Compassion, self-compassion and psychopathology

In addition to being an innate capability, a tendency towards compassionate responding is a trainable attribute [4]. Descriptions of such training, and its benefits, are often attributed to (practitioners of) the Mahayana branch of Buddhism [5, 6]. This tradition has strongly influenced and deepened our contemporary understanding of compassion, and provided a basis for the development of compassion-focused meditation approaches, which have become increasingly mainstream and accessible, even to those with no previous background in these practices [7]. Compassionate behavior (including the subjective state of compassion) is typically understood as being directed outwards, toward other (suffering) beings. However, the value of self-compassion as a cognitive-emotion regulation strategy and an antidote to pathological self-criticism is also increasingly recognized [8]. Certain prevalent social practices (e.g. maladaptive upward social comparisons [9]) breed a kind of toxic self-criticism and shame, which in turn, are thought to contribute to the increasing prevalence of anxiety and depression, despite improved availability of treatments for these common mental health problems [10]. Established psychological treatments (e.g. ‘standard’ cognitive behavioral therapy), are generally ill-equipped to deal with this increasingly prevalent transdiagnostic phenomenon of toxic self-criticism. These limitations of extant psychological therapies have motivated the development of a ‘third wave’ of behavioural treatments which have tended to incorporate theoretical and technical insights from Buddhist psychology [11]. For example, interventions that specifically aim to foster self-compassion to counteract toxic shame and harmful self-criticism have been developed for a range of psychological disorders. Although this field is still in its infancy and the likely true effect size of compassion-focused therapies remains to be determined [12], early syntheses of the evidence suggest promising effects across a variety of psychiatric disorders and outcomes [13]. Nonetheless, as with all psychological treatments, strategies for augmenting compassion-focused therapies may be required to optimize their efficacy, or to improve outcomes for specific patient groups. Recently there has been growing interest in augmentation of psychological therapies using neurostimulation [14]. In the case of compassion-focused therapies, neurostimulation strategies that modulate the vagus nerve might be especially appropriate as adjunctive treatments with behavioural and cognitive psychotherapies.

The vagus nerve and self-regulation

The vagus nerve (VN)—a major component of the parasympathetic nervous system, and the nexus between peripheral organs and the brain—is largely comprised of sensory afferents [15], feeding brainstem structures with information about bodily states [16]. However, the VN also contains efferents originating from nuclei in the medulla that regulate cardiac, pulmonary and gastrointestinal activity. Optimal regulation of this bidirectional central-peripheral traffic—which includes the interaction between brain-stem structures, elements of the central autonomic network and the prefrontal cortex—is thought to be associated with mental well-being and has a role in a variety of emotion-regulation functions [17]. These include basic cognitive-affective capabilities such as evaluating facial expressions of emotions [18], and, based on theoretical accounts, more complex social-regulatory functions, such as generating feelings of safety [19] and promoting social engagement [20]. We direct interested readers to authoritative reviews summarizing the proposed role of the VN in social-cognitive-affective functioning [20, 21].

The VN has a critical role in regulating breathing and modulating heart rate, momentary fluctuations in which, partly reflect this modulatory influence. In fact, ‘heart rate variability’ (HRV) is a widely recognized biomarker of VN functioning and, by extension, emotion regulation [22]. Because self-compassion is itself viewed as a form of emotion regulation [23], it is perhaps unsurprising that (linear) associations between compassionate responding and HRV have been reported in a number of studies (see [24]). Although most of these studies have reported a positive association between compassion and HRV [24], many had small sample sizes, lacked adequate control conditions, used measures with unknown reliability, and did not pre-specify the relationships of interest. As such, the overall robustness of these findings remains unclear. Indeed, related studies on a variety of positively valenced, low-arousal emotional states (e.g. safeness, contentment, warmth) and positive affective dispositions (e.g. agreeableness, gratitude, meaning in life) suggest a quadratic rather than linear association with HRV [25–30]. This may also be true for HRV’s association with compassion, although few studies have examined this possibility (c.f. [27]).

To date, experimental studies examining the relationship between vagal activation and positive affect—including compassion—have tended to use top-down cognitive-behavioral strategies to induce/enhance these states while measuring HRV(e.g. [31–33]). In addition, however, bottom-up methods for activating the VN have been proposed [17, 34]; these tend to involve some form of regulated breathing or HRV biofeedback. Although such methods have been shown to influence compassion-adjacent behaviors (i.e. altruism [35]), we are not aware of published studies that have specifically examined the effects of such bottom-up behavioral VN regulation techniques upon compassionate responding.

