When touch predicts pain: predictive tactile cues modulate perceived intensity of painful stimulation independent of expectancy

2.1 Participants

The study was advertised using flyers, social media (www.bodyinmind.org and university channels) and via word-of-mouth. Participants were reimbursed AU$20 for their participation. Participants were excluded if they reported current pain, or any psychological or neurological impairment. Participants were aware that the study was investigating the relationship between pain and touch but were not informed about the experimental stimulus contingencies or the hypothesis. Sample size was chosen to ensure adequate (80%) power to detect a small to medium effect. Fifty- eight healthy, pain-free individuals (33 females, mean age = 25 years, SD= 9) volunteered. The study was conducted at the University of South Australia and approved by the institutional ethics committee (Protocol #0000031100).

2.2 Stimulus material

Vibrotactile stimuli (CSs) were delivered by 10 mm, 3 V, off- centre, electric motors (tactors). The CS+ and CS-tactors were set to vibrate at the same, easily detectable, non-painful intensity, at 220 Hz. Tactors were not placed over superficial bone. The intensity of the tactors was calibrated on a pilot sample and location of the CS+ and CS-tactors (proximal or distal forearm) was counterbalanced among participants to eliminate any systematic effect on sensation of the anatomical location of the tactor. The tactors were attached to the skin using a double-sided adhesive sticker.

Electrical stimuli (USs) were delivered using two Digitimer DS700 electrical stimulators (Digitimer Ltd, Welwyn Garden City, UK) as 20 ms bursts with instantaneous rise and fall. The US electrodes were placed on the forearm midway between the two CS tactors, which were 8 cm apart (see Fig. 1). This arrangement was chosen to enable spatially distinct conditioned stimuli while maintaining a close anatomical relationship between each CS and the USs.

Fig. 1

Experimental set-up showing the placement of the tactile and electrocutaneous stimuli. Seated participants rested one arm on a table behind a board, so that it was out of view. Painfulness and unpleasantness of each US were rated using the other hand

The US intensities were individually calibrated, serving as the US^low, US^med and US^high. Calibration was performed using a staircase method with a single ascending run of 0.1 mA increments. During this procedure, the intensity of the stimulus was rated using a 0-10 numerical rating scale (NRS) where 0 = no sensation at all, 1 = you feel a sensation, but it is not painful, 2 = you feel a strong sensation, but it is not painful, 3 = starts to be painful, but is a very mild pain, 8 = significantly painful and demanding some effort to tolerate, and 10 = the worst imaginable pain. The stimulus intensity (mA) corresponding to ratings of 2, 4 and 8 on this NRS were used as the US^low, US^med (test stimulus) and US^high, respectively. Participants were free to stop before they reached a rating of 8, but if they did not reach at least 6 for the US^high, the experiment was discontinued.

2.3 CS duration

Participants were randomised to one of three groups. The only difference between groups was the duration of the CS, which was varied in order to differentiate expectancy and non-expectancy mediated effects (see Fig. 2). In the Long CS group, the onset of the CS+ was 4 s prior to the US. In this condition, any effect could be explained by the participants’ conscious expectation and anticipation of pain. In the Short CS group, the onset of the CS was 110 ms prior to the US. According to Pastor (2004), temporal discrimination of tactile stimuli on the forearm occurs at intervals of greater than 30-50 ms. Therefore, this timing enabled temporal differentiation of the CS and US, yet did not provide sufficient time for participants to deduce and anticipate the outcome. That is, the duration of the CS was less than that which would enable it to be recognised and analysed with respect to past experience, a process that would be required before an outcome could be consciously predicted (common estimates of pre-conscious processing are in the order of 250 ms [18]). In the CS-US Indistinguishable group, the onset of the CS was only 20 ms prior to the US. In this condition, not only could the US not be consciously predicted on the basis of the CS, the two stimuli could not even be temporally differentiated. As can be seen in Fig. 2, the CS tactors overlapped with the 20 ms US (delay conditioning), probably more closely mimicking the clinical relationship between somatosensory and nociceptive input.

Fig. 2

Duration and timing of the vibration and electrical stimuli in the three CS duration groups. The black bar represents the vibration and the red bar represents the 20 ms electrocutaneous stimulus.

