Gut Feelings: Vagal Stimulation Reduces Emotional Biases

—The vagus nerve is a key physical constituent of the gut–brain axis. Increasing attention has recently been paid to the role that the gut, and the microorganisms inhabiting it, play in emotion and cognition. Animal studies have revealed the importance of the vagus nerve in mediating communication between the gut microbiome and the central nervous system, resulting in changes in emotional behaviour. This has renewed interest in understanding the role of vagal signalling in human emotion, particularly since human studies have also shown that alterations in gut microbiome composition can aﬀect emotion. While stimulating the vagus nerve can help treat some cases of severe depression, here we investigate whether vagal aﬀerent signalling can inﬂuence emotional processing in healthy subjects. We use the dot-probe task to determine the eﬀect of transcutaneous vagus nerve stimulation on attentional biases towards emotional stimuli in 42 volunteers. Participants received both active and sham treatments using a within-subject design. We show that transcutaneous vagus nerve stimulation reduces the emotional bias towards faces expressing sadness and happiness, indicating a decrease in emotional reactivity. While our novel ﬁndings reveal the eﬀect that vagal signalling can have on emotional biases in healthy subjects, future studies should seek to develop our understanding of the ways in which the microbiome interacts with, and stimulates, the vagus nerve. Since we ﬁnd a reduction in emotional bias, most notably towards sadness, this may partly account for the eﬀective use of vagus nerve stimulation in treatment-resistant depression. While its clinical application currently involves surgical stimulation, our results support the potential beneﬁt of transcutaneous vagus nerve stimulation as a non-invasive, intermittent adjunctive therapy for patients with depression, given its frequent association with emotional biases. (cid:1) 2022 The Authors. Published by Ltd on behalf of IBRO.


INTRODUCTION
Emotional biases and impairments in emotion regulation are commonly observed in people suffering from depression or anxiety (Beck, 2008;Fossati, 2008;Cisler and Koster, 2010). Such psychological conditions are multifactorial, with both genetic and environmental risk factors playing a part. Recently, there has been growing interest in the role played by the gut and its microbial environment, known as the gut microbiome, especially given the evidence that it interacts with the host's central nervous system, influencing neurochemistry, emotion, cognition and behaviour (Sampson and Mazmanian, 2015;Sharon et al., 2016;Cryan et al., 2019). In animals, the gut microbial community has been shown to alter emotional behaviour and its associated neurochemical pathways, while there is an increasing number of human studies linking the gut microbiome to personality and emotion (Johnson, 2020;Johnson and Burnet, 2020;Sarkar et al., 2020). Furthermore, neuroimaging has shown that consumption of probiotics in healthy volun-teers can alter neural signatures of emotion (Tillisch et al., 2013;Bagga et al., 2018). For instance, probiotic administration over a period of four weeks was found to modulate brain activity, leading to decreased activity of brain regions involved in emotional processing when participants were presented with negative facial expressions (Tillisch et al., 2013).
There are a number of mechanisms that may mediate this interaction between the gut, its microbiota and the brain, including communication via the nervous, immune and endocrine systems (Collins et al., 2012;Johnson and Foster, 2018;Cryan et al., 2019). In terms of neural signalling, the vagus nerve provides a direct link between the gut and the brain. While it consists of both motor (i.e. efferent) and sensory (i.e. afferent) nerve fibres, approximately 80% of the nerve fibres are afferent, transmitting information to the brain from viscera including the gut (Bonaz et al., 2019). The nucleus tractus solitarius in the brainstem receives around 95% of vagal afferents, with secondary projections to limbic and cortical structures (Nemeroff et al., 2006;Yakunina et al., 2017). The importance of the afferent vagus nerve in emotional processing and mood disorders is apparent from animal studies. A recent study in mice, for example, showed that the https://doi.org/10.1016/j.neuroscience.2022.04.026 0306-4522/Ó 2022 The Authors. Published by Elsevier Ltd on behalf of IBRO. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). antidepressant effect of oral selective serotonin reuptake inhibitors (SSRIs) depends on afferent vagal activity (McVey Neufeld et al., 2019); exposure of the gut to SSRIs increased vagal afferent activity whereas blocking vagal signalling from gut to brain via vagotomy nulled all behavioural antidepressant effects. Hence, the authors conclude that mood disorders may be treated by vagus nerve stimulation. Animal studies also show afferent activity of the vagus nerve is modulated by the internal environment of the gut since nutrients, peptides and the microbiota can all interact with vagal afferents and influence subsequent brain function and behaviour (Bonaz et al., 2019). For instance, similar to SSRIs, the probiotic Lactobacillus rhamnosus was shown to reduce anxiety and depressive-like behaviours in mice only when the vagus nerve was functionally intact (Bravo et al., 2011) and this probiotic has been found to increase the firing rate of vagal afferents (Pe´rez-Burgos et al., 2013). Administration of Lactobacillus reuteri in mice has also been shown to increase plasma oxytocin levels via the vagus nerve (Poutahidis et al., 2013).