Vagus nerve stimulation

In fact, until relatively recently, direct manipulation of VN activity was only achieved through invasive means (requiring surgical implantation of electrodes on the VN). However, the past two decades have witnessed rapid developments in non-invasive VN stimulation technologies, including transcutaneous stimulation of the VN (tVNS). This non-invasive approach is possible because of the superficial nature of vagal innervation of parts of the face and neck, including the ear. This allows mild current pulses to be safely and painlessly applied to the auricular branch of the VN through the skin of the tragus or concha and cymba conchae of the external ear. An important feature of vagal innervation of the external ear in particular is that these nerve fibers do not extend far beyond the specified areas [16]. Not only does this provide an opportunity for anatomically-targeted stimulation, but also allows stimulation of nearby areas (e.g. the earlobe) to serve as a credible sham condition (i.e. credible to participants receiving a tVNS ‘intervention’).

Given the proposed role of the VN in regulating heart rate (variability), HRV would appear to serve as a positive control and bioassay for effective (i.e. ‘successful’) vagal modulation. Additionally, through the locus coeruleus-noradrenergic pathway, tVNS might influence attentional mechanisms, the activation of which can be assessed using pupillometry [36] and other measures of attention and attentional bias (eye fixations and dwell times) in response to salient stimuli (e.g. positively valenced facial expressions). As such, various psychophysiological indices may provide objective and non-intrusive means for assessing the degree of vagal modulation by tVNS.

The ‘compassionate vagus’?

Beyond these advances in the psychophysiology of compassion, recently developed measures that are based on consensus definitions of compassion, enable (self-)compassionate ‘traits’ [37], and ‘states’ [38] to be reliably assessed. In addition, stimuli that appear to tap general ‘compassion sensitivity’ and are less reliant on introspective reports have also recently been developed (e.g. facial expressions of compassion; [39]). These various experimental advances now provide an opportunity to address—at least preliminarily—questions related to the sufficiency of VN activity in mediating compassionate responding [20]. Such peripheral, VN-mediated (bottom-up) control over complex psychological capabilities may be possible because of the bidirectional influence of VN-mediated physiological states on brain activity [17]. Although there has been a considerable amount of theoretical discussion about the role of the VN in compassion/compassionate behavior [20, 21], the empirical evidence for such a role is currently very limited [24].

Experimentally manipulating vagal activity to determine causality in compassionate responding: Study rationale

Taken together, the developments in autonomic and contemplative science, and non-invasive methods for stimulating the VN, provide a basis for testing the effects of tVNS on compassionate responding (and related outcomes). We therefore outline a study protocol for an experiment that is intended to preliminarily examine the separate and combined effects of tVNS and self-compassionate mental imagery on self-compassion, self-criticism and related outcomes (state mindfulness, positive affect and HRV). In some respects, this may appear to be a high-risk approach: the main variables of interest (self-compassion and self-criticism) are assessed using self-report measures rather than fine-grained computationally-derived metrics [40], or objective outcomes measured with millisecond precision [41], which might be more likely to detect weak signals. However, while small effects can have important mechanistic implications, their direct clinical or other applied relevance may be more limited. The adoption of bioelectronic stimulation methods—for example in clinical psychological settings—will rely on convincing demonstrations of efficacy on clinically relevant outcomes—such as those used here. Nonetheless, we will also assess ‘compassion-sensitivity’, using high-precision eye-tracking metrics, although these are secondary outcomes.

Importantly, in addition to examining the stand-alone effects of tVNS on self-compassion and related outcomes, we will test the effects of combining tVNS with a behavioral intervention (compassionate mental imagery) that itself reliably increases subjective self-compassion [33, 42–44]. In fact, according to a psychophysiological model of compassion [20], vagal activation generates the physiological conditions necessary for compassionate responding, but is insufficient for generating such responding. As such, participants who undergo active tVNS and a sham imagery intervention, rather than the active tVNS and self-compassion imagery intervention might be expected to show minimal effects on self-compassion. Alternatively, those who receive the combined tVNS + self-compassion intervention—in which tVNS is hypothesized to produce the appropriate physiological conditions for (self-)compassion—would show the largest responses (e.g. increased subjective self-compassion and decreased self-criticism) following stimulation.

