Introduction

Musculoskeletal pain is a major health problem, and research into the underlying neurophysiological mechanisms is required to improve its understanding and management (Arendt-Nielsen et al. 2011). It is a clinical condition that is characterized by ongoing pain and discomfort in the deep tissues such as muscle, bones, ligaments, nerves and tendons (Kehl and Fairbank 2003; Arendt-Nielsen et al. 2011). Clinical musculoskeletal pain is usually not a short-lasting and time-bound sensation; even acute pain can last from minutes to days or weeks. In this regard, tonic rather than phasic experimental pain models better resemble clinical musculoskeletal pain (Svensson and Arendt-Nielsen 1995). Although many surrogate models can closely mimic musculoskeletal pain (e.g. injections of saline solution or nerve growth factor), they are invasive, and the time course and intensity of pain are hard to control. In this regard, recent developments in cuff pressure algometry allow for a non-invasive, more controllable setup (Polianskis et al. 2001).

People who suffer from musculoskeletal pain describe it as a sustained, intense and prolonged sensation (Soares and Jablonska 2004), all of which emphasize the disruptive ability of a painful stimulus to capture attention even without behavioural relevance and voluntary effort (Eccleston and Crombez 1999). Thus, pain has the intrinsic attribute of maintaining its salience over the duration of the stimulus, and it has been shown that the anterior insula (AI) and the anterior cingulate cortex (ACC) display sustained responses during painful transcutaneous electrical stimulation (Downar et al. 2003). These two brain regions are often activated together and are involved in detecting and filtering salient stimuli that are behaviorally relevant (Legrain et al. 2011). Since sustained deep-tissue pain can remain salient for a prolonged period, describing the activity of these regions may provide new insights into musculoskeletal pain mechanisms.

Previous studies that employed brain neuroimaging techniques such functional magnetic resonance (fMRI) and positron emission tomography (PET), have found that musculoskeletal pain is characterized by brain activation in numerous areas (Duerden and Albanese 2013). Among these regions, the primary somatosensory cortex (S1), the anterior insula (AI) and ACC are consistently reported in the literature as being involved in musculoskeletal pain (Kupers et al. 2004; Thunberg et al. 2005; Kim et al. 2013). Still, fMRI or PET are not suitable to assess neural oscillations with frequencies higher than 4 Hz (Ritter and Villringer 2006). On the contrary, techniques with high temporal resolution such as electroencephalography (EEG) or magnetoencephalography (MEG) provide a better measurement of neural oscillations, which are known to be in the range of 0.01 Hz to 100 Hz, and even higher (Cohen 2017). This characteristic makes EEG an invaluable tool to assess the neural dynamics during tonic pain (Ploner et al. 2017).

An important limitation of scalp EEG is that it does not provide direct information about the spatial localization of the underlying sources within the brain, because the electrodes record the superpositions of brain signals originating from the entire cortical grey matter. In this regard, there is a considerable number of inverse solution algorithms that allow estimating the active neuronal populations inside the brain, yet with limited precision (Michel et al. 2004; Michel and Murray 2012). Previous studies have characterized the neural correlates of superficial tonic painful stimulation using these techniques (Chang et al. 2002; Hansen et al. 2017). However, EEG studies that characterize brain response at the source level during sustained deep-tissue pain are scarce.

The main goal of this study was to characterize the current source density (CSD) estimations of brain source oscillations in healthy subjects during tonic, deep-tissue painful and non-painful mechanical pressure stimulation (Le Pera et al. 2000; Lavigne et al. 2004). It was hypothesized that sustained deep-tissue pain would result in increased power of neural oscillations compared to non-painful stimulation, reflecting higher activity in regions involved in pain processing.

Materials and Methods

Subjects

Twenty-three healthy human subjects (nine males, fourteen females, mean ± SD age: 24.5 ± 2.5 years) participated in the study. The participants were recruited by advertisement at the local university. None of the participants suffered from chronic pain, had any pain or used pain medication during one week before participation. Written informed consent was obtained from every participant. This study was approved by the ethics committee of Northern Jutland, Denmark (N-20170047) and conducted in accordance with the Declaration of Helsinki.