Given the evident role of the vagus nerve in bringing about antidepressant and anxiolytic effects in animals, vagal signalling has also been hypothesised to potentially influence human emotion and behaviour (Fu¨lling et al., 2019). Since we know that the gut microbiome stimulates the vagus nerve, investigating the effects of vagal stimulation is relevant to understanding the ways in which the gut environment may be able to affect emotion (Fig. 1). As it is not possible to measure afferent vagal tone in humans, we can instead investigate the effect of externally applied stimulation of the vagus nerve. Vagus nerve stimulation is indeed prescribed for severe cases of treatment-resistant depression (Rush et al., 2005;Nemeroff et al., 2006;Aaronson et al., 2013), indicating that the vagus nerve has some ability to modulate human emotion. However, it is not currently known whether vagal stimulation can detectably influence emotional biases in healthy subjects, or whether in fact its effect on emotion is limited to those suffering from depression.
Manipulating vagal signalling in healthy humans using transcutaneous vagus nerve stimulation (tVNS) provides a tool to study the role of the afferent vagus nerve in human emotional processing (Dietrich et al., 2008;Frangos et al., 2015). This non-invasive method of Fig. 1. Understanding vagal stimulation in the context of the microbiome-gut-brain axis. Many studies in humans and animals have shown that the gut microbiome can affect emotion (right arrow), most commonly by using probiotics, prebiotics or antibiotics to alter the gut microbiome or by transferring microbiota between individuals. Both animal studies and in vitro studies have demonstrated that the gut microbiota and their metabolites can stimulate the vagus nerve (left lower arrow). Given studies in animals showing that the vagus nerve can mediate the effects of the gut microbiome on emotional behaviour, it is commonly stated that vagal signalling may play a role in the human microbiome-gut-brain axis too. However, it is unknown whether vagal stimulation can affect emotion in healthy humans (left upper arrow) which is what we seek to determine in this study. stimulation can be applied to an area of the ear known as the auricular concha which is rich in vagal afferents distributed near the skin surface (Peuker and Filler, 2002). Importantly, tVNS has been shown to trigger the same neural pathways as invasive stimulation of the vagus nerve (Assenza et al., 2017). Imaging studies in healthy humans have confirmed that tVNS activates the afferent vagus nerve (Frangos et al., 2015;Yakunina et al., 2017), which then conveys the signal to the nucleus tractus solitarius and from there to the insula (Kraus et al., 2007;Dietrich et al., 2008). Notably, the insula is a key neural structure involved in emotional processing (Adolphs, 2002) and has projections with the orbitofrontal and anterior cingulate cortices (Mayer et al., 2006). Previous studies have shown that tVNS can have a beneficial effect on well-being (Kraus et al., 2007) and can influence emotion recognition Steenbergen et al., 2021), reward sensitivity (Neuser et al., 2020) and learning and memory (Jacobs et al., 2015;Hansen, 2019;Weber et al., 2021).