By examining the effects of tVNS longitudinally, with repeated stimulation (active vs. sham) and mental imagery (active vs. sham) practice over the course of a week, we will also determine whether any augmenting effects on self-compassion and related outcomes are ‘dose-dependent’, becoming more evident after repeated stimulations. Cumulative effects might be expected to emerge if—in addition to providing an acute neurophysiological receptivity to compassion—tVNS also produces a central physiological state that is conducive to neuroplasticity [45]. If so, such effects would again only be expected if tVNS has an appropriate ‘behavioral substrate’ (i.e. the self-compassion intervention) upon which to act (i.e. the effects of stimulation are task- and use-dependent [45]). In the current study, the compassionate mental imagery intervention serves the role of this behavioural substrate. As a separate, confirmatory line of investigation, we will test the presence and nature (e.g. linear v quadratic) of the relationship between HRV and positive affect (including state/trait self-compassion). Importantly, like many ongoing and recently published tVNS studies, the current study protocol and analysis plan have been pre-registered (https://osf.io/4t9ha). In addition, as a point of clarification, while the study is registered as a clinical trial (ClinicalTrials.gov, Identifier: NCT05441774) the aims are not to examine the safety, efficacy or effectiveness of a treatment for a specific condition, nor to establish a ‘treatment dose’.

Design

This is a proof-of-concept, experimental-medicine study that examines the separate and combined effects of active auricular tVNS and compassion-focused imagery using a parsimonious 2 x 2 between-subjects factorial design, with repeated assessment of outcomes. The between-subjects factors are stimulation condition (active versus sham) and imagery condition (compassionate imagery versus sham imagery). Thus, participants are assigned randomly to one of four conditions (n = 30/condition):

• Sham (earlobe) stimulation + sham imagery (draw-a-face-in-imagination)
• Active (tragus) stimulation + sham imagery
• Sham stimulation + Active (compassion-focused) imagery
• Active stimulation + Active imagery

Participants attend two in-person lab sessions separated by one-week (Session 1/Day 1; Session 2/Day 8), during which psychophysiological (eye-tracking and HRV) measures are taken, alongside self-report measures of subjective state (see procedure description below). In addition, participants are instructed to use the tVNS device according to the condition (active versus sham) to which they were randomly assigned on Session 1, alongside their assigned imagery task at home for six sessions, between Sessions 1 and 2 (i.e. Days 2–7), during which they repeat the same subjective state measures used in the lab sessions (Fig 1; Standard Protocol Items: Recommendations for Interventional Trials, SPIRIT).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Standard protocol items: Recommendations for Interventional Trials (SPIRIT) schedule of enrolment, interventions, and assessments.

https://doi.org/10.1371/journal.pone.0282861.g001

As such, the within-subjects factor of ‘time-point’ will be used to analyse acute (within-session) effects of the stimulation and imagery manipulations (pre-stimulation, peri-stimulation; post-imagery; Fig 2), and the separate within-subjects factor of ‘day’, will be employed in the analyses of sustained (between-session) effects.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2.

Procedural sequence for the two lab sessions (top panel). PRE-STIM refers to the assessment period immediately before the stimulator is switched on. PERI-STIM (before the imagery tasks) assessments occur after 30 min of continuous stimulation in the absence of any concurrent task. POST-IMAGERY measures are taken ~15 min after the PERI-STIM measures on lab Session 1 (Day 1) and ~10 min after on Session 2 (Day 8). The different timings are a result of the briefer recorded instructions used on Session 2 (i.e. only section 3 of the audio imagery instructions is played to participants; see details under Mental Imagery Tasks). The block of ‘State’ measures at the three timepoints consists of the iPANAS, SCSC Scale, 5-items from the SMS, TPAS and VFIQ (see text for details). The bottom panel illustrates the 2 x 2 design and the four conditions used in the study. Participants complete the same stimulation and imagery procedures on both lab sessions (and during the intervening at-home sessions).

https://doi.org/10.1371/journal.pone.0282861.g002

Study setting

This is a single-center, laboratory-based study conducted in the Clinical Psychopharmacology Unit at University College London, UK.

Participants

In line with experimental medicine study design standards, participants are non-clinical (healthy) volunteers, recruited from University College London and surrounding areas using online adverts (e.g. Call For Participants). They receive £80 for taking part. Written consent is obtained and recorded electronically before any experimental procedure is performed.