Stimulation

Sustained Deep-Tissue Stimulation

A computer-controlled cuff-pressure algometer (CPA) (NociTech, Denmark, and Aalborg University, Denmark) (Polianskis et al. 2002a) was used to deliver pressure stimulation to the right forearm. The CPA consists of a 10-cm wide silicone tourniquet cuff (VBM, Germany), a compressor (Condor MDR2, JUN-AIR International A/S, Nørresundby, Denmark) connected to an electric–pneumatic converter (ITV2030, SMC Corp., Tokyo, Japan) and controlled by a computer through a data acquisition card (PCI 6024E, National Instruments, Austin, TX, USA). The cuff was placed around the right forearm making sure that it was attached over the belly of the extensor carpi radialis brevis (ECRB) muscle 3-cm proximal to the cubital fossa.

Vibrotactile Stimulation

Continuous vibrotactile stimuli (VT) was delivered to the right forearm using a vibrotactile stimulator (g.STIMbox, g.tec medical engineering GmbH, Austria). The vibrotactile was placed under the cuff and over the ECRB muscle, to ensure that the same skin area was stimulated with both modalities. Vibrotactile stimuli consisted of constant-amplitude sinusoidal mechanical vibration delivered at 30 Hz using a vibrotactile transducer (g.VIBROstim, length: 3 cm; width: 1 cm, g.tec medical engineering GmbH, Austria).

Behavioural Data

The participants were instructed to relax, pay attention only to the stimulation and use their left hand to manipulate a slider in a 10-cm visual analogue scale (VAS) device to score the perceived intensity of each stimulus. Besides, the participant could press a button to stop the stimulation. The intensity of the perceived sensation was rated on a scale anchored from 0 (no sensation at all) to 10 (maximal pain), where 5 represented the pain detection threshold (Hansen et al. 2017). Therefore, VAS scores from 0 to 5 were considered as non-painful, and VAS above 5 up to 10 were regarded as painful. The participants were instructed to start rating the stimulation intensity continuously on the VAS device (sampled at 10 Hz) once the stimulation became noticeable and to press the button to stop the stimulation if the pain became intolerable.

Electrophysiological Data

EEG data were collected using a g.HIamp amplifier (g.tec medical engineering GmbH, Austria) from 64 scalp electrodes placed according to the international 10/20 system. Electrode impedance was kept below 10 kΩ and the sampling rate was 1200 Hz. Electrodes were referenced to the left earlobe (A1) and AFz served as the ground. At the beginning of the experiment, three minutes of eyes-open resting EEG (REEG) recordings were conducted for quality control. To prevent eye movements during resting, the participants were instructed to fix their sight to a fixation cross displayed on the computer screen placed in front of them. Furthermore, to minimize eyes movement during stimulation, participants were instructed to fix their sight to a vertical VAS bar displayed on the computer screen that showed their actual VAS score.

Experimental Setup

Data acquisition was carried out in a single session. Participants were seated on a comfortable chair, with both forearms resting on the armrest and facing a computer screen. Participants wore foam earplugs (Earplugs, TaperFit 3 M, Minnesota, United States) to reduce ambient noise. Prior to the acquisition of data, familiarization trials were conducted to introduce the participants to the cuff pressure sensation and to train them to score with the VAS device. Familiarization trials consisted of the administration of constant stimulation for 2 min at a random pressure that ranged from 10 to 100 kPa. After familiarization, the average of the cuff pressure that elicited VAS 7 and VAS 3 in three consecutive ramps (rate = 1 kPa/s, interstimulus interval = 5 min, pressure limit = 110 kPa) were used to estimate the pressure of the sustained deep-tissue pain (SDTP) and sustained deep-tissue no-pain (SDTnP) conditions, respectively.

The session continued with three stimulation blocks of three minutes each: SDTP, SDTnP, and VT. Both VAS and EEG were recorded simultaneously during stimulation. There was a pause of five minutes between every block, during which no stimulus was presented. The order of the stimulation blocks was randomized across participants. In order to reach the pressure used in the SDTP and SDTnP conditions, the cuff was inflated at a rate of 1 kPa/s until the pressure level was reached, and it was kept constant for three minutes.

EEG Pre-processing

EEG data were exported to Matlab R2016b (The Mathworks Inc, Natick, MA, USA) for off-line processing using EEGLAB toolbox v14.1.1 (Delorme and Makeig 2004). EEG data were digitally band-pass-filtered between 0.1 and 30 Hz and resampled to 500 Hz. The recorded signals were visually inspected to exclude noisy segments (non-cerebral source activity). In the event that a bad/noisy channel was found in a certain participant, such channel was rejected for all conditions for that said participant. A maximum of three channels were removed from the EEG dataset. Moreover, independent component analysis was carried out to remove remaining noise from blinks, muscle activity, and electrode artefacts (Jung et al. 2000). The continuous clean datasets were segmented in 4-s epochs. EEG data length was further reduced to a total of 160 s for each condition to account for differences in the number of segments eliminated during cleaning, ensuring the same amount of data for all the subjects. Furthermore, the first two epochs were rejected to avoid transient evoked responses (Chen et al. 2008). In total, 38 epochs were used in the subsequent analyses per condition and per participant.

