossover design (taVNS‐sham; sham‐taVNS) carried out in two sessions that took place 1 week apart. Both sessions were identical except for the stimulation condition. After arrival, participants entered the experimental, sound‐attenuated, dimly lit room, were seated in a comfortable chair and electrocardiography (ECG) and electroencephalography (EEG) electrodes were applied. Before stimulation, participants performed the first block of the experimental tasks (i.e., HCT and SCT). Afterward, blood pressure, heart rate, and salivary alpha‐amylase (sAA) levels (indirect marker for central noradrenergic activity (Chatterton et al., 1996; Giraudier et al., 2022; Warren et al., 2017)) were collected. Thereafter, stimulation electrodes were applied to the left ear and the intensity was adjusted as described in the following taVNS section (Ventura‐Bort et al., 2018). After applying the stimulation, participants were instructed to close their eyes for the upcoming resting phase. At the end of the 6‐min resting phase, they were instructed to open their eyes and to perform two more blocks of the experimental tasks (see Figure 1a).

After completion of the third block, the ECG and EEG electrodes were detached, the stimulation electrodes removed, and blood pressure, heart rate, and sAA levels were measured again. Finally, participants were asked to report on a seven‐point scale (ranging from 1, not at all, to 7, very much) how strongly they experienced the following symptoms during the stimulation: headache, nausea, dizziness, neck pain, muscle contractions in the neck, stinging sensations under the electrodes, skin irritation in the ear, fluctuation in concentration or feelings, and unpleasant feelings. Furthermore, at the end of the second session, participants were asked about their beliefs and knowledge about resting heart rate (i.e., “How many times do you think the average person's heart beats in 60 s when they are at rest?”; “How many times do you think your own heart beats in 60 s when you are at rest?”).

2.3. Transcutaneous auricular vagus nerve stimulation

The taVNS stimulator consisted of two titan electrodes attached to a mount wired to a stimulation unit (CMO2, Cerbomed, Erlangen, Germany). In the taVNS condition, the stimulator was placed in the left cymba conchae, an area innervated exclusively by the auricular branch of the vagus nerve (Ellrich, 2011; Peuker & Filler, 2002). As in previous studies using taVNS (Kraus et al., 2007), electrodes for the sham condition were positioned in the centre of the left ear lobe, an area known to be free of vagal innervation (Ellrich, 2011; Peuker & Filler, 2002). Prior to application of the electrodes, the ear surface as well as the electrodes were cleaned with disinfectant solution. Thereafter, the stimulator was mounted on the ear and the device was shortly switched on at the minimum intensity (0.1 mA) to ensure that the stimulation was properly delivered. If the stimulation was not working properly, the position of the stimulator was adjusted to ensure a proper contact with the area of interest and, if necessary, attached to the surface of the ear with tape. In addition, the electrodes were also soaked in disinfectant solution to reduce resistance. Stimulation alternated between on and off phases every 30 s and was delivered with a pulse width of 250 μs at 25 Hz. The stimulus intensity was set to be perceived but not to cause discomfort. To individually adjust the stimulation intensity, participants received increasing and decreasing series of stimulation and rated the subjective sensation of the stimulation on a 10‐point scale, ranging from nothing (1), light tingling (3), strong tingling (6), to painful (10). The increasing series of trials started from an intensity of 0.1 mA and increased in 0.1 mA steps on a trial‐by‐trial basis until participants reported a sensation of 9. Before starting the decreasing series, the same intensity was repeated and then reduced trial by trial in 0.1 mA steps until a subjective sensation of six or below was experienced. This procedure was repeated a second time. The final stimulation intensity used for the experimental procedure was calculated based on the average of the four intensities rated as 8 (i.e., two from increasing and two from decreasing series). The average stimulation intensity for both conditions were as follows: 0.98 mA (0.1–3.7 mA) for taVNS and 0.96 mA (0.1–3.2 mA) for sham condition. The stimulation intensity did not differ between both conditions (b = 0.005, SE = 0.13, t (42.9) = 0.042, p = .97). The stimulation lasted for about 35 minutes. Although the felt intensity might decrease with time due to habituation effects, the stimulation intensity was kept constant throughout the task.