Here we test the hypothesis that afferent vagal signalling influences people's emotional biases using non-invasive vagus nerve stimulation during an emotional faces dot-probe task. The dot-probe task is one of the most common tasks to study emotional attentional biases in individuals suffering from, or at increased risk of, mood disorders. Results show these individuals process both negative stimuli (i.e. increased vigilance) and positive stimuli (i.e. avoidance or inhibition) in a biased way (Winer and Salem, 2016). Biases have also been found to depend on the specific nature of the negative or positive stimulus, for example a bias towards sad but not angry faces is commonly found in depressed patients (Gotlib et al., 2004). Assessing the dot-probe task in healthy volunteers has been proposed as a sensitive measure to detect subtle anxiolytic and antidepressant effects (Pringle and Harmer, 2015). By applying a within-subject sham-controlled design, we therefore investigate the effect of tVNS on the processing of negative and positive information in healthy volunteers.

EXPERIMENTAL PROCEDURES Participants
The required sample size was estimated using a power analysis (parameters: power = 0.8, Cohen's d = 0.5, alpha = 0.05), assuming a moderate effect size based on previous tVNS protocols . While the estimated sample size was 34, to provide a buffer the study sought to recruit approximately 40 participants for a within-subject design with two groups (active and sham). Participants were recruited through an online university recruitment system, calling for volunteers to participate in a two-session study on the effect of brain stimulation on social decisions in exchange for institutional participation credit or monetary renumeration. Following consensus (Farmer et al., 2021) and previous protocols Steenbergen et al., 2020), participants were screened and considered suitable for participation if they met the following criteria: aged between 18 and 30 years old; no self-reported excessive (>25 units per week) alcohol use; no use of soft or hard drugs in the past month; not pregnant; no gastrointestinal disease; no cardiovascular disease; no use of any psychoactive medication; no mental or physical disability that could hinder participation; no history of neurological or psychiatric disorders; no history of brain surgery; no intracranial metal implants; no pacemaker or other implanted device; no history of stroke or brain injury; no recent experience of, or susceptibility to, fainting, panic attacks, dizziness or headaches; no claustrophobia; no epilepsy or firstdegree relative with epilepsy; and no skin abnormality, such as eczema, in the left ear. While 42 healthy individuals were considered eligible and gave consent for participation, one participant dropped out and the dot-probe recording failed for three participants in the second session, resulting in 38 participants considered for the main analyses (Table 1).

Procedure and design
The current study followed a two-session within-subject, counterbalanced active-sham design; participants received sham or active stimulation in a first session and the opposite stimulation in a second session, separated by at least seven days. All participants received verbal and written explanation of the procedure and possible side effects (e.g. vibrating, itching or a slight burning sensation of the skin below the electrodes). To minimise expectation effects, information was not provided about the type of stimulation applied in the specific sessions nor about the expected direction of effects.
Upon arrival to the first session, participants read and signed the informed consent, following which their height and weight were assessed (using an OMRON Karada Scan) to calculate body mass index (BMI). Participants were then placed in an individual sound-attenuated cubicle, instructed to turn off mobile devices and asked to rate on four visual analogue scales (ranging from 0 to 100) how anxious, nervous, insecure and stressed they felt. These scales provide a simple and quick way to measure the participants' feelings at various timepoints, notably before, during and after the stimulation. Stimulation was applied from 15 min before the start of the computerised tasks until their completion, resulting in a stimulation period of approximately an hour. Consistent with previous protocols Steenbergen et al., 2020), participants were asked to fill out a number of questionnaires during the first 15 min of stimulation, referred to as the waiting period. These questionnaires were used to infer mood (Positive and Negative Affect Scale (Watson et al., 1988), PANAS), interpersonal reactivity (Interpersonal Reactivity Index (Davis, 1980(Davis, , 1983, IRI), empathy (Empathy Quotient (Baron-Cohen and Wheelwright, 2004), EQ), autistic-like traits (Autism Spectrum Quotient ( Baron-Cohen et al., 2001), AQ) and alexithymia (Bermond-Vorst Alexithymia Questionnaire (Vorst and Bermond, 2001), BVAQ-40). Whereas the PANAS was filled out during both sessions to assess current mood state, the other questionnaires were filled out only once since they are assumed to reflect trait measures (specifically, the IRI in session 1 and the EQ, AQ and BVAQ in session 2). This battery of questionnaires has been used in similar previous tVNS procedures Steenbergen et al., 2020). Our rationale to include them here was two-fold: firstly to assess the variation in these traits present in our cohort of healthy participants and secondly to conduct an exploratory analysis to determine whether these scores (assessing interpersonal reactivity, empathy, autistic-like traits and alexithymia) are related to emotional biases. Just before starting the computerised tasks, participants were again asked to rate how anxious, nervous, insecure and stressed they felt using the visual analogue scales. Next, participants were asked to perform three tasks, the order of which was balanced but kept consistent for individual participants over the two sessions. Performance on the dot-probe task will be evaluated here, while results from the other two tasks (a facial emotion recognition task and a visual approachavoidance task) will be reported elsewhere. After completion of these tasks, stimulation was stopped and participants again rated how anxious, nervous, insecure and stressed they felt. This was followed by a questionnaire rating on a scale from 1 to 5 any minor physical sensations experienced: headache, neck pain, nausea, stinging sensation under the electrodes, burning sensation under the electrodes, muscle contractions in the neck or face, and generic uncomfortable feelings. Lastly, participants were asked to guess what stimulation (active or sham) they had received. Upon completion of the second session, participants were debriefed and reimbursed.