Before further data analysis, across-subject grand average scalp topographic of spectral power were plotted for each condition and each frequency band for quality control, since further signal processing steps are highly dependent on artefact-free data (Hoffman 2006). Since eye movements generate increased power in the delta band of the frontal electrodes (Babiloni et al. 2020), this entire frequency range would be disregarded from the analysis in the event that it exhibits clear remnants of artefacts in the delta band topography.

Estimation of Cortical CSD Distribution

For the estimation of the cortical three-dimensional current source density distribution, the exact low-resolution brain electromagnetic tomography (eLORETA) method was used (Pascual-Marqui 2007). The eLORETA (LORETA-KEY software v20150415, http://www.uzh.ch/keyinst/loreta.htm) allows estimating the cortical current density from the EEG scalp recordings. This is performed by solving the so-called EEG inverse problem (Michel et al. 2004). The solution space is limited to the cortical grey matter and has a volume of 6239 voxels at 5 mm3 spatial resolution. In this solution, the lead field was computed using a realistic head model based on the Montreal Neurologic Institute average MRI (MNI152) (Lancaster et al. 2000; Fuchs et al. 2002; Brett et al. 2002). The current density was calculated from the cleaned EEG epochs for each condition and for each subject. The eLORETA CSD was estimated for five frequency bands: delta (1–3.9 Hz), theta (4–7.9 Hz), alpha (8–13.9 Hz) and beta (14–29.9 Hz).

The CSD was extracted from seven specific regions of interest (ROIs), which are frequently reported to be active during pain processing (Peyron et al. 2000; Downar et al. 2003; Apkarian et al. 2005; Duerden and Albanese 2013; Jensen et al. 2016), see Table 1. Note that, in the present study, the insulae were the only anatomical regions that were analysed bilaterally. The other selected regions (Table 1) were those located contralateral to the stimulation site. In previous literature, however, some of these areas are reported as being active bilaterally during pain, namely the posterior parietal cortex (PPC), pre-motor cortex (Pre-MC) and dorsolateral-prefrontal cortex (DLPFC). Despite these ipsilateral areas are likely to be active during pain, higher brain activity can be expected in the contralateral counterparts (Shenoy et al. 2011; Yücel et al. 2015). Since the experimental procedure involved the use of the left hand to score their sensation using a VAS, and this task certainly elicits brain activity in the ipsilateral side that could be confounded (Fuchs et al. 2000), it was decided to focus primarily on the contralateral regions, marked with the “L-” (left) prefix throughout the text. The ROIs were defined as the average of all voxels that fell within a 15-mm radius sphere around the seed point (Canuet et al. 2012). Additionally, the absolute CSD of the ROIs during REEG, SDTP, SDTnP and VT was estimated.

Table 1 Spatial locations of regions of interest considered for the analysis
Full size table

Statistical Analysis

Perceived Stimulus Intensity

To assess the effects of the stimulation conditions (SDTP vs. SDTnP vs. VT) on the perceived stimulus intensity, a point-by-point, one-way repeated measures analyses of variance (RM ANOVA) with permutation test (Maris and Oostenveld 2007) was performed on the continuous VAS scores over the 180-s interval using Letswave software (https:// github.com/NOCIONS/letswave6/). The permutation test was performed to control for false positives (type I error) that may result from the multiple comparisons performed for each time point.

CSD Estimations

To assess possible differences in CSD estimations between conditions, a within-subject generalized linear mixed model was built in SPSS (version 24; IBM Corporation, New York, NY). The CSD was defined as the target, while the experimental conditions (three levels: SDTP, SDTnP and VT), the ROIs (seven levels: L-ACC, L-S1, R-AI, L-DLPFC, L-AI, L-PPC, L-S2), the EEG frequencies (three levels: theta, alpha and beta), and all possible combinations between them were included as fixed factors. Furthermore, subject was included as a random effect and the covariance type structure for the residual was specified as compound symmetry. Since the CSD is non-normally distributed (Tzyy-Ping Jung et al. 1997; Cohen 2014), a gamma distribution with a log link was used to fit the model (Bolker et al. 2009). The absolute criterion for parameter convergence of the model was set at \({1\cdot 10}^{-5}\). Post-hoc Sidak corrections were applied to adjust for Type I error.