2.4. Autonomic measures

To test the effects of stimulation on cardiovascular measures, blood pressure (systolic and diastolic) and heart rate were measured before (baseline) and after stimulation with an upper arm cuff placed on the left arm using the Intelli Wrap Manschette M500 device (Omron Healthcare, Medizintechnik Handelsgesellschaft mbH, Mannheim, Germany). In addition, mimicking the protocol of our prior studies (Fischer et al., 2018; Giraudier et al., 2020; Ventura‐Bort et al., 2018; Ventura‐Bort Wirkner, et al., 2021) sAA was also measured as a marker of endogenous noradrenergic activation (Chatterton et al., 1996; Giraudier et al., 2022; Warren et al., 2017). Saliva was collected by instructing participants to drool the saliva into a polypropylene tube. Saliva samples were stored at −20°C. In a recent mega‐analysis carried out while data collection of the current study was undergoing, we observed that the effects of taVNS on sAA levels were rather small, hardly observable in single studies, and only discernible when data were analyzed collectively (Giraudier et al., 2022). These results suggest that in smaller samples, sAA might not be informative of the efficiency of taVNS, and thus we decided not to analyze sAA levels in the current study.

2.5. Electrocardiography

Raw ECG signal was recorded continuously during the experiment, using a two‐lead set‐up with the MP‐160 BIOPAC system (BIOPAC Systems, Goleta, CA). The 8 mm Ag‐AgCl electrodes were filled with electrolyte and placed on the right arm and left ankle, following the Einthoven's triangle configuration (lead II). The raw ECG signal was recorded at 2000 Hz. After automatic R‐peak detection (using in‐house scripts), data were segmented (i.e., during the resting phase: 6‐min segment was extracted; during experimental tasks: the length of the trials segments was determined by the starting and ending acoustic signals) and visually inspected to detect and correct for possible artifacts. Thereafter, we computed the number of heartbeats per participant during each trial to calculate the objective IAcc (see below). Furthermore, the inter‐beat intervals (distance between RR intervals in milliseconds) were extracted from all segments for the calculation of the HRV index.

Trials were rejected if not all R‐peaks could be correctly identified (e.g., due to movement artifacts) or if, due to technical issues, triggers signaling the start or end of the trial were not recorded.

2.6. Electroencephalography

EEG was recorded using five Ag/AgCl electrodes (8 mm diameter) filled with EC2 Genuine Grass Electrode Cream (West Warwick, R) and placed at Fpz, Fz, Cz, Pz, and Oz according to the international 10‐20‐system. Two more electrodes (Ag/AgCl, 8 mm diameter) placed at the left and right mastoids served as online reference. To correct the EEG data from potential eye movement artifacts, an electrooculogram (EOG) was recorded with two additional electrodes (Ag/AgCl, 8 mm diameter) placed above and on the right side of the right eye, using the mastoids as reference. The EEG and EOG activity were measured with the MP‐160 BIOPAC system (BIOPAC Systems).

2.7. EEG preprocessing and extraction of HEPs

EEG data were downsampled by a factor of 8 (250 Hz) and low‐pass filtered (at 20 Hz to avoid any potential noise from the stimulation). Then, ocular artifacts were corrected using the revised aligned artifact average procedure (Croft & Barry, 2000). Given that participants performed the task with eyes closed and avoided corporal movements, the quality of the EEG was generally good. Overall, only 1.9% of the trials under taVNS and 1.7% of trials under sham had to be excluded. The percentage of excluded trials did not differ between stimulation conditions, t (45) = 1.63, p = .11. After artifact removal, data were segmented into epochs from 300 prior to 600 ms after R‐peak onset and baseline corrected (−200 to −100 ms). Trials with bad quality data (mean amplitude, standard deviation, or maximum gradient larger than 3 standard deviations) in one or more sensors were removed. Epochs were then averaged across sensors and blocks (or trials within blocks for mediation analysis; see below).

2.8. Analysis

2.8.1. Self‐report and autonomic outcomes

To test for potential side effects induced by the stimulation, ratings on side effects were compared between taVNS and sham stimulation for each reported subjective symptom, separately, using linear mixed models with the fixed factors Session (Session 1 vs. Session 2) and Stimulation (taVNS vs. sham), as well as their interaction. Participant intercepts were modeled as random effects. To test the effects of stimulation on blood pressure, and heart rate, linear mixed models were employed with the fixed factors Time (pre‐ vs. post‐stimulation), Session (Session 1 vs. Session 2), and Stimulation (taVNS vs. Sham), as well as their interactions. Participant intercepts were modeled as random effects,

$$\mathsf{\text{Side effect}}~{\mathsf{\text{Stimulation}}}^{*}\mathsf{\text{Session}}+\left(\mathsf{1}|\mathsf{\text{Participant}}\right)$$

For If interaction effects were found (or exploratory analysis conducted), they were followed up by pairwise post hoc comparisons using lsmeans (Lenth, 2016).