2.4 Experimental design

We used a differential classical conditioning design with stimulus type (CS+ vs. CS-) as the within-subjects factor, and CS duration (4 s, 110 ms or 20 ms) as a between-subjects factor.The experiment consisted of two phases: the acquisition and the test phases. During the acquisition phase, a tactile cue at one location on the forearm (CS+) was always followed by a high intensity electrical stimulus (US^high), and a tactile cue at another location on the forearm (CS-), was always followed by a low intensity electrical stimulus (US^low). During the subsequent test phase, the effects of the CS+ and CS- on pain evoked by a medium intensity painful stimulus (US^med) were determined.

During the acquisition phase, the CS+/US^high and CS-/US^low pairs were presented 12 times each. During the test phase, the CS-/US^med were paired 10 times, while the CS+ was paired 5 times with the US^med, and 5 times with the US^high. In this phase, both CSs were paired with the US^med so that their modulating effect on the pain evoked by an identical test stimulus could be determined. We included CS+/US^high presentations during the test phase as ongoing reinforcement, to prolong differential learning and prevent extinction. Including CS-/US^low stimuli might have assisted this further, but the experimental set-up restricted us to two US intensities per phase. Stimuli were delivered according to three pre-selected, semi-randomised sequences, with the restriction that no stimulus pair could be presented in more than three consecutive trials. Participants were allocated to one of the three sequences according to a pre-randomised order, with each sequence allocated to six people in each group. Each CS-US paired presentation was separated by as inter-trial interval.

2.5 Measurement

During the experiment, participants were asked to rate pain intensity and unpleasantness for each electrical stimulus, immediately after the stimulus presentation, during the inter-trial interval. Participants recorded their pain intensity ratings on a 150 mm visual analogue scale (VAS) where 0 = no pain and 150 mm = worst imaginable pain. Participants similarly recorded their pain unpleasantness ratings on a 150 mm VAS where 0 = not at all unpleasant and 150 mm = most unpleasantness imaginable. Data were extracted and entered by an assessor who was blinded to stimulus type.

2.6 Protocol

After being informed of the broad nature of the study, participants gave signed informed consent. The electrodes and tactors were then attached (see Fig. 1) and calibration of the different US types was initiated. Once the acquisition phase was complete, the intensity of the current stimulator delivering the US^low was increased to the pre-calibrated US^med intensity for the test phase. Thus there were no US^low presentations in the test phase.

2.7 Software and apparatus

The experiment was executed using a custom-built, Arduino™- based module, which controlled the timing and intensity of the two tactors, and triggered the two Digitimer DS700 electrical stimulators via TTL signalling. The pre-randomised stimulus presentation sequences were uploaded to the Arduino module using customised Lab VIEW^TM software. Electrodes and tactors were placed on the left arm, which was positioned behind a screen to eliminate visual feedback. Participants wore white-noise-emitting headphones to eliminate auditory feedback and background noise. During the procedure, the experimenter was in a partitioned section of the laboratory and did not interact with the subject.

2.8 Differential conditioning manipulation checks

Because potential limitations in the experiment set-up could influence differential learning, aspects of the set-up were assessed in order that their effect on conditioned responding could be discussed and accounted for by secondary analyses. These potential limitations were: 1. insufficiently distinct US^low and US^high to enable differential learning, and 2. inability to spatially discriminate between the conditioning stimuli. Furthermore, because evidence suggests that differential conditioning is highly correlated with propositional learning [14,15], awareness of the contingency between tactile cues and painful stimulation was also assessed.

In order to assess whether the US^low and US^high were sufficiently distinct with respect to quality (non-painful vs. painful) and intensity, a criterion was set to score this feature of calibration as ‘valid’ or ‘not valid’. To be valid, the mean US^low had to be less than 30 mm on the 150 mm VAS, because this was considered to be relatively non-painful, and the mean US^high had to be at least 30 mm greater than the US^low. Thus, only subjects whose calibration was classified as ‘valid’ were expected to show a meaningful differential effect. In order to ensure that the CSs were spatially distinct, each participant’s ability to discriminate between CSs was confirmed upon completion of the study by stimulating each location, asking participants to identify the locations of the stimuli as the same or different, and to correctly identify each location (proximal or distal).