The procedures followed the ethical standards of the 1975 Declaration of Helsinki (World Medical Association, 2013) and subsequent amendments, and were approved by the local ethics committee (CEP18-0816317, Psychology Research Ethics Committee, Institute of Psychology, Leiden University).

Transcutaneous vagus nerve stimulation (tVNS)
The NEMOSÒ tVNS instrument, consisting of two titan electrodes attached to a gel frame and connected to a wired neurostimulating device (CM02, Cerbomed, Erlangen, Germany), was used to stimulate vagal afferents. Since efferent fibres of the vagus nerve modulate cardiac function, cardiac safety has always been a concern in the therapeutic use of vagus nerve stimulation (Kraus et al., 2007;Sperling et al., 2010). Hence, following previous protocols (Maraver et al., 2020; Steenbergen et al., 2020), we stimulated vagal afferents located at the cymba concha of the left ear to avoid efferent heart interference (Nemeroff et al., 2006;Kreuzer et al., 2012). Following the same previous protocols, in the sham condition electrodes were placed at the centre of the left earlobe instead of the cymba concha. In both the active and sham condition (Fig. 2), stimulation followed a schedule of 30 s on and 30 s off, with an intensity of 0.5 mA and delivered in pulses of 200-300 ms at 25 Hz. Emotional dot-probe task E-Prime 2.0 software (Psychology Software Tools, Inc., Pittsburgh, PA) was used to programme the task, present the stimuli and collect participants' responses. In this case we used the emotional dot-probe paradigm ( Fig. 3) to assess selective attention for emotions (emotional bias).
This task starts with participants using the mouse to click on a black dot appearing in the middle of the white screen as fast as possible. Upon clicking this dot, two pictures (a face expressing an emotion and a face with a neutral expression) are presented for 500 ms; one on the left-hand side of the screen and one on the righthand side. Immediately after, the pictures disappear and at one of the locations the probe (which is depicted as two dots) re-appears. The task of the participants is to click the probe as fast as possible. Individuals are generally faster at clicking the probe that replaces the stimulus that immediately caught their attention (usually the emotional stimulus) compared to a probe replacing the neutral stimulus. Next, a 1000 ms blank screen is presented, after which the sequence is re-run starting with a black dot appearing in the middle of the screen. Firstly, the participants had a short 10-trial practice block with irrelevant and non-emotional stimuli (i.e. fruits and flowers) before commencing the experimental task In both instances the device is held in place by an earplug in the auricle. Diagram adapted from Koenig et al., 2021. in which the pictures represented facial emotional expressions. The experimental task consisted of 120 trials comprising six expressions (sadness, fear, anger, happiness, surprise and neutral) portrayed by 20 actors that were counterbalanced regarding location of the emotional picture (left or right) and location of the probe (congruent with the emotional picture or incongruent with the emotional picture). Stimuli were stills of peak emotional expressions of eleven male and nine female actors from the Amsterdam Dynamic Facial Expressions Set ( Van der Schalk et al., 2011).
