Neural networks involved in pain processing are intimately linked to the autonomic nervous system (ANS) (Benarroch, 2006): On the one hand, the body’s response to pain is defined by changes in ANS parameters (Kyle and McNeil, 2014); on the other, alterations in autonomic arousal can also influence the experience of pain (Terkelsen et al., 2004). There is growing interest in mindfulness-based and other mind-body interventions in the treatment and management of pain (Stanos, 2012; Goyal et al., 2014), with changes in autonomic balance one of the likely underlying mechanisms of action (Tang et al., 2015). Whilst various regions of the central nervous system (CNS) are known to play a role in both pain and the ANS, currently there is a lack of knowledge concerning how pain-autonomic interactions may be reflected by functional connections in the brain.

One possible mechanism underlying this pain-autonomic interaction is the baroreflex, the negative feedback loop used to maintain stable blood-pressure (Suarez-Roca et al., 2018). This mechanism has been associated with observed reduction in pain perception in healthy controls during spontaneous high blood pressure (during which baroreceptors are activated) and also during mechanical stimulation of baroreceptors (Edwards et al., 2001; Duschek et al., 2007; Reyes del Paso et al., 2014). Decreased baroreceptor sensitivity has also been described in some chronic pain conditions (Davydov et al., 2018).

Heart rate variability (HRV), which is derived from variations in interval length between consecutive heart beats (Task Force, 1996; Berntson et al., 1997; Thayer et al., 2010a, 2012), is often estimated to assess the autonomic response to experimental pain (Koenig et al., 2014). Low frequency (LF)-HRV spectral power is thought to represent the baroreflex-dependent outflow to the heart, whereas high frequency (HF) is interpreted as indicator of vagal cardiac control, and the ratio between the two measures (LF/HF) as a reflection of sympathovagal balance (Pagani et al., 1986, 2012; Montano et al., 2009; Thayer et al., 2010a; Goldstein et al., 2011; Reyes del Paso et al., 2013). Experimentally induced pain increases LF power and the LF/HF ratio (Koenig et al., 2014), indicating an increased engagement of the baroreflex. Few studies have employed tonic cold pain stimuli, but results from cold water hand immersion tests (i.e., the cold pressor test) point toward a similar tendency of increased LF-HRV (Mourot et al., 2009; Streff et al., 2010).

A pain-suppressive effect associated with larger HF spectral power at baseline has been found (Nahman-Averbuch et al., 2016; Tracy et al., 2018), which might be explained by superior capacity to engage vagal cardiac control (Reyes del Paso et al., 2011; Busch et al., 2013; Zunhammer et al., 2013; Tracy et al., 2018). Conversely, higher resting LF-HRV has been found to predict reduced thermal pain sensitivity (Appelhans and Luecken, 2008; Tracy et al., 2018). Patients with chronic pain conditions such as fibromyalgia often show reduced HF power in addition to increased LF and LF/HF, suggesting dysregulated autonomic cardiac control (Meeus et al., 2013) and altered baroreflex engagement (Bruehl et al., 2017). In brief, HRV indices respond to noxious stimuli and relate to the subjective intensity of pain, but can also predict the experience of pain, indicating a bi-directional association where nociception influences the ANS and, conversely, the ANS modulates the experience of pain.

Modern neuroimaging techniques, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET), have been used to investigate the brain regions associated with autonomic activity during various tasks and conditions. Significant correlations between HRV, amygdala and medial prefrontal cortex (mPFC) activity have been demonstrated (Thayer et al., 2012; Beissner et al., 2013; Steinfurth et al., 2018). In resting-state fMRI (rs-fMRI) paradigms, HRV measures were associated with functional connectivity of different resting state networks (RSNs). For example, Jennings et al. (2016) found participants’ HF-HRV at rest to correlate with resting state connectivity of the mPFC, but not with salience (SN) or default mode networks (DMN). Functional connectivity of the dorsal anterior cingulate cortex (dACC) to the thalamus and brainstem co-varies with HF-HRV, whilst LF-HRV relates to dACC connectivity with the temporoparietal junction (Chang et al., 2013). Further, measures of vagal output have been shown to be associated with functional connectivity between cortical areas and parts of the brainstem (e.g., Smith et al., 2015; Bär et al., 2016).