A target sample size of n = 120 was deemed to be feasible given our resource constraints. A sensitivity analysis was conducted using the pwr2ppl package in R [46] with a variety of simulated mean values on the Self-Compassion-Self-Criticism (SCSC; [38]) scale. This indicated that for the final within-lab-session timepoint (post-imagery), a mean difference of ≥11.1 points (common SD = 15) between the double-sham and double-active conditions on either subscale of the SCSC could be detected using independent samples t-tests with 80% power (two-sided; α = 0.05) with group sizes of n = 30. Although an average difference of 11.1 points on the SCSC scale represents a relatively large effect size (SMD = 0.74), it is in line with group differences reported in previous studies using the self-compassion subscale of the SCSC scale following a variety of biological and behavioral manipulations [42–44]. Larger effects (reductions) on the self-criticism subscale were reported in these studies [42–44]. As such, the design may have greater power to detect between-group effects on self-criticism. A sample size of n = 60 per (stimulation) condition is also adequate to detect within-condition changes in either SCSC subscale of ≥ 5.6 points (standardized mean difference = 0.37) between pre-stimulation and peri-stimulation or between pre-stimulation and post-imagery timepoints collapsed across imagery conditions (based on two-sided, paired-samples t-tests). Of course, the strength of conclusions derived from such within-condition comparisons will be lower than between-group effects.

Inclusion criteria: participants are required to be 18–35 years old, fluent in English and have good (including corrected) vision and hearing. Exclusion criteria: participants are not included if they have any of the following characteristics: current use of any medication for a psychiatric condition, regular use of any medication used to treat a cardiovascular or inflammatory condition, use of any illicit recreational drug >2/week, regular weekly consumption >14 standard UK servings (’units’) of alcohol. Additional exclusions are: current treatment for any mental health condition, scores on screening measures of depression (Patient Health Questionnaire; 2-item version [47]) or anxiety (Generalized Anxiety Disorder rating scale, 2-item version [48]) indicating significant levels of anxiety or depression (scores on either >4), history of serious mental health problems, past or current cardiovascular disease or neurological problems, current/past chronic/recurrent facial or ear pain, skin irritation/broken skin at the stimulation site, (likelihood of) pregnancy during study, and previous adverse response to meditation. Participants self-declare an absence of these conditions during screening.

Interventions

Transcutaneous vagus nerve stimulation.

In line with minimum reporting standards for tVNS studies [49], stimulation parameters are described in Table 1. Sham stimulation of the earlobe and active stimulation of the tragus is delivered using a Parasym tVNS device (Parasym Ltd, United Kingdom). The pulse width, frequency and other parameters used here are similar to those reported in the majority of studies on cognitive effects of tVNS (see [50]). For each stimulation session, participants are instructed to increase the current level from 0 mA in one-unit increments on the tVNS device (corresponding to steps of 0.8 mA per increment, based on a fixed resistance of 500 Ohms). The stimulation level used for each participant is set separately on each of the eight lab/at-home sessions such that it exceeds the participant’s sensory threshold (the point at which they feel a clear ‘tingling’ or pulsing sensation that is not painful). Because skin resistivity can change within a stimulation session (e.g. due to changes in moisture or slight changes in the position of the electrodes), participants are instructed to adjust the stimulation level appropriately during each lab/at-home sessions such that it remains comfortably above threshold level. Stimulation intensity on the device (range 0–40) is recorded during the session.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Stimulation parameters in accordance with minimum reporting standards.

https://doi.org/10.1371/journal.pone.0282861.t001

An initial (pre-imagery, ‘offline’) 30-min active/sham stimulation period begins immediately after the block of pre-stimulation state measures. During this 30 min period of stimulation, participants are asked to relax and watch scenes from a nature documentary (Blue Planet II, 2017) which were selected to be emotionally neutral. At the 30 min mark, the imagery instructions are presented (see below) and stimulation continues throughout the imagery task (‘online’ stimulation). Approximate total stimulation periods for the lab and at-home sessions are outlined in Table 1.

Measures

Procedure

Statistical analysis

Assuming that the usual assumptions of linear modelling are met, linear mixed models (LMMs) will be used to analyze repeated measures data with >2 timepoints. LMMs are appropriate for modelling the serial dependence in outcome from the same participant and are more effective at handling missing data than analysis of variance. Random intercepts per participant will be included in models and the effects of including interaction terms and random (Timepoint or Day) slope terms will be tested. AIC and BIC minimization will be used as criteria for model selection. Univariate analyses of variance will be used to test imagery/stimulation effects at the post-imagery timepoint for the lab sessions. Specific tests of hypotheses will be performed using independent- or paired-samples t-tests of outcomes. If assumptions for linear modelling are not met (e.g. in cases of serious violations of distributional assumptions as indicated by visual inspection of QQ plots), generalized linear mixed models will be used with appropriate distribution and link functions. This analysis approach will be applied to the primary outcomes: SCSC subscales, and HRV (RMSSD and HF power), as well as the secondary subjective measures that might be responsive to active stimulation and/or compassionate imagery including the safe/content subscale of the TPAS, iPANAS-positive affect (effects on iPANAS-negative will also be assessed in exploratory analyses) and SMS 5-item total. These secondary subjective measures will also be taken pre-stimulation, peri-stimulation and post-imagery on lab Sessions 1 and 2 (Day 1 and 8) and post-imagery on Days 2–7, as for the primary measures.