Results

Perceived Stimulus Intensity

There was a marked difference in the perceived intensity scores between the SDTP, SDTnP, and VT conditions differently across time (p < 0.001) in accordance with the intended stimulation levels, see Fig. 1 Neither SDTnP nor VT mean intensity scores surpassed the pain threshold, and the SDTP did not fall below the pain threshold either. Furthermore, the three curves reached a plateau, and habituation was not observed during most of the stimulation period. The mean cuff pressure that elicited SDTnP was 29.69 ± 9.65 kPa and SDTP was 62.02 ± 14.62 kPa.

Fig. 1
figure 1

Mean (dashed) and standard deviation (shaded) visual analogue scale (VAS) scores for all subjects (N = 23) during SDTP (red), SDTnP (blue) and continuous vibrotactile (VT, black) stimulation. Point-by-point, one-way repeated measures analyses of variance with permutation test revealed significant differences across the whole-time interval (p < 0.001) in the VAS scores between conditions. The orange dotted line represents the pain threshold anchored at VAS = 5. (red) SDTP sustained seep tissue pain, (blue) SDTnP sustained deep-tissue no-pain, (black) VT continuous vibrotactile (Color figure online)

Full size image

Topographic Representation

The grand-average scalp topographies for REEG, SDTP, SDTnP and VT for each frequency band are represented in Fig. 2. Higher activity over parieto-occipital electrodes is displayed during REEG in the alpha band, which appears to decrease during stimulation. Moreover, there was higher activity in the frontal electrodes than in the rest of the scalp for the delta band. The estimation of spectral components from frontal electrodes are heavily affected by eye movements (Hagemann and Naumann 2001), and the present scalp maps clearly showed that not all ocular artefacts were separated and rejected from the brain signals. Therefore, the delta band was not included in the statistical analysis.

Fig. 2
figure 2

Grand-average topographic distribution of absolute EEG power across frequency bands and experimental conditions. REEG resting EEG, SDTP sustained seep tissue pain, SDTnP sustained deep-tissue no-pain, VT continuous vibrotactile

Full size image

CSD Estimation

Figure 3 exhibits the current source density of three particular areas (L-ACC, R-AI and L-S1) across the 1–30 Hz frequency range for REEG, SDTP, SDTnP and VT conditions. Overall, the CSD of REEG showed higher values in the delta frequency range in the three regions. Furthermore, a clear peak in the alpha band was observed in S1 during REEG.

Fig. 3
figure 3

source density (CSD) under resting EEG (REEG), sustained deep-tissue pain (SDTP), sustained deep-tissue no-pain (SDTnP) and vibrotactile (VT) of the left anterior cingulate cortex (L-ACC), right anterior insula (R-AI) and left primary somatosensory cortex (L-S1)

Across subject mean (± SD) of the Log current

Full size image

The statistical analysis showed a main effect of the condition (F3,1.386 = 99.92, p < 0.001), ROIs (F6,1.386 = 105.19, p < 0.001) and frequency (F2, 1.386 = 553.89, p < 0.001) on the CSD. The two-way interactions condition × ROIs (F12,1.386 = 2.11, p = 0.014) and condition × frequency (F6,1.386 = 6.86, p < 0.001) showed significant effect. No effect was found in the tree-way interaction ROIs × condition × frequency (F24,1.386 = 0.67, p = 0.885). Adjusted pairwise comparisons of relevant interactions are shown in Tables 2, 3, Fig. 4 and 5.

Table 2 Pair-wise comparisons of the ROI × condition double interactions
Full size table
Table 3 Pair-wise comparisons of the Frequency × condition double interactions
Full size table
Fig. 4
figure 4

source density (CSD) estimations of the interaction between conditions and the regions of interest. The grey bar represents the interquartile range and the white dot the median. The whiskers represent 1.5 times the interquartile range. The outer shape represents the distribution density of the log10(CSD). Each point in the violin represents an individual value. The black line represents the mean of the log10(CSD). *P < 0.05; **P < 0.01; ***P < 0.001. SDTP sustained deep-tissue pain, SDTnP sustained deep-tissue no-pain, VT continuous vibrotactile, L-ACC left anterior cingulate cortex, L-S1 left primary somatosensory cortex, R-AI right anterior insula, L-AI left anterior insula, L-DLPFC left dorsolateral prefrontal cortex, L-PPC left posterior parietal cortex, L-PreMC left premotor cortex