In order to assess post-experimental propositional knowledge of the CS-US association, participants were exposed to the CS stimuli and asked: “Are you aware which of these stimuli were most often paired with higher pain intensity and, if so, which?” If participants were aware and correctly identified the CS+, they were classified as contingency aware.

2.9 Statistical analysis overview

3.1 Participants

One participant was excluded because of complete loss of electrode conductance. Three participants were excluded because of technical failure. As such, fifty-four healthy, pain-free individuals (31 females, mean age = 26 years, SD = 9) completed the study.

3.2 Stimulus characteristics

The perceived intensities of the electrical stimuli for each phase are shown in Fig. 3. On average, the US^high was rated more painful that the US^low (p < 0.001). The US^med was successfully delivered at an intensity between the intensities of the US^high and the US^low. All participants were able to discriminate between the CS locations when individually presented post-experimentally.

Fig. 3

Mean pain intensity ratings for each stimulus type. Intensity ratings measured in millimetres on a 150 mm VAS. Error bars denote SE.

3.3 Testing the primary hypotheses: effect of predictive cues and CS duration on pain intensity

Pain evoked by a moderately painful stimulus was greater when it followed a tactile cue that had predicted a high intensity stimulation during acquisition, than when it followed a tactile cue that had predicted a low intensity stimulus during acquisition. That is, there was a main effect of CS type; F (1, 51) = 8.1, p = .022, η²_p = 0.1. The relative duration of the tactile CS did not significantly alter the effect of CS on perceived pain (CS type x Group interaction; F (2, 51) = 0.9, p =0.4, n.s. (Fig. 4, panel 1). Mean (95% CI) difference in pain intensity between the CS+/US^med combination and the CS-/US^med combination was 3 mm (CI 0.4-4.8 mm). Post hoc comparisons examining the effect of predictive cue (CS-type) on pain intensity ratings within each CS duration group did not support the main finding (Holm-Bonferroni corrected p = 0.24,0.26 and 0.09 for the 4 s, 110 ms and 20 ms durations respectively).

Fig. 4

Panel one shows the mean test phase difference between CS+/US^med and CS-/US^med VAS pain ratings (mm) for each CS duration group (4 s, 110 ms and 20 ms) and overall (pooled). The second and third panels show the mean test phase difference between CS+/US^med and CS-/US^med pain ratings (mm) for subgroups where calibration was classified as ‘valid’ or ‘not valid’ and for participants who were aware or not aware of the CS-US contingencies. Error bars denote 95% CI.

3.4 Secondary analyses—unpleasantness

Perceived unpleasantness of the US^med was not significantly different following a tactile cue that had predicted a high intensity stimulation during acquisition, than following a tactile cue that had predicted a low intensity stimulus during acquisition (Fig. 5). That is, there was no main effect of CS type on unpleasantness; F (1, 51) = .83, p = .37. Post hoc comparisons examining the effect of predictive cue (CS-type) on pain unpleasantness ratings within each CS duration group were also non-significance (Holm-Bonferroni corrected p = 0.42,0.42 and 0.14 for the 4 s, 110 ms and 20 ms durations respectively).

Fig. 5

Mean test phase difference between CS+/US^med and CS-/US^med unpleasantness ratings (mm) overall and for each CS duration group. Error bars denote 95% CI.

3.5 Effect of US intensity calibration

Twenty nine of the 54 participants met the criteria for successful calibration (US^low <30 mm, US^high >30 mm with respect to the US^low) and were classified as valid (see Table 1). The mean CS+/US^med vs. CS-/US^med difference was 4.8 mm (CI 0.9-8.7) in the valid group and 0.2 mm (CI -1.5 to 2) in the non-valid group (Fig. 4, panel 2). Repeated measures ANOVA [2 (CS type: CS+ vs. CS-) × 2 (Validity Group: ‘valid’ vs. ‘non-valid’)] confirmed the interaction between the observed effect and validity of calibration, F (1,1) = 4.7, p = .034, η²_p = 0.1. Post hoc t-tests revealed that the ‘valid’ group showed a differential effect, t (27) = 2.6, p = .01, whereas the ‘non- valid’ group did not t (25) = .26, p = .4.