2.8.2. Interoceptive accuracy

Objective IAcc was extracted from the HCT (and SCT; see Supplementary Material Section S9 for description and results of other interoceptive indexes commonly extracted from the HCT). IAcc was derived from the counted heartbeats (or seconds) reported by participants compared to the objectively measured heartbeats (or seconds passed). The accuracy score was calculated using the following formula:

$$\mathit{\text{IAcc}}=1-\left(\frac{\left|N_{\mathit{\text{measured}}}-N_{\mathit{\text{counted}}}\right|}{N_{\mathit{\text{measured}}}}\right)*100$$

N_measured refers to the number of measured heartbeats (based on ECG recordings; or seconds passed). N_counted refers to the number of heartbeats (or seconds counted). The accuracy score was calculated for each participant and trial and averaged across trials for each block and participant. To ensure normalization of the data, IAcc scores were log‐transformed.

Because we were interested in the modulation of IAcc scores during active stimulation in comparison to sham stimulation, an increase IAcc score was calculated by substracting the IAcc score during block 1 (i.e., averaged across trials) from the IAcc score from blocks 2 and 3 (averaged across trials). The effects of stimulation were examined using linear mixed models. We included Stimulation (i.e., taVNS vs. sham), Task (i.e., HCT vs. SCT), Session (Session 1 vs. Session 2), and Block (Block 2 vs. Block 3) as well as the interaction between factors. Previous studies have highlighted the need to control potential confounds when assessing IAcc in the HCT (Murphy et al., 2018; Prentice & Murphy, 2022). Following these recommendations, we also included baseline blood pressure and heart rate (measured prior to stimulation), the beliefs about one's heart rate, the body mass index (BMI), age, and gender as fixed factors in our model. Furthermore, participant intercepts were modeled as random effects,

$$\mathsf{\Delta }\mathsf{\text{Accuracy}}~{\mathsf{\text{Task}}}^{*}{\mathsf{\text{Stimulation}}}^{*}{\mathsf{\text{Block}}}^{*}\mathsf{\text{Session}}+\mathsf{BMI}+\mathsf{\text{baseline}}\mathsf{HR}+\mathsf{Age}+\mathsf{\text{Gender}}+\mathsf{\text{baseline systolic blood pressure}}+\mathsf{\text{baseline diastolic blood pressure}}+\mathsf{\text{Subjective heart rate estimation}}+\left(\mathsf{1}|\mathsf{\text{Participant}}\right)$$

If interaction effects were found (or exploratory analysis conducted), they were followed up by pairwise post hoc comparisons using lsmeans (Lenth, 2016).

2.9. Heart‐evoked potentials

Individual ERP averages were computed for each sensor, block, task, stimulation type, and participant. To examine the effects of stimulation on HEPs, we submitted all sensors and time points to a 2 × 2 × 2 repeated‐measures within‐subjects ANOVA, including the factors Stimulation (taVNS vs. sham), Task (HCT vs. SCT), and Block (Block 2 vs. Block 3). Significant effects were corrected for multiple comparisons on the spatiotemporal dimension from 0 to 600 ms using the cluster‐based permutation test implemented with ElectroMagnetic EncephaloGraphy Software (Peyk et al., 2011). This test uses a two‐step procedure to identify significant effects between conditions. In a first step (i.e. sensor‐level criterion), F‐tests are performed for each time point and sensor. If significant effects (p < .05) that last for at least five consecutive time points (i.e., 20 ms) are detected, their F‐values are summed into the so‐called empirical “cluster masses.” In a second step (cluster‐level criterion), Monte Carlo simulations are used to draw 1000 permutations across experimental conditions and participants. As for the empirical data, cluster masses are extracted from the permuted data, following the above‐mentioned criteria (i.e., p < .05 for at least 20 ms). To determine whether the empirical cluster masses reach significance, these are compared to the “critical cluster mass” defined as the 95th percentile of the permuted cluster masses. If an empirical cluster mass is bigger than the critical cluster mass, it is considered significant.

The detection of significant effects was restricted to a minimum time window of 20 ms. Given that only five sensors were used, no spatial restrictions were made. Permutation tests are limited to ANOVAs requirements (e.g., analysis is restricted to cases with data in all possible cells), given that for some participants data were only available in one session (see Supplementary Material Section S1 for details). If any main or interaction effects of stimulation emerged, they were tested over all available data using linear mixed models. For task‐related effects, we included Stimulation (i.e., taVNS vs. sham), Task (i.e., HCT vs. SCT), and Block (Block 2 vs. Block 3) as well as their interaction as fixed factors. To ensure that the significant interaction effects were due taVNS, analyses were repeated for the HEP increase, calculated by subtracting the amplitudes extracted during block 1 (i.e., averaged across trials) from those from blocks 2 and 3 (averaged across trials), using Stimulation, and Task as well as their interaction as fixed factors. Participant intercepts were modeled as random effects in both analyses,

$$\mathsf{\Delta }\mathsf{HEP}\mathsf{\text{amplitudes}}~{\mathsf{\text{Task}}}^{*}{\mathsf{\text{Stimulation}}}^{*}\mathsf{\text{Block}}+\left(\mathsf{1}|\mathsf{\text{Participant}}\right)$$

If interaction effects were found (or exploratory analysis conducted), they were followed up by pairwise post hoc comparisons using lsmeans (Lenth, 2016).

