Participants. Participants were 63 neurotypical right-handed university student volunteers, aged between 18 and 29 years (32 women: M = 21.56, SD = 2.41, men: M = 23.03, SD = 2.63). We excluded one male participant from data analyses because we detected outliers in his data. Thus, only data from 62 participants were analyzed.
The experimental protocol was conducted under the Helsinki Declaration (1964) and approved by the Institutional Review Board (IRB) of the Department of Psychology of Sapienza University of Rome (protocol number 0001291 issued on 07/12/2017). Informed consent was obtained from each participant (see Supplementary Information, section S1, for more details).
Questionnaires. The participants completed the RST-PQ32 measuring three major systems: the BAS, BIS, FFFS. The BAS is composed of the following facets: Goal-Drive Persistence (BAS-GDP), Reward Interest (BAS-RI), Reward Reactivity (BAS-RR), and Impulsivity (BAS-I). The total BAS (BAS-TOT) measure is obtained by summing the BAS-GDP, BAS-RI, BAS-RR, and BAS-I scores.
We also administered the ECQ34 consisting of five facets. From The ECQ facets, we derived the following principal scores (see34): Cognitive Empathy (CE), Affective Empathy (AE), Empathic Drive (ED), Total Empathic Ability (TEA), and Cumulative Total Empathy (CTE) scores. More details are provided in Supplementary Information, section S1. Participants also completed the state anxiety form of the State-Trait Anxiety Inventory (STAI-Y1)35.
Experimental trials and treatments. To investigate pain-related empathy, we benefit from a known paradigm developed by Singer and colleagues36. Rutgen et al.33 and ourselves23 employed this paradigm to test empathic experience wherein the object of empathy experience was a real person seated on the left side of the participant chair (see Fig. 1). In the self-pain condition, participants were exposed to individually calibrated, short-lasting painful electric stimuli (duration from 18 to 30 ms) and non-painful electric stimuli delivered to the back of their right hand. In the other-pain condition, participants experienced empathy for the pain of the confederate seated next to which we delivered painful stimuli and non-painful electric stimuli to the back of her right hand. Each of the self-pain and other-pain conditions took about 24 min, where painful stimuli were delivered respectively to the participant and the confederate in random order. We used the e-prime 2.0 system to program the self-pain and empathy for pain trials (the trial structure, stimulation, and timing are provided in Fig. 1; see supplementary materials, section S2 for more details). In the self-pain condition, participants rated, after the presentation of a painful stimulus their experienced pain and unpleasantness on a numerical 7-point Likert scale (from 1 = “not at all” to 6 = “very painful”) to obtain a numerical pain score (NPS) and unpleasantness score. Equivalently, participants used the same 7-point Likert scale to rate the inferred unpleasantness experienced by the confederate. Pain and unpleasantness ratings were presented in a quasi-random order. To evaluate the level of relative pain and unpleasantness reduction induced by placebo treatment, we calculated numerical pain and unpleasantness difference scores (NPDSs and NUDSs) by subtracting NPSs, and NUSs rated during placebo from scores rated during pain. We used these difference scores for statistical analyses.
Procedure. This experiment consisted of two sessions conducted over two days. Participants first signed approved informed consent forms on the first session and then completed the RST-PQ and EPQ. The participant and the confederate were invited for electrophysiological recordings on the second experimental day. The confederate was always a female, as well as the experimenter. Before EEG recordings, each participant underwent a psychophysical pain calibration procedure to determine sensory and pain thresholds making possible a reliable electrical stimulation intensity for painful and no painful stimuli. Details are provided in23 and in the Supplementary Information, section S2. After the calibration procedure, participants were exposed to two experimental testings: a control pain condition and a PA treatment. In the control condition, participants experienced pain without any prescription. In the PA condition, each participant had to ingest a placebo pill and then participate in a pain manipulation procedure known to reduce the first-hand experience of pain (Supplementary Information, S2). The PA treatment made it possible to test whether it modulates empathy for pain. Control and PA treatments turned up in a counterbalanced order across participants. The confederate did not receive any medication, and all participants were purposely informed about this. The partner was seated next to the participant’s left side with the mandatory request to fix their gaze to the ground to prevent direct observation of the other. In addition, each participant also received a mandatory injunction to maintain a fixed eye on the screen and avoid directing the gaze to the confederate. The testing session in total took about 1.9 hours. At the end of the experiment, we dismissed participants after filling the state anxiety inventory (STAI-Y1).
A power increase relative to baseline level can be observed in response to all stimuli during pain compared to placebo treatments. This increase is much more pronounced in the self-pain between 100 to 250 msec. The maximum relative increases during the pain of TF power were at 7 Hz, 11 Hz, 18 Hz, 31 Hz, and 39 Hz, as shown by the arrows in the upper-right panel. The power increases can be observed at all midline electrodes but are more assertive at central locations.
