*2.5. EEG Processing*

The EEG was processed using Matlab (The Matwhorks, Inc., Natick, MA, USA) and the EEGLab toolbox [67], following a previously proposed standard pipeline [54,68].

First, the data were re-referenced to the linked earlobes and down-sampled to 512 Hz. Then, a band-pass filter (0.1–30 Hz) and a notch filter (50 and 100 Hz) were applied in order to remove the physiological and external noise. The Artefact Subspace Reconstruction (ASR) with a default cut-off parameter (k = 20) was applied in order to remove non-stationary artefacts [69]. The data were then decomposed into Independent Components (ICs) using the SOBI algorithm [70]. By using the neural-net based classifier ICLabel [71], artefactual ICs were identified as those with brain probability Pr{brain} ≤ 0.7 and set to zero, while non-artefactual ICs were back-projected to the original sensor space. A Current Source Density (CSD) reference was then applied in order to increase the spatial resolution of the EEG at the sensor level [72].

Finally, the cleaned EEG was epoched according to the onset and the duration of the EYC and BSL stimuli, as well as the narrative sequences. For each subject, we computed the Individual Alpha Frequency (IAF), which is defined as the centre of gravity of the PSD within the extended alpha range (7.5–12.5 Hz) [73]. In the IAF calculation, we considered, as PSD, the mean PSD aver-aged across all the occipital channels. The occipital PSDs were computed using the EYC data. Finally, we computed 2 canonical EEG bands as: *δ* = [0; IAF − 6] Hz and *α* = [IAF − 2; IAF + 2] Hz [74].

In order to have the highest temporal resolution, all indices were computed following the filtering approach, which is based on filtering and averaging an appropriate set of EEG channels to produce a cluster [54]. The Hilbert Transform was applied to the filtered channels before the averaging to compute the smoothed instant power [75]. The AWI was obtained by subtracting the *α*-filtered right-frontal (FP2, F4, F8, FT8, FC4) and left-frontal (FP1, F3, F7, FT7, FC3) clusters [54], while MI was obtained as the *θ*-filtered left-frontal (FP1, F3, F7, FT7, FC3) cluster [57].

### *2.6. SC and PPG Processing*

The SC and PPG signals were processed using Matlab (Mathworks, Inc.), following a previously proposed standard pipeline [68,76].

The SC signal was first band-pass filtered (0.001–0.35 Hz); then, a threshold for SC extreme values (0.05–60 μS) and extreme rate of changes (±8 μS/s) was used in order to detect artefacts [77]. The artefactual points were replaced by a linear interpolation using adjacent points. From artefact-corrected SC, the tonic Skin-Conductance Level (SCL) was extracted by means of the cvxEDA algorithm [78].

The BVP signal was first low pass filtered (5 Hz); then, all peaks were identified using the Pan–Tompkins algorithm [79], and the instant HR was computed from the inverse of the peak-to-peak distance. Finally, the HR signal was linearly interpolated and filtered with a 2s-long moving average filter in order to obtain a smoother signal.

By means of a trigonometric transformation, SCL and HR were converted into the uni-dimensional EI [57].