Violin-plots representing the current

Full size image
Fig. 5
figure 5

source density (CSD) estimations of the interaction between conditions and frequency bands. The grey bar represents the interquartile range and the white dot the median. The whiskers represent 1.5 times the interquartile range. The outer shape represents the distribution density of the log10(CSD). Each point in the violin represents an individual value. The black line represents the mean of the log10(CSD). *P < 0.05; **P < 0.01; ***P < 0.001. SDTP sustained deep-tissue pain, SDTnP sustained deep-tissue no-pain, VT continuous vibrotactile

Violin-plots representing current

Full size image

Analysis of the condition × ROIs interactions showed that the participants presented higher brain activity in the L-ACC, both anterior insulas, the L-DLPFC, L-PPC and L-PreMC when they reported pain (SDTP) compared to when they did not experience pain (SDTnP and VT). Furthermore, the participants had higher activity in the L-S1 for increasing stimulus intensity.

Analysis of the condition × frequency interactions showed that the participants had higher alpha and beta activity when they reported pain compared to when did not. Participants had higher theta activity when increasing the stimulus intensity.

Discussion

The purpose of this study was to characterize the effects of sustained deep-tissue painful stimulation on the brain oscillations in regions involved in pain processing. The results show that evoked activity by painful cuff-pressure algometry can be differentiated from non-painful cuff-pressure algometry. More precisely, increased activity within the L-ACC, AI, L-DLPFC, L-PPC, L-PreMC were consistent with increasingly painful input. In contrast, the activity in L-S1 may be related to somatosensory discrimination of stimulus intensity with higher CSD evoked by pressure algometry in comparison with vibrotactile stimulations.

Stimulus Intensity and Evoked Sensations

In this study, the participants reported steady sensations when the stimulus intensities were held constant. Similarly, previous studies have reported that continuous cuff-pressure algometry and vibrotactile stimulations elicit stable sensations that do not easily adapt or build-up (Polianskis et al. 2002a). This contrasts with other experimental tonic pain models. For instance, intramuscular infusion of hypertonic solution induces temporal summation of pain after its injection, which can last several minutes (Stohler and Kowalski 1999; Arendt-Nielsen and Svensson 2001). Moreover, tonic heat stimulation initially induces temporal summation, in which a non-painful stimulus can become painful (Huber et al. 2006); however, it is followed by pain habituation that may even lead back to non-painful perception (Hashmi and Davis 2009). These adaptive processes certainly modulate the perception of pain and involve different levels of salience (Davis 2011). Thus, conclusions about tonic pain processing and salience can be affected when either habituation or temporal summation are present (Jepma et al. 2014; Weissman-Fogel et al. 2015). Alternatively, using a stimulus that evokes a constant perception of pain intensity may help to disentangle the mechanisms underlying tonic pain and salience. Consequently, computer-controlled cuff pressure algometry can provide deep stimulation and evoke continuous aching sensation, which consequently decreases the variability in pain ratings associated with the mechanisms previously mentioned (Polianskis et al. 2001).

Effect of Sustained Simulation in the Overall CSD

The present results showed that vibrotactile, sustained deep-tissue painful and non-painful stimulation elicited different activity levels in the studied regions. Particularly, these regions exhibited an overall higher CSD when the participants reported pain in comparison to when they did not feel pain. This is in agreement with several observations that found that self-reported pain intensity strongly correlates with the magnitude of brain oscillations (Tracey and Mantyh 2007; Nir et al. 2012; Zhang et al. 2012; Schulz et al. 2015). These findings are similar to those reported by studies that investigated phasic evoked responses to painful stimulation (Ohara et al. 2004; Iannetti et al. 2005; Zhang et al. 2012), and further supported by neuroimaging evidence (Schneider et al. 2001; Davis et al. 2002; Baliki et al. 2008; Lin et al. 2018).

Acute muscle pain serves to protect the body and is subject to an appropriate motor response to avoid further damage and contribute to recovery (Graven-Nielsen 2006). In this experiment, participants continuously slid the VAS scale to quantify the perceived stimulus intensity, which activates sensory and motor cortical areas. Thus, the difference found in the CSD estimations between SDTP and SDTnP may be related to the preparation and the execution of motor responses. Yet, previous evidence suggests that painful and non-painful stimuli do not differ in the influence on motor preparation in the human brain (Postorino et al. 2017). Furthermore, the participants repeated this scoring activity in all three experimental conditions. Therefore, the observed differences in the CSD estimations between conditions cannot be attributed to the scoring task.