3.6 Relationship between contingency learning and pain

Twenty-two of the 54 participants were able to report the contingency verbally (six, nine and seven people in the 4 s, 110 ms and 20 ms groups respectively). Interestingly, while awareness did not relate to longer CS duration, it seemed to relate distinctiveness of the US^high and US^low. That is, of the 22 aware participants, were also classified as having a ‘valid’ calibration defined by high distinctiveness of the US^high and US^low. While the aware participants had, on average, a greater conditioning effect (see Fig. 4, panel 3), the RM ANOVA [2 (CS type: CS+ vs. CS-) × 2 (Contingency Group: ‘aware’ vs. ‘non-aware’)] did not demonstrate an interaction between contingency awareness and stimulus type F (1, 52) = 1.2, p = .28. Interestingly, those who did not become aware of the CS- US contingency experienced a significantly greater increase in pain responses in the test phase with respect to the US^high (see Fig. 6), t(52) = 2.21, p = .03.

Fig. 6

Mean differences between US^high ratings in acquisition vs. test phase for participants who became aware of the contingencies and those who did not. Error bars denote 95% CI.

4.1 Pain-predicting cues modulate pain intensity independent of expectancy

Previous studies that have also shown that pain can be enhanced by the sensory cues that predict the painful stimulus [4,7,8] have attributed the effect to conscious expectancy. One attempt to isolate the non-conscious effects of pain-associated cues on pain intensity used a paradigm in which one facial expression was used to signal the US^high and another the US^low [6,8]. In the test phase these faces were presented only 94 ms prior to painful stimulation, for just 12 ms. Because of this short duration, participants were unable to consciously discriminate the CSs, yet the CS+/US^med evoked significantly more pain than the CS–/US^med.

The current study extends that finding to the tactile domain. Here, the effect seen in two of the three groups (where the difference in stimulus onset was just 110 ms and 20 ms) could not reasonably be explained by conscious expectation [18]. Indeed, the lack of significant interaction between learning the contingency and the observed effect also supports that conscious learning was not a critical factor in this study. The magnitude of the effect in this study was smaller than that reported by Jensen et al. [6,8]. That we did not have sufficient power to detect the effect in post hoc tests in individual groups emphasises the small effect. This suggests that tactile cues do not have greater associability with noxious stimuli compared with visual cues such as those used by Jensen et al. [6,8]. That we used fewer acquisition trials (24 pairings rather than 50); and that they used visual stimuli with sociological salience and complexity (pictures of faces), may also explain the differences.

Surprisingly, the statistically supported modulation of pain intensity was not paralleled in the domain of pain unpleasantness. It may be premature, however, to conclude that pain intensity may be preferentially modulated by this procedure; particularly because intensity and unpleasantness results were comparable among two of the three groups.

4.2 Theoretical considerations

The effects can be explained by classical conditioning theory. Most applications of classical conditioning to pain research assign pain as the US and assign a behavioural response consistent with fear as the unconditioned response (UR) (e.g. [19]). However, the view of pain as a response [20] invites us to consider a parallel process, whereby the nociceptive stimulus assumes the role of the US and pain assumes the role of a perceptual UR—which may later be contributed to, or perhaps evoked, by a previously inert CS [3]. The ability of a CS to influence US-related responses is thought to develop through adaptation at synapses where neurons encoding the CS converge with those encoding the US [21].

4.3 Adaptive and maladaptive modulation of pain intensity

Like nociception, innocuous pain-associated stimuli signal bodily danger. Where this suggestion of danger is accurate and proportionately enhances pain, the effect might be adaptive, and subserve protection. Where the effect persists, is disproportionate or not specific to the cues truly predicting danger, the effect may constitute maladaptation, and contribute to pain states. The potential for, and determinants of, these maladaptations may be specifically considered in future studies. Notably, the US^high increased in painfulness between acquisition and test phases by a much greater amount in participants who did not learn the contingency than those who did not (VAS change 16 mm vs. 5 mm, see Fig. 6). That contingency naive participants experienced greater increases in pain and had reduced differential (CS+ vs. CS-) conditioning suggests that they experienced a general rather than CS+-specific increase in pain sensitivity in response to classical conditioning. This corroborates the previous finding from our group that failure to learn the predictors of pain results in a non-specific increase in pain responses, likely because of contextual anxiety [33]. That the general pain modulation associated with not learning the contingency was greater than the specific effect of the CS+ relative to the CS-suggests that non-selective learning may be a key maladaptation.