During the resting‐state phase, the modulatory effects of stimulation on HEP amplitudes were examined by comparing the effects of stimulation (taVNS vs. sham) using a cluster‐based permutation test for period of 0–600 ms.

2.10. Heart rate variability

The root mean square of successive inter‐beat (RR) interval differences (RMSSD) was calculated as a measure of vagally mediated HRV (Penttila et al., 2001). During the task performance, HRV was extracted for each trial and averaged across blocks. To adjust for deviations from a normal distribution and unequal variance, RMSSD was logarithmically transformed (natural log).

To assess the effects of taVNS during task performance, linear mixed models were used, including the factors Stimulation (taVNS vs. sham), Task (HCT vs. SCT), and Block (Block 2 vs. Block 3) as well as their interaction. Participant intercepts were modeled as random effects,

$$\mathsf{\text{LogRMSSD}}~{\mathsf{\text{Task}}}^{*}{\mathsf{\text{Stimulation}}}^{*}\mathsf{\text{Block}}+\left(\mathsf{1}|\mathsf{\text{Participant}}\right)$$

If interaction effects were found (or exploratory analysis conducted), they were followed up by pairwise post hoc comparisons using lsmeans (Lenth, 2016).

2.11. Mediation analysis

To test whether the relationship between the detection of cardiovascular signals (i.e., IAcc) and allostatic regulation (i.e., HRV) was mediated by interoceptive prediction error processing (i.e., HEP amplitudes), we performed mediation analysis during the HCT for the taVNS and sham condition, separately. We extracted values of each measure at a trial level (i.e., to increase the number of observations) and modeled IAcc scores as predictor variable, RMSSD as outcome variable and HEP amplitude increase as mediator variable. The total effect of IAcc on HRV was calculated by regressing RMSSD scores on IAcc (path c). HEP amplitude increase (interoceptive prediction error processing) was regressed on IAcc (path a), and HEP amplitude increase was regressed on HRV, while controlling for the effect of IAcc (path b).

3. RESULTS

3.1. Self‐report side effects, heart rate, and blood pressure

To test for potential side effects related to the stimulation, at the end of each session, participants reported on a series of potential physical symptoms on a scale from 1 to 7 (results are described in detail in the Supplementary Material Section S2).

Overall, reported symptoms were low (scale from 1 to 7; M_taVNS = 1.79, SD_taVNS = 1.28, M_taVNS = 1.68, SD_taVNS = 1.21) and did not differ between conditions. The only exception was a trend difference in reports in ear irritation (b = 0.57 (0.30), t (75) = 1.87, p = .06), which was slightly higher in the taVNS (M = 1.71, SD = 1.35) than in the sham condition (M = 1.27, SD = 0.58).

Tonic heart rate and blood pressure levels were measured before and after stimulation. Overall, no significant modulation of stimulation over time (pre vs. post stimulation) was observed in any of the measures (ps > .23). Results are described in detail in the Supplementary Material Section S3.

3.2. Interoceptive accuracy

Figure 1b shows the mean accuracy results in the HCT during session 1 and session 2, divided by blocks without (Block 1; NO) and with stimulation (Blocks 2 and 3 averaged, ON; see Figure 1a for design details; see also S5 from Supplementary Material for depiction of the SCT). Table 1 contains the descriptive values of the HCT and SCT.

TABLE 1.

Mean (SD) accuracy scores (log transformed) in the HCT and SCT.

		NO stimulation HCT	Stimulation ON HCT	Increase	NO stimulation HCT	Stimulation ON HCT	Increase
taVNS	Session 1	2.67 (1.48)	2.59 (1.53)	−0.32 (0.72)	4.20 (0.29)	4.23 (0.23)	0.03 (0.25)
	Session 2	2.26 (1.01)	1.81 (1.27)	−0.45 (0.93)	4.35 (0.23)	4.33 (0.20)	−0.026 (0.13)
Sham	Session 1	2.56 (1.06)	1.91 (1.27)	−0.65 (0.79)	4.24 (0.24)	4.31 (0.21)	0.07 (0.19)
	Session 2	2.49 (1.67)	2.23 (1.59)	−0.29 (0.65)	4.27 (0.20)	4.25 (0.21)	−0.014 (0.08)

Abbreviations: HCT, heartbeat counting task; SCT, second counting task.