During cuff-pressure algometry not only deep-tissue is stimulated, but also the skin. Therefore, differences in CSD estimations between painful and non-painful cuff stimulation may be attributed to factors other than deep pain. However, Polianskis et al. (2002b) induced hyperalgesia in the skin using topical capsaicin cream, yet the participants did not report increased pain sensitivity to cuff stimulation in the treated area. In another study (Manafi-Khanian et al. 2015), the researchers investigated the mechanical stress and strain distribution in superficial and deep tissues during cuff algometry using a computational three-dimensional finite element model of the lower leg; showing that cuff algometry preferentially stimulates deep somatic tissue. On the other hand, vibrotactile stimulation mostly stimulates mechanoreceptors in the skin (Verrillo 1985). Thus, differences in CSD found in this study between cuff pressure and vibrotactile are likely due to different tissues and receptors being stimulated.

The L-ACC and AI are central hubs of a network involved in salience detection called the salience network. Observing brain activity within these regions when a person experiences pain is not sufficient to assume that such activity is a direct correlate of pain perception (Iannetti et al. 2013). A body of work argues that the measured brain activity not only reflects the processing of the perceived stimulus intensity, but it also reflects its salience (Legrain et al. 2003; Iannetti et al. 2008; Wiech et al. 2010). A salient stimulus thus evokes activity in the brain that directs the attention towards any perceived event that may have a potential impact on the organism (Costantini et al. 2008). In other words, a painful experience will generally stand out from other stimuli, since maintaining tissue integrity is crucial. Therefore, the observed increase in brain activity in the most painful condition might not strictly be related to the change in stimulus intensity, but to an increase in the salience of the stimulus (Iannetti et al. 2008; Legrain et al. 2009).

In the present study, each stimulus block lasted three minutes, which highly contrasts to the duration of high-intensity laser or electrical pulses (in the tens of milliseconds) used in the majority of previous studies that investigated the transient brain responses elicited by these stimuli and its functional significance in pain neurophysiology (Legrain et al. 2011). It can be argued that the long-duration, constant stimulation loses its novelty and its ability to capture attention. Still, novelty might not be the only factor that determines the salience of a stimulus, but rather its ability to stand out (Ronga et al. 2013). Indeed, non-painful somatosensory stimulation requires voluntary attention or behavioural relevance to maintain salience; for instance, voluntary attention to score the perceived sensation continuously on a scale (Frith 2001). On the other hand, constant painful stimulation tends to maintain salience even without voluntary attention, because pain is able to keep attention both involuntarily and continuously (Downar et al. 2003). Therefore, the observed changes in the CSD estimations during SDTP might indeed reflect the processing of the salience of cuff-pressure painful stimulation that makes it stand from the environmental context.

Effect of Sustained Simulation in the ROIs

The present results showed differences in CSD at L-S1 among conditions. It is well documented that S1 is mostly involved in the sensory-discriminative aspects of somatosensory processing, such as stimulus intensity, localization and frequency, (Kenshalo et al. 1988; Jousmäki and Forss 1998; Bushnell et al. 1999). It is worth noting that vibration and pressure are mediated by different receptors. When using a cuff-pressure algometer, the pressure sensation originates from activation of free nerve endings in the deep tissues, conducted via group III (Aδ) and IV (c) afferent fibres that project to S1 via the thalamus (Graven-Nielsen et al. 2004). On the other hand, the vibrotactile sensation originates from activation of Meissner and Pacinian corpuscles in the skin, conducted via group II (Aβ) afferent fibres which also project to S1 (Breitwieser et al. 2012). Furthermore, neuroimaging studies confirm that S1 is not only involved in discriminating the stimulus intensity but also its localization and quality discrimination (Peyron et al. 1999; Hofbauer et al. 2001; Lemus et al. 2010). Therefore, the observed differences in oscillatory activity within S1 between all conditions possibly reflect the differences in somatosensory input given by the different tissue and fibres stimulated by the cuff and vibrotactile stimulator.