4.4 Clinical relevance

Non-nociceptive cues that might act as predictors of harm have been shown to reduce pain threshold during movement [22] as well as during touch [23], possibly via a classical conditioning pathway [3]. In this sense, the current findings are consistent with the Imprecision Hypothesis of chronic pain [3], which outlines an associative pathway to chronic pain, mediated by maladaptive over-generalisation of conditioned responses.

Because pain-associated cues can enhance later painful experiences, procedures aiming to extinguish these associations may help treat persistent pain. Such treatments could aim to exercise neural networks that encode the non-nociceptive cues independent of those that encode nociception (for example, graded motor imagery [24])—or could aim to disassociate the cues from threat through exposure therapy [25] or pain education [26,30]. The use of approaches to understanding and treating chronic pain that are founded in learning theory is strengthened by evidence suggesting that people with chronic pain may show learning characteristics consistent with maladaptive pain-related learning [27,31,32]

4.5 Limitations

Behavioural studies such as this one rely on self-report of pain as a primary outcome. Although self-report is the gold standard approach to measuring pain, the inclusion of a physiological correlate of pain, for example via brain imaging, would add weight to our finding. Unfortunately, the current design does not allow us to confidently interpret the result as either a hyperalgesic effect of the CS+ or a hypoalgesic effect of the CS-. In a similar study using visual conditioning stimuli, Jensen et al. found that the CS+ resulted in higher pain and the CS- resulted in less pain, relative to a previously unpaired control stimulus [6,8]. They interpreted this to mean that the CS+ vs. CS- difference was due to both hyper- and hypo-algesic effects (i.e. CS+ related nocebo and CS- related placebo effects). There are several reasons, however, that cast doubt over this interpretation. Firstly, the novel control stimulus is not neutral, but rather one that is likely associated with surprise and uncertainty. Secondly, both CS+ and CS-were paired with painful stimuli, and thus both predicted aversive stimulation. Therefore, it is possible that what is actually being observed is three levels of pain amplification by pain-predicting and surprising stimuli. Therefore, while inclusion of a novel control stimulus in our study might have been satisfying, it could not have assisted in answering this question. Another possibility would be to compare the painfulness of the US^med alone (either at calibration or test phase) to when paired with the CS+ and CS-. Unfortunately, comparison to baseline ratings would be confounded by changes in sensitivity and response criteria over time, such as those due to repeated stimulation and emotional modulation. Within-phase comparison to US^med alone, on the other hand, would be confounded by the lack of concurrent tactile stimuli that might modulate pain via neural, attention mechanisms.

Another limitation of this study that could be investigated in future designs is the possibility that the perception of the CSs themselves may have been altered by the conditioning process. That is, the CS+ may evoke more unpleasant or intense perceptions than the CS-, because of its association with the aversive US. This question, however, has been investigated using electrocutaneous CSs with no apparent effect [5]. This suggests that more lengthy and/or aggressive conditioning procedures, than typically used in laboratory studies, may be needed to shift the sensory qualities of somatosensory CSs, if they can indeed be shifted.

Retrospective measurement of contingency learning was used in this study so as to prevent confounding of the primary outcome by experimental demand [28]. Furthermore, there was no way to introduce questions about variables such as expectancy awareness in-between CS and US presentations, because timing was critical to this study. Nonetheless, offline ratings do not necessarily relate well to online measures [29], which may be more sensitive [15].

We also note that we did not account for individual differences in psychological variables that may interact with the conditioning process and with pain reports. Given the limited sample size within each group, these differences may not have been accounted for in the randomization process. This reduces our confidence in the between-group comparisons and should be considered in future studies.

Finally, the magnitude of the induced change in pain observed in the primary outcome may not be clinically meaningful. That is, the presence of the CS+ resulted in only a 5.3% higher intensity rating compared with the CS-. However, this too reflects our aim and a broader limitation of laboratory studies, where a limited number of trials, an artificial context and the use of experimental pain are likely to reveal only glimpses of what is clinically possible. Available data suggest that pain-associated cues may be more potent in clinical states [22,23]. Nonetheless, on the basis of our data, it remains possible that the potential effect of pain-associated tactile cues is not clinically relevant. It should be noted, however, that the effects seen in subgroups of people where the calibration was optimal and where the contingency was learned produced greater effects.