The outcome variable of this task is an emotional bias score, which is calculated separately for each emotion. The bias is calculated using the following formula: RT neutralprobe -RT emotionalprobe . Thus, the bias is calculated by subtracting participants' mean response time to probes replacing emotional stimuli (congruent trials) from their mean response time to probes replacing neutral stimuli (incongruent trials). Although no response deadline is defined, responses slower than 2000 ms were filtered out as is common practice (Fox et al., 2001;Ma et al., 2019).

Statistical analyses
All statistical analyses were conducted using R 3.2.3 software (R Development Core Team, 2015) unless otherwise stated. Paired t-tests were performed to check whether there were any significant differences in PANAS scores between the sham and active sessions and also in ratings for physical sensations recorded. Wilcoxon signed rank tests were conducted when the assumption of normality was not met (determined using the Shapiro-Wilk test).
Since participants were asked to rate how anxious, nervous, insecure and stressed they felt before, during and after stimulation, repeated-measures ANOVA analyses were conducted in JASP 0.14.1 (University of Amsterdam) with treatment (active versus sham) and time (before, during or after) as within-subject factors. This was to ensure that there was no effect of an interaction between treatment and time on participants' feelings of anxiety, nervousness, insecurity and stress as otherwise this could be a confounding factor in subsequent analyses.
Linear mixed-effects models were used to determine whether there was any effect of tVNS on emotional bias scores for the five emotions included in the dot-probe task. Since participants' reports for the strength of burning sensation were higher for the active session, the interaction term between treatment and burning sensation, as well as the main effects, were included in the model, along with participant ID as a random effect to account for the within-subject design. For each model, the Shapiro-Wilk test was used to confirm that the residuals were normally distributed and the Levene's test was conducted to check for homogeneity of variance.
An intercorrelation analysis was performed using pairwise Kendall's Tau-b correlation coefficients to determine the relationship between participants' emotional bias scores and their responses to the questionnaires assessing interpersonal reactivity, empathy, autistic-like traits and alexithymia. The P values were then adjusted for multiple comparisons using the Benjamini-Hochberg method to control for the false discovery rate (FDR). Results are considered significant if P < 0.05, unless otherwise stated.

Participant characteristics
There were no significant differences between PANAS scores for the active and sham sessions for either positive mood (t 37 = 0.448, P = 0.657) assessed using the paired t-test or negative mood (Z = À0.207, P = 0.836) assessed using the Wilcoxon signed rank test (given the non-normal distribution of negative mood scores). For the summary statistics of participants' reaction times in the dot-probe task, please see Table 2.

Participant reporting during sessions
The repeated-measures ANOVA analyses showed that there was no effect of an interaction between treatment (active versus sham) and time (before, during or after the session) on how anxious (F 2,105 = 0.268, P = 0.765), stressed (F 2,105 = 0.584, P = 0.559), insecure (F 2,105 = 0.208, P = 0.812) or nervous Fig. 3. Schematic of the emotional dot-probe task. The task presents two faces simultaneously (one with an emotional expression and the other a neutral expression), followed by a probe in place of one of the faces. The reaction time to click the probe is taken as a measure of attention. In congruent trials, the probe appears at the position where the emotional face was previously shown and for incongruent trials the probe appears at the position where the neutral face was previously shown. Reaction times from the trials can be used to assess the degree to which participants are biased towards emotional expressions (sadness, fear, anger, happiness and surprise) versus neutral ones.
(F 2,105 = 0.430, P = 0.652) the participants felt during the stimulation. Two participants failed to complete the mood visual analogue scales so data from 36 participants were included in these analyses.
In terms of the physical sensations that might be experienced during the active and sham sessions (Table 3), there was no significant difference between the intensity of sensations reported, except there was some evidence that the burning sensation under the electrodes was greater for participants in the active session (Z = À1.954, P = 0.051). However, the participants were not able to accurately guess in which session they had received the active or the sham stimulation, with an accuracy of 44.7% for the active session and 47.4% for the sham session.

