Pain Tolerance Test

Pain tolerance was determined using a tonic heat pain model (Figure 1). Pain was induced by exposing each participant’s hand to hot air at 60.0°C ± 0.5°C. Hot air was produced by Peltier modules and circulated in a hermetic box made by our laboratory. This model at 70.0°C can produce intensive pain perception within 3 min after the pain threshold (Zhou et al., 2015a); we therefore choose a lower temperature to extend ET during pain perception. The heat pain model shows less confounding with thermoregulatory cardiovascular reactivity observed in tonic cold pain model (Naert et al., 2008).

Paradigm of tonic pain procedure.

Participants were instructed to insert their entire left hand into the box and keep it open and still. All participants were right-handed and, consequently, were able to score precisely their pain perception on the visual analog scale (VAS) with their dominant hand after withdrawal of the left (non-dominant) hand. Hand temperature was monitored continuously throughout the entire procedure. Subjects were required to announce three stages of sensation: (1) the first pain sensation (pain threshold); (2) significant pain, that is, when pain was rated 50 on a 100-point VAS; and (3) when pain became intolerable. The test ended when each participant took his/her hand off the box or 10 min after the pain threshold was reached. This maximum ET was set because our preliminary tests of the model showed that an air temperature of 60°C induces a mean maximum hand temperature of 42°C ± 1°C in 8 ± 2 min, followed by a consistent hand temperature with small fluctuations within 1°C. Perception of pain accordingly remained unchanged or was even reduced in a few subjects.

Pain tolerance was estimated by the ET to pain. Three types of ET were measured (Figure 1): ET-mild pain, defined as the time from the pain threshold to significant pain; ET-significant pain, defined as the time from the announcement of significant pain to the withdrawal of the hand; and ET-total, defined as the time between the pain threshold and the removal of the hand. Perceived pain intensity and unpleasantness were assessed using VAS. Participants rated pain intensity from no pain (0) to worst possible pain (100) and unpleasantness from not unpleasant (0) to extremely unpleasant (100).

Electroencephalograms were recorded during 2-min eyes-open resting state (i.e., pain-free state) and subsequently the tonic pain test (for details, see section “Procedure”) from 32 Ag/AgCl active electrodes (BioSemi^®; Amsterdam, Netherlands) mounted in an elastic cap. Electrode positions included the standard International 10–20 system location and intermediate positions (Klem et al., 1999). The EEG was digitized at 512 Hz, with an amplified band-pass of 0.1–100 Hz. Additional electrodes were placed on earlobes as references (averaged offline). Vertical and horizontal electro-oculographic (EOG) potentials were recorded from bipolar derivations to detect ocular movements. In order to minimize noise originated from movement, participants were instructed to fixate a black computer screen placed in front of them, keep still, and not talk except for verbal announcing three stages of sensation (i.e., “pain started,” “significant pain started,” and “intolerable pain”). Pain ratings were assessed after the tonic pain test (i.e., after withdrawal of the hand), to avoid motor artifacts of ongoing EEG recording during the test. To focus on brain responses related to pain processing, GBO data of pain were limited to the ET-significant pain period (varying from 1.1 min to 8.6 min, on average 4.3 min; Figure 2) to ensure that brain activities of pain processing were represented.

Pain measures during pain tolerance tests. (A) Exposure time, (B) perceived pain ratings for pain intensity and unpleasantness, (C) correlation between frontal composite score and exposure time to significant pain. *p < 0.05 for post hoc analyses of paired comparisons.

The EEG data were preprocessed using EEGLAB, an open source toolbox running in the MATLAB environment (Delorme and Makeig, 2004), and in-house MATLAB functions. Data preprocessing consisted of the following steps: (1) the recorded EEG signals were band-pass filtered between 0.5 and 100 Hz; (2) a notch filter was used to eliminate 50-Hz line noise; (3) EEG data were re-referenced to a common average reference; (4) data portions with large drift were removed; (5) channels with bad activation were interpolated using a spherical spline method; (6) EEG epochs were then visually inspected, and trials contaminated by artifacts due to saccadic movement detected by EOG were removed; (7) the Blind Source Separation was used to correct the data portions contaminated by eye blinks and movements, electromyography, or any other non-physiological artifacts; (8) the data portions reflecting resting-state period (RS) and significant pain period (SP) were extracted, respectively; (9) the EEG signals within the above two periods were segmented into 1,000-ms epochs; (10) EEG epochs with amplitude values exceeding ±80 μV at any electrode were rejected. For a SP, an average of 79 epochs, 199 epochs, and 318 epochs (71.1%, 68.3%, and 88.1% of the total number of epochs) were remained for elderly LPs, elderly HPs, and young participants, respectively.

The EEG spectral power was analyzed with in-house MATLAB functions. For each participant and each period, the segmented epochs were transformed to the frequency domain based on fast Fourier transforms (FFTs), yielding FFTs ranging from 1 to 100 Hz with frequency resolution 1 Hz. The power spectra were calculated as the magnitude-squared FFTs averaged across epochs. The EEG power of each frequency band was obtained for each electrode, each period, and each participant through averaging the power spectral of each frequency bin within its corresponding frequency limits: delta (∼1–4 Hz), theta (∼4–8 Hz), alpha (∼8–13 Hz), beta (∼14–30 Hz), low gamma (∼30–48 Hz), and high gamma (∼52–100 Hz). We focused on high-frequency brain oscillations in the gamma band (52–100 Hz). We computed the pain-related change of powers between the two periods (i.e., the RS and SP) for each electrode and each participant, using the following equation: pain-related change of powers% = (SP - RS)/RS, where SP is the spectral power within the SP, and RS is the spectral power within the RS. The SP was chosen to ensure the recorded oscillations were induced by tolerating pain.

Artifact-free EEGs were quantitatively analyzed to determine source localization using standardized low-resolution brain electromagnetic tomography (sLORETA), determined using LORETA-KEY^©®, a publicly free academic software, located at http://www.uzh.ch/keyinst/loreta.htm (Pascual-Marqui et al., 2011). Among all the source localization methods, the sLORETA provides a weighted minimum norm inverse solution and has correct localization even in the presence of structured noise, albeit with low spatial resolution (Aoki et al., 2015). The solution space of sLORETA is restricted to cortical and divided into 6,239 cortical gray matter voxels at 5-mm resolution using the MNI152 template. The amplitude of each EEG rhythm (i.e., delta, theta, alpha, beta, and gamma) was obtained for each participant and each period (i.e., RS and SP). The source signals were analyzed through the following two approaches. In the first approach, we explored the cortical distribution of group-level source solutions of GBOs during SP state. The 6,239 cortical gray matter voxels were divided into 84 Brodmann areas. Group-level source solutions were obtained via averaging the source solutions across subjects in each group. The estimated localization of GBOs during SP was identified as the 10 Brodmann areas with largest amplitudes. In the second approach, changes in GBOs induced by pain were estimated by one-sample t test conducted on the pain-related change of GBOs (i.e., comparing SP from RS) for each group and each voxel. The localization of GBO changes was resulted in three-dimensional images where cortical voxels of statistically significant differences were identified. The significance level (p value) was corrected using false discovery rate procedure for multiple comparisons.