Many brain regions involved in ANS activity are also active during the experience of pain, including the anterior cingulate cortex (ACC), amygdala, and periaqueductal gray (PAG) (Leone et al., 2006; Heinricher and Fields, 2013). The dACC is consistently found to be involved in the CNS response to noxious stimulation (Vogt et al., 2003; Apkarian et al., 2005; Leone et al., 2006; Duerden and Albanese, 2013; Jensen et al., 2016). Whilst the more rostral parts are associated with the affective component of pain, the most dorsal aspect (bordering the mid-cingulate cortex) encodes the objective aspects of pain (e.g., stimulus intensity) (Rainville et al., 1999; Singer et al., 2004; Wager et al., 2004; Amodio and Frith, 2006; Boccard et al., 2014). Taken together, the dACC forms part of a network of brain regions involved in the detection of salient sensory events, including the multimodal context-dependent experience of pain, but is also intimately linked to the ANS and the sensing of internal body states (i.e., interoception) (Craig, 2002; Tracey and Mantyh, 2007; Legrain et al., 2011). Similarly, the midbrain PAG receives both peripheral nociceptive input and descending projections from the hypothalamus, amygdala, and rostral ACC. Providing output to medullary centers, the PAG is an essential component of a descending pain modulatory system that inhibits or facilitates nociceptive processing within the spinal dorsal horn (Ossipov et al., 2010; Heinricher and Fields, 2013). PAG neurons projecting to autonomic centers in the medulla are also involved in cardiovascular changes observed during opioid-dependent and independent endogenous analgesia (Benarroch, 2006; Green et al., 2006). In addition, stimulation of the PAG alters baroreflex sensitivity and cardiac control (Pelosi et al., 2007; Pereira et al., 2010; Benarroch, 2012; Lagatta et al., 2016).

Due to extensive functional overlap between structures involved in autonomic control, nociception, and pain sensation at different levels of the spinal cord, brainstem, midbrain, and cortex (Benarroch, 2001, 2006), the reciprocal ability of the ANS to modulate nociceptive information is not surprising. Amongst others, emotional appraisal (Wiech and Tracey, 2009), attention (James and Hardardottir, 2002; Bradshaw et al., 2012), mood (Carter et al., 2002), hypnotic suggestion (Rainville et al., 1999), and stress (Terkelsen et al., 2004; Ballegaard et al., 2014) can influence the subjective pain experience. Any of these states corresponds with changes in ANS activity (Wiech et al., 2008; Kreibig, 2010) and baroreflex sensitivity (Fejes et al., 2020).

Whilst previous studies have described brain networks underlying pain-induced sympathetic reactions in response to tonic pain (Kobuch et al., 2017, 2018), only one study so far has investigated how variations in HRV relate to brainstem functional connectivity changes observed during a prolonged painful experience: Sclocco et al. (2016) convolved HRV with the hemodynamic response function during 6 min of pressure pain, identifying several brainstem nuclei (specifically, rostral ventromedial medulla, ventral nucleus reticularis/nucleus ambiguous, and pontine nuclei) associated with pain-evoked HRV alterations. To date, the bi-directional association between pain-engaged higher brain networks (involving, for example, the anterior cingulate cortex) and HRV (either resting or in response to pain) remains unexplored.

In this study, we combined sampling of HRV and rs-fMRI, both during rest and tonic noxious cold stimulation, in a group of healthy participants. We adopted a hypothesis-driven region of interest (ROI) approach, drawing upon regions known to play important roles in both ANS function and nociceptive processing, namely the dACC and the PAG (Benarroch, 2006; Leone et al., 2006; Tracey and Mantyh, 2007; Legrain et al., 2011). Following previous studies, we hypothesized that HRV measures at baseline and during noxious stimulation will be associated with subjective pain intensity ratings. Further, we hypothesized that PAG and dACC RSNs would be affected by a tonic noxious stimulus, and that these pain-induced rs-fMRI changes would correlate with HRV parameters and subjective ratings of pain. Lastly, we explored whether baseline HRV is associated with the brain response to a noxious stimulus.