We observed a main effect of Stimulation, b = 0.56 (0.23), t(209) = 2.40, p = .02, indicating overall higher accuracy under taVNS compared to sham. Task × Stimulation effects were also found, b = −0.75 (0.29), t(258) = −2.54, p = .01, as well as Stimulation × Session effects, b = 0.76 (0.36), t(258) = −2.03, p = .04. Most critically, a triple interaction Task × Stimulation × Session was observed, b = 1.05 (041), t(258) = 2.53, p = .016 (see for details in Section S4 of the Supplementary Material). Following‐up on this triple interaction, we compared sham and stimulation performance across sessions for each task, separately (see Figure 1c). We observed that in the first session, the decrease in IAcc was stronger in the sham than in the taVNS condition, first sham vs. first taVNS: t(114) = −3.70, p < .001, ES = −1.03, CI[−1.54, −0.47] but in the second session, no differences were observed between conditions, second sham vs. second taVNS: t(113) = 0.72, p = .47, ES = 0.20, CI[−0.35, 0.76]. Interestingly, those participants who received taVNS in the first session showed a similar IAcc as in the second session in which they received sham stimulation, taVNS first session vs. sham second session: t(270) = 1.17, p = .24, ES = 0.26, CI[−0.18, 0.69]. However, participants who received sham in the first session showed a lower decrease in IAcc in the second session in which they received taVNS stimulation: sham first session vs. taVNS second session: t(264) = −2.38, p = .02, ES = −0.57, CI[−1.05, −0.09]. Additionally, session differences were observed in the sham condition. Those participants receiving sham stimulation in the second session (presumably benefitting from taVNS during the first session) showed lower IAcc decrease than those who received sham stimulation in the first session, sham first session vs. sham second session: t(117) = −2.51 p = .001, ES = −0.77, CI[−1.35, −0.19]. No such session differences in the taVNS condition were found, taVNS first session vs. taVNS second session: t(113) = 1.66 p = .1, ES = 0.46, CI[−0.09, 1.01]. No significant differences were observed in the SCT (see also Supplementary Material S5 for results removing potential outliers).

Altogether, these findings indicate that taVNS facilitated the maintenance of IAcc levels across time. During sham stimulation, a decrease in IAcc was observed (Richter et al., 2021), however, taVNS helped maintaining IAcc levels. The session interacting effects suggest that the beneficial effects of taVNS persisted even 1 week later, when participants repeated the task (under sham stimulation).

3.3. Heart‐evoked potentials

To determine whether the task in which participants were engaged in modulated the HEPs, we applied cluster‐based permutation tests. Similar to previous studies, we found a modulatory effect of Task, irrespective of stimulation (see for details in Section S7 in the Supplementary Material): HCT versus SCT in the sham condition: t(504) = −2.43, p = .015, ES = −0.29, CI[−0.52, −0.05]; HCT versus SCT in the taVNS condition: t(504) = −2.41, p = .016, ES = −0.28, CI[−0.52, −0.04].

To investigate the effects of stimulation, HEP extracted during task performance and resting state were compared between taVNS and sham condition. Figure 2a–c depicts the Stimulation × Task interaction effects. The cluster‐based permutation test revealed an interacting effect of Stimulation and Task over frontal and central electrodes in the 0–600 ms time window, as indicated by significant masses (mass 1: 97.5, mass 2: 87.7, mass 3: 44, mass 4: 49.4) that surpassed the critical cluster mass of 38. Because there was time overlap between masses (mass 1: 272–308 ms; mass 2: 176–204 ms; mass 3 = 292–320 ms; mass 4: 176–200 ms) probably indicating that the same effects extended across sensors, we decided to focus on the two maximal masses located at Fz. In cluster 1, subsequent analysis using linear mixed models showed a main effect of Task, b = 0.42 (0.24), t(317) = 2.35, p = .02, and critically, a significant Stimulation × Task interaction, b = −0.75 (0.35), t(317) = −2.13, p = .03. No other main or interacting effects were found (Table 2; Section S6 of the Supplementary Material). Follow‐up analysis revealed that in the HCT, HEP amplitudes were reduced in the sham compared to the taVNS condition: sham vs. taVNS: t(319) = −2.22, p = .03, ES = −0.33, CI[−0.62, −0.03]. Furthermore, HEP amplitudes in the HCT under sham stimulation were reduced compared to the SCT, t(316) = −2.54, p = .01, ES = −0.37, CI[−0.66, −0.08]. No other significant effects were observed. Comparable effects could be observed when repeating the analysis using the HEP amplitude increase (blocks 2 and 3 vs. block 1) as dependent variable: HCT sham vs. HCT taVNS: t(322) = −2.28, p = .02, ES = −0.34, CI[−0.62, −0.05]; HCT sham vs. SCT sham: t(317) = −1.95, p = .05, ES = − 0.29, CI[−0.58, 0.001] (Table 2; Section S5 of the Supplementary Material). In cluster 2, linear mixed models revealed a main effect of Stimulation, b = 0.5 (0.24), t(319) = 2.04, p = .04, but the critical interaction effect did not reach significance, b = −.05 (0.35), t(316) = −1.41, p = .16. Exploratory, follow‐up analysis revealed larger HEP amplitudes in the HCT under taVNS than under sham stimulation: t(321) = −2.74, p = .006. However, these effects vanished when analysis were performed using the HEP amplitude increase as dependent variable: t(321) = −1.42, p = .15.