Analysis of eyetracking metrics will involve comparing neutral expressions with high intensity (100%) compassion expressions, with Stimulation and Imagery condition as independent variables. Additional analyses will include ‘Intensity’ (0, 25, 50, 75, 100%) as an additional explanatory variable.

Two-tailed tests will be used, with a p value threshold of ≤0.05. The latter might be considered liberal in light of the number of pre-specified primary outcomes (self-compassion, self-criticism, RMSSD and HF power). However, given the preliminary nature of the study and the risk of type II errors, no downward adjustment to the α level was pre-specified.

Further details on analyses, including exploratory approaches are described on the Open Science Framework (https://osf.io/4t9ha).

Limitations of the study design

There will likely be unknown/unexplained sources of variance in our data that could substantially affect our ability to detect the effects of tVNS. For example, there appears to be substantial individual anatomical variability in vagal innervation of the tragus, and in the number of myelinated axons in the auricular branch of the VN. An absence of reliable effects of tVNS, particularly on the positive control variables, namely HRV, might reflect ineffective stimulation of the VN in some individuals. While a number of studies have shown that active tVNS increases HRV [71–73], other researchers have raised doubts about whether increased HRV is indeed a reliable indicator of successful stimulation [62, 74] (in fact some researchers even doubt that HRV reflects vagal tone: [75]). Part of the difficulty in establishing reliable biomarkers of tVNS efficacy is the inordinately large number of combinations of stimulation parameters that might affect successful stimulation, and by extension, HRV.

A strength of many published tVNS studies is their use cross-over designs. Although such within-subjects designs are desirable from a statistical noise-reduction perspective and are feasible if all conditions can be performed within a single or small number of session(s), our use of repeated sessions of stimulation and imagery makes it highly impractical to employ a cross-over design. In particular, it would be very challenging to recruit and retain participants to perform daily tasks that would become highly repetitive and burdensome over the required period. As such, on balance, we decided that a between-subjects design was necessary, although this is only powered to detect relatively large between-group effects.

Our design is based on the assumption that the stimulation location, and other stimulation parameters, as well as our selection of measured variables are the ‘correct’ ones for generating and detecting the effects of tVNS. It is possible that slightly different stimulation parameters or targeting the cymba conchae or cervical branch of the VN would be more effective. Indeed, even supposing that the tragus is the optimal locus of stimulation, inter-individual anatomical variability in vagal innervation [16], might contribute to sufficient statistical noise to obscure group level average differences. Our current state of knowledge is too limited to have high confidence that any single set of parameters is optimal for any particular cognitive-affective task, let alone for enhancing compassion, for which there is no precedent in the existing literature. We therefore use the manufacturer default settings of the tVNS device, which in one study were shown to increase HRV [72], and have been found to have therapeutic effects in clinical trials, though on unrelated outcomes (e.g. [76]). We believe that this is at least a reasonable starting point for testing the effects of interest.

Unlike the increasing use of double blinding in tDCS studies, it is still common for tVNS studies to be single-blind. Although researchers’ influence on participants’ responses is minimized by the limited and scripted nature of their interaction with participants in the current study, the researchers in the current study are not blind to treatment allocation. Moreover, although the study aims are nominally masked from participants, there was no way within our design to protect against curious and diligent participants from learning about the nature of active versus sham stimulation, and about the nature of the experiment as a whole.

These limitations notwithstanding, we believe the study protocol provides an opportunity to test the direct, causal role of the VN in modulating compassionate responding. Even if effects are relatively small (and statistically non-significant) but nonetheless point to a possible additive/synergistic effect between compassionate imagery and active stimulation at a descriptive level (i.e. largest effects on self-compassion/self-criticism or other outcomes in the active tVNS + active/compassionate imagery condition), this would still provide valuable information (e.g. likely effect size; informativeness of control groups) for planning future related studies. Alternatively, a more convincing demonstration of synergy/additivity between the two interventions might provide the basis for preliminary clinical testing of this combination strategy, for example in participants with high levels of self-criticism/low self-compassion.