The aim of this study was to examine the relationship between HRV and brain functional connectivity during painful experimental stimulation. Specifically, we investigated the effect of a tonic noxious cold stimulus in a group of healthy participants, implementing simultaneous rs-fMRI and HRV sampling. Our results support a role of ANS activity, as indexed by HRV, in brain connectivity during pain and specify functional connections of our dACC and PAG seed regions that are associated with measurements of LF-HRV. In particular, we identified a three-way relationship between HRV, cortical brain networks known to underpin pain processing, and participants’ subjectively reported pain experiences. Baseline PAG-vmPFC functional connectivity was associated with higher LF-HRV during cold stimulation and lower subjective cold pain ratings, suggesting that the role of the ANS in the modulation of nociception might relate (at least in part) to PAG-cortical functional connections.

Our findings combine two separate streams of research: the association between LF-HRV and pain and the role of PAG functional connectivity in pain perception. Higher baroreflex activation and parasympathetic activity (indexed by LF and HF-HRV) prior to and during noxious stimulation is associated with reduced pain intensity or higher pain thresholds (Duschek et al., 2007; Appelhans and Luecken, 2008; Nahman-Averbuch et al., 2016; Tracy et al., 2018). In addition to the well-established role of the PAG in descending nociceptive modulation (Ossipov et al., 2010; Benarroch, 2012), PAG-mPFC connectivity (Sprenger et al., 2011) and activity of the mPFC alone (Bogdanov et al., 2015) have been linked to the efficacy of endogenous analgesic mechanisms and to vagal cardiac control (Critchley et al., 2011). A recent brainstem-focused fMRI study showed that subjective pain intensity is not only influenced by the reaction of the brain to a noxious stimulus, but also by the connectivity of the PAG prior to stimulation (Stroman et al., 2018). These results indicate that the PAG is functionally associated with the hypothalamus and several brainstem areas involved in autonomic regulation, highlighting the potential importance of homeostatic autonomic control in the descending modulation of nociception. Our findings provide direct support for this theory. Meta-analytical evidence shows that right mPFC is associated with HRV during both emotional and cognitive/motor tasks (Thayer et al., 2012). Others have reported an association between mPFC resting state connectivity and HRV, often linked to emotion regulation. Our findings expand this work by showing that the functional communication between the PAG and the mPFC is associated with LF-HRV reactions to noxious stimulation as well as the subjective perception of cold pain intensity.

During noxious cold stimulation, functional connectivity of the right dACC with the mPFC was positively correlated with LF-HRV. A reduction in the dACC-(ventral)mPFC functional connectivity has been shown to partially mediate heart rate increases during socially stressing experiences (Wager et al., 2009). Whilst we did not formally assess mediation effects, our finding extends this notion to include pain, an experience which requires both emotional and physiological regulation. In more general terms, this finding underlines the known role of the mPFC in context appraisal and autonomic adaptation, not least in situations of threat (Schiller et al., 2008; Thayer et al., 2012).

Higher baseline LF-HRV, which is often considered as a measure of baroreflex activation (Goldstein et al., 2011), was associated with an increase in PAG-precuneus functional connectivity during tonic noxious stimulation. Areas of the DMN (including the precuneus) show reduced functional connectivity during pain (Baliki et al., 2008; Kong et al., 2010a). In view of the opinion of higher LF-HRV as a likely anti-nociceptive mechanism, sustained connectivity between the PAG and the precuneus is thus consistent with a reduced nociception-associated brain response. A potential underlying mechanism is offered by Kucyi et al. (2013), who associated greater functional connectivity between PAG and precuneus/mPFC with participants’ tendencies to disengage their attention from the noxious stimulation (“mind wandering”), thus achieving pain relief by means of distraction (Sprenger et al., 2012). Possibly, effective engagement of descending pain modulation as associated with LF-HRV allows for, or is part of, participants’ natural tendencies to “take their minds off the pain.” Accordingly, this finding offers preliminary support for theories of a protective role of mind wandering in optimizing states of bodily arousal during challenges to homeostasis (Baars, 2010; Ottaviani et al., 2013).