FIGURE 2.

Heart‐evoked potential (HEP) and mediation analysis. (a) The grand averaged HEP over Fz for transcutaneous auricular vagus nerve stimulation (taVNS) and sham stimulation, during the heartbeat counting task (HCT) and second counting task (SCT). The blue bar represents the time window, in which the permutation tests showed significant interacting effects. In the inset, the significant effects are shown in the time window between 272 and 308 ms at an adjusted amplitude. (b) The averaged HEP amplitudes over the significant sensor and time window for taVNS and sham stimulation during the HCT and SCT. (c) The HEP increase calculated as the difference between stimulation blocks (blocks 2 and 3) and no stimulation block (block 1). In (b) and (c), bar plots and dark dots represent the average scores across participants. Black lines represent standard errors. taVNS condition is colored pink and sham condition is colored gray. Pink and gray dots represent individual amplitudes. (c) The results of the mediation analysis. In the taVNS, but not in the sham condition, the positive association between interoceptive accuracy and heart rate variability (HRV) was mediated by the HEP amplitude increase.

TABLE 2.

Mean (SD) HEP amplitudes of the significant Stimulation × Task effects (time window 272–308 ms) in the HCT and SCT.

	NO stimulation HCT	Stimulation ON HCT	Increase	NO stimulation HCT	Stimulation ON HCT	Increase
taVNS	1.85 (1.70)	2.22 (1.53)	0.36 (1.53)	1.87 (1.61)	2.05 (1.29)	0.13 (1.46)
Sham	1.91 (1.68)	1.79 (1.48)	−0.12 (1.38)	1.92 (1.61)	2.23 (1.84)	0.32 (1.29)

Abbreviations: HCT, heartbeat counting task; SCT, second counting task; taVNS, transcutaneous auricular vagus nerve stimulation.

To ensure that the observed interactions did not reflect taVNS effects on ECG activity, the same analysis was repeated for the ECG waveform in the same time window (272–308 ms), but no significant effects were observed (ps > .26).

Cluster‐based permutation tests revealed significant effects of stimulation during the resting‐state phase in one cluster (mass = 79.6, time window: 520–556 ms) that surpassed the critical cluster mass of 40 at the Poz electrode. However, subsequent linear mixed model analysis did not reveal any Stimulation effects, b = −0.85 (0.59), t (48) = −1.45, p = .15.

3.4. Heart rate variability

To assess the effects of stimulation on HRV, log RMSSD was calculated for each trial, block, and participant. No main effect of Stimulation was found, b = −0.04 (0.04), t(308) = −0.87, p = .38, and no Stimulation × Task, b = 0.04 (0.06), t(308) = .63, p = .53, Stimulation × Block, b = 0.09 (0.06), t(308) = 1.59, p = .11, or Stimulation × Task × Block interaction effects were observed, b = −0.09 (0.08), t(308) = −1.11, p = .27 (results are described in detail in the Supplementary Material, Section S8).

3.5. Mediation analysis

Finally, we investigated the contribution of IAcc and HEP amplitudes on HRV by performing a mediation analysis on the HCT for the taVNS and sham condition, separately. IAcc scores were mediated as predictor variables, logRMSSD (as index of HRV) was used as outcome variable and HEP amplitude increase as mediator variable. In the taVNS condition we observed an indirect positive mediation of HEP increase on the relationship between IAcc and HRV that approached significance (indirect effects = z = 1.84, p = .065), whereas the direct effects of IAcc on HRV showed a negative trend (direct effects = z = −1.73, p = .08). These effects were related to a negative relation between IAcc and HRV (c = −0.09, p = .08), but a positive relation between IAcc and HEP increase (a = 0.13, p = .01), and between HEP increase and HRV (b = 0.15, p = .006). On the other hand, in the sham condition, no indirect significant effects were found (indirect effects = z = −0.52, p = 0.61), but a positive direct effect of IAcc on HRV was observed (direct effects = z = 2.26, p = .02), indicating a positive relation between IAcc and HRV (c = 0.12, p = .02). No relation between IAcc and HEP increase (a = 0.03, p = .54), or HRV (b = −.05, p = .31; see Figure 2d) were observed.