Trait measures and emotional bias
The participants' self-reported scores assessing their interpersonal reactivity (IRI), empathy (EQ), autistic-like traits (AQ) and ability to identify their own emotions (BVAQ) are summarised in Table 5. Intercorrelation analysis (Fig. 5) revealed expected correlations between some of these questionnaire scores but there were few significant correlations between the emotional bias scores and questionnaire scores once the P values were adjusted for multiple comparisons. The only notable findings were in relation to bias towards surprise in the sham session where BVAQ scores were positively related (Analysing: s = 0.37, P = 0.001; Cognitive dimension: s = 0.31, P = 0.006; Total: s = 0.32, P = 0.004) and the AQ subscale measuring attention switching was negatively related (s = À0.33, P = 0.006) and these correlations were significant at FDR < 0.1, with the positive relationship between the BVAQ Analysing subscale and bias towards surprise retaining significance at FDR < 0.05.

DISCUSSION
Here we show that afferent vagal signalling attenuates emotional biases. The most notable finding is the reduction in emotional bias towards sad faces in the dot-probe task during tVNS, though we also find a reduction in bias towards happy faces. In addition, we show a reversal of the main treatment effect in participants who report feeling a strong burning side effect from the active simulation, such that they show increased bias. Biases in emotional processing are key markers of depression and anxiety and many studies have now shown that depression is associated with a stronger emotional bias towards sad stimuli (Beck, 2008;Fossati, 2008;Cisler and Koster, 2010). There is also evidence that greater bias towards sadness is exhibited by individuals at risk of depression (Joormann et al., 2007), as well as those currently suffering from it. Antidepressant pharmacotherapy has been shown to decrease the neural response to sad faces, measured using functional magnetic resonance imaging (Fu et al., 2004). Since vagus nerve stimulation can be effective for some cases of Fig. 4. Coefficient plots from linear mixed-effects models predicting emotional bias. Estimated coefficients are plotted from the five models predicting the bias for each facial emotional expression with bars indicating 95% confidence intervals. Note that the sham treatment was set as the reference level for the models such that a negative coefficient indicates decreased bias in the active stimulation. Table 4. Output from linear mixed-effects models predicting emotional bias. Results of the five models predicting the bias for each facial emotional expression are given with the estimate for the coefficient, standard error and 95% confidence limit (CL). Note that the sham treatment was set as the reference level for the models such that a negative coefficient indicates decreased bias in the active stimulation treatment-resistant depression (Rush et al., 2005;Nemeroff et al., 2006;Aaronson et al., 2013), our findings suggest that one possible mechanism of action could be through attenuating emotional biases, particularly towards sadness, which may help play a role in alleviating depressive symptoms. Indeed, a relevant avenue for further research would be to investigate whether this reduced emotional bias mediates the beneficial effect of tVNS on depression.
Notably, we did not find an effect of tVNS on mood visual analogue scales reported by the participants before, during and after stimulation. However, this is not that surprising since active stimulation was only applied for one session and the premise of the dot-probe task is indeed that cognitive effects precede subjective reporting, and thus it is a much more sensitive method to detect changes in affective processing (Pringle and Harmer, 2015). Currently, the majority of studies manipulating the gut-brain axis in humans (particularly via probiotic supplementation) focus on measuring the effects through subjective self-reports but to achieve a greater understanding more studies should seek to also incorporate tests from cognitive neuroscience (Johnson and Steenbergen, in prep).
It is noteworthy that vagus nerve stimulation decreased attentional bias towards happy faces as well as sad faces. Interestingly, antidepressants, particularly SSRIs, are often found to cause emotional blunting, with patients reporting a reduction in positive emotions as well as negative ones (Price et al., 2009;Goodwin et al., 2017). Imaging studies have revealed that tVNS results in decreased synaptic activity in the limbic system (Kraus et al., 2007. This deactivation might at least partly account for the reduced bias shown by participants towards both sad and happy faces. In addition, probiotic consumption in healthy participants for a month has been shown to reduce attention paid to negative faces and decrease the activity of brain regions involved in emotional processing such as the insula (Tillisch et al., 2013). In fact it has been suggested that probiotics may also function by attenuating emotional reactivity to both positive and negative stimuli (Sarkar et al., 2018). These findings do suggest that manipulation of the microbiomegut-brain axis, either through using probiotics or stimulating the vagus nerve, can lead to changes in the limbic system. It should be noted that while not all studies find an effect of probiotics on emotional processing or mood (Papalini et al., 2019;Bloemendaal et al., 2021), numerous meta-analyses and systematic reviews have concluded that probiotics can improve symptoms of depression and anxiety, including subclinical symptoms in healthy subjects (Pirbaglou et al., 2016;McKean et al., 2017;Wallace and Milev, 2017;Liu et al., 2019).