Higher baseline LF-HRV was also associated with a stronger decrease in functional connectivity between both the PAG and dACC seeds with the dlPFC. DlPFC activity is commonly reported in pain experiments. Functionally part of the central executive-control network (CEN), the dlPFC shows increased activation during the performance of cognitively demanding tasks (Seeley et al., 2007) and is associated with cognitive difficulties seen in chronic pain (Bushnell et al., 2013). Albeit based on experimentally induced pain, our results suggest a link between cognitive and homeostatic systems, and indicate a possible common neuronal substrate for cognitive difficulties (Apkarian et al., 2004; Bushnell et al., 2013) and HRV alterations (Meeus et al., 2013) often reported in chronic patients. Functional connectivity of the dACC with the dlPFC has previously been shown to co-vary with LF-HRV (Chang et al., 2013), supporting a role of the ANS in this functional link.

Higher baseline resting LF-HRV also related to decreases in functional connectivity between the dACC and the left AI during noxious stimulation. Activity of the insular cortex is frequently coupled to that of the cingulate cortex. Together, they form part of an interoceptive network which facilitates emotion and self-awareness but also modulates autonomic function (Craig, 2002). A specific role in cardiovascular control has been postulated (Nagai et al., 2010). Importantly, insular activity is associated with top down suppression of baroreflex activity induced by a stress challenge (Gianaros et al., 2012). We suggest that the ANS-associated connectivity between AI and dACC plays a central role in interoception during pain and might also contribute to baroreflex-induced anti-nociception (as reviewed by Suarez-Roca et al., 2018). Also, activity in the AI is commonly seen during pain (Legrain et al., 2011). More specifically, the AI is involved in the cognitive evaluation of pain (Brooks et al., 2002; Wiech et al., 2010; Segerdahl et al., 2015). Furthermore, combined AI and ACC activation was found to relate to the emotional components of pain, such as psychological pain, and empathy for a loved one in pain (Singer et al., 2004; Immordino-Yang et al., 2009). Less functional connectivity between the dACC and AI with high baseline LF-HRV is thus consistent with maintained baroreflex activity during noxious stimulation, and reduced activity in an interoceptive and cognitive/emotional key area of pain perception.

Our data provide reason for further investigation into the role of autonomic cardiovascular modulation in central network dynamics. The mPFC and the precuneus are considered part of the DMN (Fox and Raichle, 2007), whilst the dlPFC is a core node in the CEN (Seeley et al., 2007). The dACC and the AI, on the other hand, form the salience network (SN) which responds to the subjective salience of cognitive, emotional, and homeostatic stimuli. The SN is thought to play a critical role in switching between the off-task DMN and the on-task CEN (Sridharan et al., 2008; Menon and Uddin, 2010). Both DMN and CEN alterations have been described in chronic pain states (Baliki et al., 2008; Cauda et al., 2009; Malinen et al., 2010; Napadow et al., 2010, 2012; Tagliazucchi et al., 2010; Bolwerk et al., 2013; Jiang et al., 2016; Alshelh et al., 2018; Androulakis et al., 2018). Here we demonstrated that by phenotyping participants based on LF-HRV, a physiological variable, we were able to predict changes in components of these functional networks during the experience of a noxious stimulus. Albeit not having assessed network dynamics explicitly, these findings suggest that the ANS may be implicated in the salience network-mediated switch from default to central cognitive-executive modes during the experience of tonic experimental pain. Speculatively, a pronounced LF-HRV at baseline (interpreted as a more pronounced engagement of the baroreflex) might predict a reduced tendency for a SN-mediated switch of network dynamics toward an executive mode, whilst LF-HRV during the experience of pain might be associated with maintained DMN dynamics. It remains to be determined whether these initial neurophysiological findings apply in patients with chronic pain.