4. DISCUSSION

According to predictive coding views of interoception, the organism continuously creates predictions about the most probable state of the body to ensure adaptive distribution of energetic resources. When a dissonance exists between bodily afferent signals and predictions (i.e., prediction errors), the organism may respond to reduce it. Consequently, allostatic changes may occur to ensure a proper reorganization of the resources, favoring homeostasis. In the current study, we examined the effects of taVNS on cardiac IAcc and HEP amplitudes, as correlates of interoceptive processing. Furthermore, the relation between the processing of interoceptive signals and allostatic adjustments, as measured by HRV, was also investigated. We observed that taVNS, in contrast to sham stimulation, facilitated the maintenance of IAcc levels, suggesting that taVNS helps to keep engagement to (i.e., precision of) cardiac afferent signals, in line with previous findings (Richter et al., 2021; Villani et al., 2019). Furthermore, during the interoceptive focus task, taVNS in comparison to sham stimulation increased HEP amplitudes, which possibly could indicate increased cardiac prediction error processing (Poppa et al., 2022; Richter et al., 2021). Remarkably, under taVNS—but not under sham stimulation—the positive relation between IAcc and HRV was mediated by HEP amplitudes. That is, higher IAcc was associated with higher HEP amplitude and higher HEP amplitudes, in turn led to better allostatic adjustments. These results suggest that when their precision is increased, by means of taVNS, the processing of cardiac prediction error signals exerts special influence on allostatic adjustments.

During taVNS, IAcc remained at similar levels as during the stimulation‐free baseline measurements, whereas during sham stimulation, IAcc decreased over time—from the initial stimulation‐free, baseline block to subsequent stimulation blocks. Interactions between session also showed that participants who received taVNS during session 1 showed less drop in the second session, in which no active stimulation was received, whereas participants who received sham stimulation in session 1 benefitted from taVNS during the second session. It should be noted that we did not expect a reduction in IAcc over time. It could be that experimental conditions (i.e., dimly lit, silent room, length of the design) and the repetitiveness of the task produced fluctuations of attention toward interoceptive signals, especially in later blocks of the task, leading to a decrease in IAcc. These changes were also unrelated to interindividual differences in IAcc: the IAcc increase was not associated with baseline levels of IAcc (Supplementary Material S4). This reduction, is, however, in line with previous studies observing a reduction in IAcc with time, unless manipulations to increase the internal focus of attention (e.g., providing feedback about the performance) are carried out (Quadt et al., 2021; Schillings et al., 2022). Importantly, our findings concord with recent reports on the effects of taVNS on IAcc. When baseline measures are considered, taVNS has been shown to facilitate the maintenance (prevent reduction) of IAcc levels (Richter et al., 2021). Considering that IAcc may reflect the extent of precision with which interoceptive signals are detected (Ainley et al., 2016), we conclude that taVNS may facilitate the maintenance of attentional resources toward cardiac afferents, and thus the precision to cardiac prediction error signals. The unexpected interacting effects with session suggest that the effects of taVNS on IAcc may extend beyond the stimulation period. Long‐lasting effects of taVNS on interoceptive processing, however, should be addressed in future studies.

For HEPs, taVNS interacted with task, such that in the interoceptive‐focus, but not in the control task, taVNS increased the HEP amplitudes compared to sham stimulation. Such increase was also observed in relation to the stimulation‐free baseline, indicating that taVNS enhances the processing of interoceptive signals when they are prioritized (i.e., as it is the case in interoceptive‐related tasks). We can, thus, conclude that when interoceptive signals become relevant, taVNS promotes the processing of interoceptive prediction error signals. Contrary to previous studies (Poppa et al., 2022; Richter et al., 2021), no other main effects of stimulation emerged either during task performance or resting state. Methodological differences related to the task, but also to the EEG signal may have hindered the replication of previous findings. In the current study, we used 5 EEG sensors located at the central electrode line (from Fpz to COz). The decision to use central electrodes was justified by a recent meta‐analysis, suggesting that the HEP are typically extracted from central‐frontal regions (Coll et al., 2021). However, previous studies found taVNS effects on HEPs over other, more lateralized, regions (Poppa et al., 2022; Richter et al., 2021). The fact that we do not replicate previous results could therefore be due to the limited number of EEG sensors used in the present study.