While we find that tVNS reduced the bias towards sad and happy faces, it had no notable effect on the participants' bias towards fearful, angry or surprised faces. We suggest some possible reasons for this but we must also acknowledge that a lack of effect of tVNS on biases towards these other emotions does not mean we can conclude there is no effect (Alderson, 2004). Since these emotional expressions are two-dimensional static images presented on a computer in a laboratory setting, participants may perceive the environment as inherently safe in both the sham and active conditions. While they may still be expected to show a bias towards the emotional stimuli, it may be less pronounced compared to if they perceived the stimuli as threatening, as should be the case with anger and fear. This may there- Table 5. Summary statistics for questionnaire data. Mean, standard deviation and range for each measure (and subscale) are given. The maximum possible range of scores are 0 to 50 for the Autism Spectrum Quotient (0 to 10 for each subscale), 0 to 80 for the Empathy Quotient, 0 to 112 for the Interpersonal Reactivity Index (0 to 28 for each subscale) and 40 to 200 for the Bermond-Vorst Alexithymia Questionnaire (8 to 40 for each subscale, 16 to 80 for the affective dimension and 24 to 120 for the cognitive dimension) fore make it less likely to detect any effect of tVNS (for a detailed discussion on the involvement of the vagus nerve and tVNS in processing of different emotional stimuli in a laboratory setting, we refer to Steenbergen et al., 2021). Additionally, regardless of stimulation, fear and surprise are typically less accurately recognised as they are often confused with one another (Roy-Charland et al., 2014), which could also affect possible biases. Perhaps it is not so surprising that no change in bias towards fear or anger was found since vagus nerve stimulation is known to be an effective treatment for cases of severe depression but there is less conclusive evidence in terms of its effect on anxiety (George et al., 2008;Burger et al., 2019). It is possible that tVNS has a greater effect on brain networks involved in the perception and experience of sadness and happiness. It would therefore be interesting for future research to repeat this study while conducting brain imaging. This may provide some insight into whether the differential effect of tVNS on the attention paid to the five emotions may be understood from the perspective of differences in brain activity during stimulation. In addition, we find an interaction effect between treatment (active versus sham) and burning sensation, a mild side effect experienced by some people during tVNS. The participants who experienced more of a burning sensation showed increased emotional bias, rather than reduced bias, during tVNS. Therefore a stronger burning sensation reversed the effect of tVNS. In fact, mild pain can alter emotional processing, particularly of sad stimuli  and negative affect has been shown to increase with physical pain (Buhle et al., 2013). However, since a stronger burning sensation also increased bias towards happy faces, perhaps this is indicative of an increase in emotional reactivity. This mild burning sensation experienced by some participants is something that could be prevented if tVNS were applied therapeutically. For the purpose of this intervention study to investigate the effect of vagus nerve stimulation on emotional bias, a uniform intensity of stimulation was used across participants. However, tVNS can also be applied at an individual intensity adjusted according to each person's sensitivity (Farmer et al., 2021).