Finally, hierarchical interactions between cognitive, emotional and autonomic processes are also a quintessential component of the neurovisceral integration (NVI) model, as proposed by Thayer and Lane (2000). This model postulates that cardiac vagal tone (i.e., the contribution of the parasympathetic nervous system to cardiac regulation) is an indicator of the functional balance between neural networks involved in the regulation of emotions and cognition (Thayer and Lane, 2000). This Central Autonomic Network (CAN) (Benarroch, 1993) consists of parts of the prefrontal cortex (anterior cingulate, insula, orbitofrontal, and ventromedial cortex), limbic cortex (amygdala and hypothalamus), and brain stem areas (i.e., PAG and ventromedial medulla). Our data fits with this model, as they suggest that the resting cardiac autonomic tone (possibly related to the baroreflex) and the cardiac autonomic reactivity to pain is associated with the functional organization of some pain-related networks (i.e., those related to the dACC and PAG, involving other structures of the CAN such as vmPFC and anterior insula), which in turn is related to the amount of experienced pain. It is of note, however, that our data do not strictly explore the vagal influence on cardiac regulation, but rather the reactivity of the baroreflex, which is more representative of sympatho/vagal balance. Importantly, the NVI model suggests that HRV is a measure of the flexibility of the entire brain-body system, with the view that flexible systems are adaptive and responsive to the environment, allowing for functional oscillations between different states (Smith et al., 2017). In the context of pain, we argue that an individual with an adequate ANS reaction to pain is more efficient in triggering those mechanisms which have the aim of re-establishing adaptive homeostasis (i.e., descending pain modulatory mechanisms). Future studies should aim to explore this model in clinical populations.

We acknowledge some methodological limitations: First, the method of HRV acquisition, photoplethysmography (PPG), is not considered gold standard. Despite this, individual studies such as Gil et al. (2010) show that while subjects are at rest, PPG is a good surrogate measure for ECG-derived HRV. Sufficient accuracy under stationary conditions has also been confirmed by a review on the topic (Schäfer and Vagedes, 2013). Furthermore, the inter-session reliability of our measurements was good (see Supplementary Material S3), providing confidence in our method of capturing HRV. We also acknowledge a relatively small sample size; this study was formally powered to detect the effect of a noxious stimulation on functional brain networks rather than to determine interactions between functional networks and the ANS. Future studies should aim to replicate our findings with larger samples in order to validate the robustness and reliability of our results. Lastly, and this note of caution applies to many studies linking brain activity to function, the associations between HRV, functional connectivity and subjective pain does by no means imply causality. Further studies will have to elucidate if there is a causal relationship between the above findings. Dynamic causal modeling techniques applied to brain connectivity data (Friston, 2009) offer a potential means to conduct these investigations.

In conclusion, we have demonstrated that the engagement of brain regions involved in the cognitive, emotional and limbic processing of pain is intimately linked to autonomic profiles and subjective pain sensitivity. As the first study to explore the association between pain-related HRV and brain functional connectivity, we provide an improved understanding of the relationship between pain perception and autonomic cardiovascular control, likely involving the baroreflex.

Future research should determine whether this functional connectivity is altered in chronic pain, and if modulation of ANS activity might protect from chronic pain. Such data may catalyze the development and utility of ANS-targeted pain treatments, such as HRV biofeedback, vagal stimulation or baroreceptor activation therapies.

At the date of publication, the datasets for this manuscript are not publicly available because of limited ethical approval to share participant data. Requests to access the datasets should be directed to EM (elena.makovac@kcl.ac.uk).

The studies involving human participants were reviewed and approved by the Ethics Committee King’s College London. The patients/participants provided their written informed consent to participate in this study.

DH-S analyzed the data and wrote the manuscript. GC, OD, JJ, SM, OO’D, JO’M, AdLR, SW, and SM contributed substantially to the conception of the experiment and to the analysis of the data and reviewed the manuscript. EM conceived and conducted the experiment and contributed substantially to the writing and revision of the manuscript. MH supervised the team, contributed to the planning and conduction of the experiment, and made major contributions to the revision of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.