In line with a recent experimental work (Poppa et al., 2022) and meta‐analysis (Wolf et al., 2021), taVNS did not show any modulatory effects on HRV. However, in the taVNS—but not in the sham—condition, mediation analysis showed that the positive association between IAcc and HRV was mediated by HEP amplitudes. These results suggest that by increasing the precision of cardiac afferents, taVNS enables the impact of cardiac prediction error signals on allostatic adjustments. It is well‐known that the vagus nerve conveys afferent cardiovascular information (Hainsworth, 1995; Prescott & Liberles, 2022; Rajendran et al., 2023) to the brain via the NST (Ran et al., 2022). From the NST, afferent interoceptive information flows continuously and hierarchically between higher‐level networks involved in the multimodal representation of the organism and lower‐level networks that send efferent signals to regulate energy expenditure (Rajendran et al., 2023). Recent neuroimaging studies have consistently shown that taVNS exerts special influence on this pathway, e.g. by increasing activity of the NST (Frangos et al., 2015; Teckentrup et al., 2021; Yakunina et al., 2017) and modulating the coupling between interoceptive signals and brain activity (Müller et al., 2022). Notably, the NST also projects to the locus coeruleus (LC), the main source of norepinephrine in the brain, which connects to a wide variety of cortical and subcortical regions, including the insula. Activity of the LC has been linked to attentional allocation processes that ensure an optimal adaptation to changing environments (Aston‐Jones & Cohen, 2005; Nieuwenhuis et al., 2005; Sara & Bouret, 2012). This adaptation requires the precise detection of prediction error signals that facilitate the update of predictions, a process that has been associated with phasic LC release (Sales et al., 2019). It could thus be that enhanced processing of cardiac signals under taVNS was due to increase activity of the LC, leading to a more precise detection of cardiac interoceptive prediction error signals in regions such as the posterior insula. It should also be mentioned that the mechanism of action of taVNS might be rather broad, affecting also other neurotransmitter systems (Colzato & Beste, 2020). Thus, our interpretation warrants further investigation. For instance, future studies examining how taVNS modulates the connectivity between the LC and interoceptive‐related regions such as the posterior insula may provide more insights into the neural mechanisms subserving the current findings.

The current findings are in line with predictive processing accounts of interoception and open new venues for the development of intervention programs focused on increasing interoceptive abilities that impact allostatic regulation (Weng et al., 2021). The latter is of particular importance given recent evidence that deficits in the precision with which interoceptive signals are detected are associated with a wide variety of clinical disorders (Brand et al., 2023; Smith et al., 2020; Smith et al., 2021; Ventura‐Bort, Wendt, & Weymar, 2021).

Despite the novelty of the current investigation, there are some considerations that need to be mentioned. To measure IAcc, we used the well‐known HCT (Schandry, 1981). Although the validity of this task has been recently criticized (Desmedt et al., 2018; Zamariola et al., 2018) (for an in‐depth discussion see Ainley et al., 2020; Zimprich et al., 2020; Corneille et al., 2020), we ensured that the present HCT task followed the current recommendations to overcome these limitations (Murphy et al., 2018). Nevertheless, replicating the current findings, using other interoceptive tasks that may extract more specific parameters of interoception (Legrand et al., 2021; Smith et al., 2020; Smith et al., 2021) may shed more light on the mechanisms underlying the effects of taVNS on interoception. Our mediation analysis indicated that the relation between IAcc and HRV was mediated by HEP amplitudes, but from the current analysis, we could not assess the bidirectionality between allostatic adjustments and the processing of cardiac signals. Assessing the dynamic interplay between allostatic adjustments and processing of cardiac activity (Candia‐Rivera, 2023), may help extend our knowledge about the dynamics between prediction error processing and allostatic changes.

In summary, in the current study, we could show that taVNS helps maintain attention toward cardiac signals and enhances their processing, which in turn mediates the relation between IAcc and allostatic adjustments. These results suggest that taVNS enhances the precision of cardiac interoceptive prediction error signals, enabling their impact on allostatic adjustments. These findings provide initial evidence for predictive processing models of interoception and open new venues for potential intervention treatments that focus on increasing the precision of prediction error signals to improve allostatic adjustments.

AUTHOR CONTRIBUTIONS

Carlos Ventura‐Bort: Conceptualization (lead), methodology, formal analysis, data curation, writing—original draft, writing—review and editing, visualization, project administration. Mathias Weymar: Conceptualization (supporting), writing—review and editing, funding acquisition.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interests.

Supporting information

DATA S1 Supporting Information.

Ventura‐Bort, C. , & Weymar, M. (2024). Transcutaneous auricular vagus nerve stimulation modulates the processing of interoceptive prediction error signals and their role in allostatic regulation. Human Brain Mapping, 45(3), e26613. 10.1002/hbm.26613

DATA AVAILABILITY STATEMENT

Data and code of the current study are available on Open Science Framework (https://osf.io/bw38g/).