CSD in the L-ACC and both AI was higher when the participants reported pain compared to when they did not (SDTnP and VT). Previous researchers have stated that the AI and ACC contribute to a large extent in detecting and attending salient events in the sensory environment (Downar et al. 2002; Wiech et al. 2010). Furthermore, these regions have also been reported to be involved in emotional, sensorimotor, homeostatic and cognitive functions (Bush et al. 2000; Taylor et al. 2009; Dolcos et al. 2011; Gasquoine 2014). The L-DLPFC, L-PPC and L-PreMC exhibited similar responses to the AI and L-ACC. Indeed, previous work indicates that the activation of the PreMC may be associated to the urge of hand withdrawal from the source of pain while the participants were instructed to remain still (Svensson et al. 1997). The activation of PPC is associated with the body awareness of the painful area and sensorimotor coordination (Forss et al. 2005) when compared to non-painful stimulation. Even though the activation of DLPFC has been associated with pain detection (Seminowicz and Davis 2007; Seminowicz and Moayedi 2017), the function of this area during pain is still debated. Except for S1, all the studied areas exhibit a similar increased neuronal activity during sustained deep tissue pain. This supports the notion that none of these regions is exclusive to pain processing but contributes to the different dimensions of pain (Peyron et al. 2000) and their activity is increased under pain.

Last but not least, these results are in accordance with the concept suggested by Davis et al. (2015) where some regions/networks of the brain should work as a “pain switch” that changes its state during pain. In other words, brain oscillations within these regions, seem to be in low activity state (“off”) when subjects perceived non-painful stimuli, while it seems to change to high activity state (“on”) when subjects perceived pain. However, with the present methods, pain cannot be disentangled from saliency, and therefore further investigations are needed to understand the “pain switch”. According to the results of this study, the “pain switch” may not be exclusive of a single region, but a shared trait of several interconnected regions.

EEG Spectral Changes During Sustained Stimulation

Consistent with other studies, increased theta activity was found during tonic pain (Navid et al. 2019). The literature suggests that theta oscillations are related to the level of attention (Fallon et al. 2017; Keller et al. 2017). In the present study, the participants were instructed to keep their attention to the stimulus source and score their perceived sensation across all conditions. Thus, the differences found in CSD in the theta band cannot be only attributed to differences in attention. The present differences were not only observed between painful and non-painful conditions, but also between non-painful stimulations (SDTnP and VT). Even though a few studies have proposed that theta oscillations have an association to pain intensity (Ray et al. 2009; Huishi Zhang et al. 2016), the present findings suggest that theta oscillations in the studied regions may not be exclusively related to pain perception, but rather associated with the amount of somatosensory input.

Increased CSD in the beta band was detected when the participants reported pain. Previous work has observed a desynchronization of the beta band, relative to pre-stimulus activity, immediately after a brief stimulus (Stancák et al. 2003). Another study found that after the desynchronization, there is a beta rebound (Hauck et al. 2007). It is noteworthy, that whereas those studies used repeated brief painful stimulus to test the beta synchronization and desynchronization, the present study applied a constant stimulus, and the first eight seconds of the EEG were excluded. Therefore, it is unlikely that this beta desynchronization may be reflected in the results. Moreover, beta oscillations are commonly associated with motor functions and increased beta activity is normally observed when a movement has to be resisted or voluntarily suppressed (Androulidakis et al. 2007; Kilavik et al. 2013). The typical defensive response to a painful stimulus is to withdraw the affected limb. In this experimental setting, however, the participants were asked to maintain the forearm in the same position across the experiment. It is therefore likely that the increase in the beta band reflects the voluntary action of suppressing the withdrawal movement (Misra et al. 2017).

The present results showed an increase in the alpha band during SDTP compared to VT and SDTnP. Le Pera and colleagues (2000) also found increased CSD in the alpha band during experimental muscle pain when compared with vibrotactile stimulation. Increased alpha activity has been observed during hand immersion in painful cold water (cold pressor test), compared to immersion in non-painful cool water (Backonja et al. 1991). In the latter, enhanced alpha-band activity was found over central electrodes contralateral to the stimulation site, at the beginning of the painful condition, and followed by a reduction of alpha activity after the first minute. A bulk of work otherwise reported that painful stimulation reduces alpha activity; importantly, these studies compared the painful condition to resting state (Nir et al. 2012; Peng et al. 2014; Hansen et al. 2017; Navid et al. 2019). Current evidence strongly suggests that the alpha band has a functional inhibitory function, i.e., the alpha band activity gates information by inhibiting task-irrelevant regions of the brain (Klimesch et al. 2007; Jensen and Mazaheri 2010). This implies that increased CSD in the alpha band found in the present study may be the consequence of increased cortical inhibition induced by painful cuff-pressure stimulation.