EEG Recordings and Wavelet Analysis. EEG activity was recorded from 30 scalp sites according to the extended 10-20 system, with the addition of two earlobes electrodes (A1, A2) using 32-tin electrodes stretch Lycra cap with a ground electrode mounted between FPz and Fz (Electro-Caps, Eaton, OH, USA). The NuAmp acquisition system (Neuroscan Acquire 4.3, Compumedics Neuroscan Inc, Charlotte, North Carolina 28269, USA) with an online notch filter at 50 Hz. The reference electrode was at the linked earlobes [(A1+A2)/2]. The electrode impedance was kept less than 5 kΩ. The EEG was recorded in DC mode (sampling frequency = 1000 Hz, gain = 200, bandpass = 0.01–100 Hz: Butterworth zero-phase filter with 24 dB/octave roll-off) with an online 50 Hz notch filter. Both vertical and horizontal eye movements and eye blinks were monitored. Trials contaminated by eye blinks, eye movements, or electromyographic (EMG) activity exceeding ±75 µV at any electrode were excluded from the analyses. Then, the EEG signals were downsampled to 250 Hz and transformed to standard average reference to obtain reference-free recordings. We removed horizontal and vertical EOGs and EMG artifacts by extracting 1 to 3 out of 30 independent components (IC; using Infomax algorithm, Brain Products; Vision Analyzer 2.2.2, Gilching, Germany)37. We reconstructed the EEG trace into discrete, single-trial 1000 ms artifact-free epochs (from 33 to 36) that were time-locked to the offset of painful electric-train stimulus delivered to the participant and to the onset of red-spark visual cue for the painful stimulus delivered to the confederate (see Fig. 1) with a 500-ms prestimulus baseline. For each treatment, we first calculated ERPs in self-pain and other-pain conditions. We subtracted ERPs in each stimulus condition from the corresponding EEG epoch to remove the phase-locked EEG activity from the EEG data.
A time-frequency (TF) representation based on the continuous Morlet wavelet transform (CMWT) of every single EEG epoch (explored frequencies: 1-40 Hz, 1 Hz step) was used to identify non-phase-locked (stimulus-induced) power modulations of oscillatory activities (for details see Supplementary Information, S3). To enhance EEG changes time-locked (but not phase-locked) to stimulus onset, the CMWT was applied to each trial. The Resulting TF power maps were then averaged across trials for each subject and within each pain condition. These maps express the average oscillation power as a function of time and frequency.
We considered the mean TF real power of the prestimulus period (between -500 and -50 ms) as a baseline level. For each frequency step, these baseline levels were subtracted from the prestimulus and post-stimulus power. Grand averages of induced TF representations of the power values at electrode Cz are displayed in Fig. 2 for the first-hand pain and other conditions. We obtained significant t-values (see right side of Fig. 2) for the following five EEG dominant sub frequencies and time-intervals: ϑ (4-8 Hz, 50-250 ms); α (9-13 Hz, 100-200 ms); β1 (14-21 Hz, 100-200 ms), β2 (22-32 Hz, 100-180 ms), γ (33-40 Hz, 120-180 ms). We first obtained the maximum amplitude for each of these frequency bands of interest and the associated frequency (7, 12, 18, 31, and 39 Hz, respectively). We then computed the current source density (CSD, µV/m²) transforms of extracted wavelet waveforms at each frequency of interest mentioned above (for more details, see Supplementary Informations S3). We used the CSD transform as a spatial filter to identify the topographical source at maximum amplitude for each waveform of interest38. These CSD maps indicated that midline frontal (Fz), central (Cz), and parietal (Pz) are sensitive sites to experimental manipulations (Fig. 3).
Color current source density maps (µV/m²) are reported at the bottom for each frequency of interest (7, 11, 18, 31, and 39 Hz) and the time corresponding to each maxima amplitude for each frequency.
HR recordings. We recorded the electrocardiogram (ECG) using two beryllium copper electrodes (1.5 cm in diameter) with a sample rate of 100 Hz. We processed the continuous ECG recording signal with Kubios HRV Analysis 3.0.2 software39 to obtain the HRV measures used in the present study. Based on our previous HRV findings20, we selected time domain, frequency domain, and sample entropy measures.
Reduction of physiological variables. We derived Control minus Placebo difference scores (∆) in the R-R time interval that we labeled as ∆tHRV (ms), the standard deviation of normal-to-normal R-R interval (∆SDNN, ms), Low-Frequency power (∆LF power, 0.04-0.15 Hz), and High Frequency (∆HF power, 0.15-0.4 Hz), LF/HF ratio, Sample Entropy (∆S-Entr). More details on HR recordings and HRV are available in39 and Supplementary Information, section S3.