The effect of vagal signalling on emotional bias that we report here is also notable given that the vagus nerve is considered a key component of the microbiome-gut-brain axis. Since depression is linked to an altered gut microbial community, it might be that such a change in microbiome composition results in reduced vagal tone, contributing to changes in emotional processing. While our study directly manipulated the vagus nerve, rather than the gut microbiota (since the latter would require concomitant measurement of afferent vagal tone which is not possible to directly measure in humans), it is interesting to note that previous studies on the microbiome-gutbrain axis in healthy volunteers have reported changes in emotion. For example, consumption of prebiotic supplements (selectively fermented indigestible carbohydrates such as fructo-and galactooligosaccharides which promote the growth and/or activity of certain gut bacteria) has also been found to affect emotional attention, with participants showing decreased bias towards negative information in a dotprobe task (Schmidt et al., 2015). In addition, cognitive reactivity to sad mood, as measured with the LEIDS-R questionnaire (Van der Does and Williams, 2003), was shown to decrease with probiotic consumption (Steenbergen et al., 2015). Interestingly, recent antibiotic use, which is well known to disrupt the gut microbiome and reduce its diversity, is linked to increased bias towards sadness in the dot-probe task (Johnson and Steenbergen, in review). To directly assess the role of the vagus nerve in the microbiome-gut-brain axis in humans, an especially relevant cohort for future studies would be patients who have undergone truncal vagotomy, a procedure generally performed to treat chronic ulcers. No research in this field has been conducted with these patients to date but it could provide a unique opportunity to understand the relative importance of the vagus nerve in mediating communication between the gut microbiome and the brain in humans.
Our results are also interesting in light of the clinical use of invasive vagus nerve stimulation for depression. The surgery required for vagus nerve stimulation, alongside the side effects of this treatment, restricts its use to patients who have failed to respond to various other therapies. Since these tVNS ear devices provide a non-invasive alternative to stimulation based on surgical implantation and have much milder side effects, intermittent use of tVNS might provide a suitable adjunctive therapy for depressed patients. Imaging has shown that tVNS leads to activation of the dorsal raphe, the brain's major source of serotonin, which may help explain the effect of vagal signalling on emotion (Frangos et al., 2015). In fact it has been found that brain regions to which vagal afferents project stay active even after tVNS has stopped being applied, which suggests any effects of stimulation may last beyond the tVNS session (Frangos et al., 2015). This may be one reason why frequent, intermittent use of tVNS has been shown to reduce depression severity (Hein et al., 2013;Fang et al., 2016;Rong et al., 2016), even following just a two-week period of daily stimulation (Hein et al., 2013). A study found that two tVNS treatments per day of 30 min each, carried out at least five days a week over a period of four weeks, reduced depressive symptoms and altered the functional connectivity of the default mode network in patients with major depressive disorder, particularly in terms of brain regions involved in emotional processing . This does suggest that the effect of vagus nerve stimulation on emotional processing is central to its beneficial effects.
In terms of the intercorrelation analysis, we did not find much evidence that the traits measured in the questionnaires were correlated with emotional biases, except for emotional bias to surprised faces in the sham condition which was positively related to BVAQ scores and negatively related to the AQ subscale assessing difficulties in attention switching. It is perhaps surprising that participants with higher scores for difficulties in attention switching showed a smaller bias towards surprised faces in the sham condition since one might expect poorer attention switching to be associated with increased bias. However, in general there was little evidence in our cohort that empathy, autistic-like traits, interpersonal reactivity or alexithymia were related to the emotional bias scores, perhaps because the study was conducted in healthy volunteers and thus there was not enough variation within the cohort. In contrast to our results, previous research has shown that empathy can influence the processing of emotional information with studies finding a stronger attentional bias towards positive stimuli in people with greater empathy  and emotional intelligence (Lea et al., 2018). However, the study of Liu et al. was restricted to participants scoring at either extreme of the empathy scale and both studies used eye tracking which may have facilitated detection of even very small changes in attention. In addition, while alexithymia has been linked to altered emotional processing in clinical cohorts (Donges and Suslow, 2017), there is little research in the general population on whether difficulty in identifying emotions is associated with emotional biases, although one study did investigate this and found no correlation (Lundh and Simonsson-Sarnecki, 2002).
In conclusion, our study in healthy volunteers finds evidence that vagal afferent signalling influences emotional biases, specifically by attenuating biases towards both sad and happy faces in the dot-probe task. This suggests that vagus nerve stimulation may function to reduce emotional reactivity which might be expected given that the vagus nerve reaches limbic and cortical structures in the brain that are important for mood regulation and emotional processing. Our findings demonstrate that vagal signalling can influence human emotion and so future research should expand this line of investigation, determining the importance of this neural pathway in mediating the microbiome-gut-brain axis in humans. The psychological benefits derived from tVNS should also be investigated further with the potential to prescribe intermittent tVNS to those suffering from mood disorders.