The observed increase in alpha activity during tonic deep-tissue pain may therefore be reflecting a lack of inhibition possibly mediated by the neurotransmitter gamma-aminobutyric acid (GABA) (Klimesch et al. 2007; Başar and Güntekin 2008). Increased GABA levels have been previously measured in the ACC and R-AI with MR spectroscopy during tonic painful stimulation (Jasmin et al. 2005; LaGraize and Fuchs 2007; Kupers et al. 2009). The neurotransmitter GABA is involved in the cortical modulation of pain trough inhibition of synapses in the brain (Olsen and Sieghart 2009). Still, the functional role of the alpha band oscillations in the brain is not fully elucidated (Palva and Palva 2011), and further studies are needed to understand its function in tonic pain processing. The observed differences between power amplitude in the frequency bands are not relevant to discuss; it is well-known that the power spectrum of EEG decays with a ratio of 1/f, resulting in EEG power bands being generally different (Cohen 2017).

Study Considerations and Limitations

Numerous studies have reported an effect of tonic pain in the gamma band (Gross et al. 2007; Schulz et al. 2015; Wang et al. 2016; De Pascalis et al. 2019). Nevertheless, it has been reported that the gamma band is highly contaminated with muscle artefacts (Hipp and Siegel 2013). Indeed, a number of studies showed that scalp EEG high oscillations have an electromyographic origin (Whitham et al. 2007; Yuval-Greenberg et al. 2008). In the last years, several procedures such as phased noise template removal for power line noise, ICA for correcting extra-ocular muscle activity and mathematical modelling to reduce muscle activity, have been applied to obtain only high-frequency brain generated signals (Nottage and Horder 2016). In this study, only visual inspection and ICA were used to eliminate EMG artefacts. Hence, it is not possible to ensure that the origin of the gamma oscillations was only from brain sources.

The present study has some limitations that are important to consider in future studies that investigate the neural correlates of deep-tissue pain. First, this study did not include a stimulus that was salient but non-painful. For instance, this could be achieved by including another sensory modality, e.g. a loud sound, whose salience could be comparable to that of the painful stimulus (Mouraux et al. 2011). This could help elucidate the specificity of the activity within the investigated regions during sustained deep-tissue pain.

Second, eLORETA, which is an improved version of LORETA, was used to estimate the activity in several ROIs involved in pain processing. This technique has been validated and reported as having no localization bias (Pascual-Marqui 2007). Still, the spatial accuracy of eLORETA, like other inverse solution methods, is highly dependent on a number of parameters, e.g. the number and location of electrodes and whether individual MRI data or a template model is being used (Michel et al. 2004; Michel and Brunet 2019).

Third, it is important to highlight that the present findings were mostly related to the contralateral regions, and, to a large extent, the discussion did not consider their laterality during pain. Numerous studies reported that unilateral painful stimulation evokes bilateral activity in a number of regions (Downar et al. 2003), but contralateral regions generally display higher brain activity compared to the ipsilateral counterparts (Shenoy et al. 2011; Yücel et al. 2015). In this experiment, the participants used their left hand to score the ongoing perceived intensity using a VAS device, and brain activity evoked by this action could mask / interact with pain-evoked activity (Fuchs et al. 2000). Hence, considering that this study aimed to identify differences in brain activity between painful and non-painful stimulation conditions, it was decided to focus on the activity of contralateral regions. Nonetheless, laterality should be considered in future experimental designs.

Finally, 64-channel configuration and a template head model were used to calculate the EEG forward solution and locate the electric brain sources. This approach simplifies the experimental procedure, circumvents the lack of MRI capabilities and avoids the labour-intensive task of measuring the electrode positions on every subject. On the other hand, individual differences in the anatomy of the head and the electrode positions are not taken into account, which can lead to a limited spatial accuracy. Although it is possible to obtain acceptable estimates of the location of the brain sources with these settings, more accurate estimations are obtained when using a whole-head, dense-array sampling (> 256 channels) and the individual MRI of each participant. Due to this constraint, other regions that have been reported to be involved in pain processing (e.g. the primary motor cortex or posterior insula) that are located within a proximity of 3 cm to other regions already included, were disregarded from the analysis (Michel and Brunet 2019). Similarly, deep structures that are important in the endogenous modulation of pain, such as those located in the brainstem, were not included in this analysis. Finally, limiting the number of ROIs is reducing the risk of false positives (type I error).

Conclusions

The present findings indicate that sustained deep tissue pain evokes different responses on brain oscillations compared with innocuous deep tissue stimulation. Results suggest that the increased activity observed among the considered areas between painful and non-painful stimulation (SDTP vs SDTnP and VT) denote physiological differences in neuronal activity that are linked to pain processing. Further studies are required in order to disentangle mechanisms related to pain processing from saliency.