For the EEG oscillation measures, to reduce skew, we derived Control minus Placebo difference scores of natural log transformation of TF mean power calculated for each of the ϑ, α, β1, β2, and γ frequency bands across Fz, Cz, and Pz leads.
We performed five varimax-rotated Principal Components Analyses (PCAs) to reduce data dimensionality, one for each of the five frequencies of interest and separately for self-pain and other-pain conditions, on the HR and EEG frequency indices (see Supplementary Information, section S4). Each of the five PCA involved six HRV difference indices, as reported above, and three EEG Control minus Placebo difference indices as obtained across Fz, Cz, and Pz midline scalp sites of interest. These analyses served to select (i) the EEG indices loading above the threshold of 0.40 in a factor together with HVR indices, (j) to reduce problems of multicollinearity, for each EEG frequency of interest, in the subsequent analyses. Results of these preliminary analyses for self-pain and other-pain for ϑ, α, β1, β2, and γ EEG frequency bands of interest are reported in Table 1.
For the self-pain condition, each of these separated PCAs (varimax rotation) yielded a three orthogonal factors solution (eigenvalues >1) that were exported as standardized factor scores and used for the correlation analyses. In terms of HRV changes, common to all these analyses was the first factor loading on frequency domain HRV difference scores (∆) that we labeled as “S_∆fHRV” (S stands for self-pain). Additionally, we obtained a combined factor loading on ∆SDNN and sample entropy changes that we labeled as “S_∆SDNN & ∆S-Entr.” In terms of EEG band power changes, we obtained two factors, one loading on β1 power and the other on β2 power, obtained at midline sites (Fz, Cz, and Pz) that we labeled as “S_∆Midl-β1Pow” and “S_∆Midl-β2Pow”. We also obtained the following composite factors including HRV measures and ϑ, α, and γ power changes: “S_∆tHRV & ∆Midl-ϑPow,” “S_∆SDNN & ∆S-Entr & ∆Cz-αPow,” “S_∆tHRV & ∆CzPz-αPow,” and “S_∆tHRV & ∆CzPz-γPow” (see loadings in boldface reported in the upper section of Table 1). Descriptive statistics for these factors are reported on the left side of Table 2.
Similar separate PCAs on physiological difference data performed for the other-pain condition yielded a three orthogonal factors solution. In terms of HRV changes, common to all these analyses was the first factor loading mainly on frequency domain HRV difference scores, and we labeled it as “O_∆fHRV” (O stands for other-pain). We also obtained a combined factor loading on time HRV and sample entropy changes labeled “S_∆tHRV & ∆S-Entr.” In terms of EEG band power changes, we obtained four factors loading on ϑ, β1, β2, and γ powers, across the three midline sites (Fz, Cz, and Pz) and labeled respectively as “O_∆Midl-ϑPow,” “O_∆Midl-β1Pow”, “O_∆Midl-β2Pow”, and “O_∆Midl- γPow.” For the α band, we also obtained a factor including the α power differences at Fz and Cz leads that we labeled as “O_∆FzCz-αPow.” All these factors can be derived from loadings in boldface reported in the lower section of Table 1). Descriptive statistics for these factors are reported on the right side of Table 2.
Statistical analyses. We first calculate partial Pearson correlation coefficients separately for the self-pain and other-pain conditions to determine the association of NPDSs, NUDSs, RST-PQ, and ECQ personality traits with difference scores (Pain Control minus PA) on physiological factors. The potential contribution of gender and state anxiety difference scores (control minus placebo) was partially out from these correlations. We also calculated a partial Pearson correlation matrix (gender scores were partially out) among personality traits of interest and NPDSs and NUDSs. The probability levels were corrected by applying the false discovery rate correction (FDR) method40 to control false-positive errors. Among physiological factors significantly correlated with a personality trait, we want to select the best predictors of this trait by avoiding collinearity among them. Thus, we first assess collinearity diagnostics using the Proc Reg procedure available in the SAS-9.4 system. We then solved the collinearity problem by implementing the Elastic Nets method provided by the Proc Glmselect procedure available in the same statistical system. This analysis can overcome the limitations on the variable selection, usually presented in other available similar methods. It can select more than one variable and achieve a better model prediction (see, e.g., 41). Separately for self-pain and other-pain conditions, we applied the above-described method to select physiological factors as predictors of pain and unpleasantness difference scores (i.e., S_NPDSs and S_NUDSs, O_NPDSs and O_NUDSs). We set a significance level at p=0.05 after FDR correction. We tested multiple and simple mediator models evaluating the role of personality traits as mediators for the influence of the selected physiological factor on NPDSs and NUDSs. We tested this effect by using the conditional process analysis42. The PROCESS macro (www.afhayes.com) tests model-6 (with two personality mediators) or model-4 (with one personality mediator) in all regression analyses.