EEG correlates of social interaction at distance

Subjects

Six Italian Caucasian healthy adults were chosen for the experiment, comprised of five men and one woman, with an average age of 35.5 years (standard deviation = 8.3).

They were selected among the members of the EvanLab, the private laboratory involved in this study. The criteria for their voluntary inclusion were their mutual friendship (> 10 years), and their experience in being able to maintain prolonged focused concentration – a product of their familiarity with meditation and other practices requiring control of mental activities.

Statement of Ethics

The use of experimental subjects is in accordance with ethical guidelines as outlined in the Declaration of Helsinki, and the study has been approved by the Ethical Committee of the University of Padova’s Department of General Psychology. Before taking part in the experiment, each subject gave his/her informed consent in writing after having read a description of said experiment.

Equipment

A software program, available at http://dx.doi.org/10.6084/m9.figshare.1466876, especially written by one of our co-authors (GW) administered the sequence of stimuli and synchronized the EEG recordings from each member of the pairs. EEG activity was measured using two Emotiv® EEG Neuroheadsets, equipped with 14 EEG channels, connected via WiFi to a Windows PC.

The technical details are: 14 electroencephalography channels based on international location from 10 to 20 (AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4, plus two reference electrodes). The mastoid electrodes (M1, M2) served as reference points against which the voltage generated from all other electrodes was compared. The sample frequency of the Emotiv® headsets is 128 Hz, with a bandwidth from 0.2 to 45 Hz, with a built-in fifth order low-pass digital filter as well as two notch filters at 50 and 60 Hz respectively as protection against noise produced by the local electricity network. The Emotiv® EEG has a proprietary wireless network connection at a frequency of 2.4 GHz.

Stimuli

The auditory stimulus was composed of a 500 Hz sinusoid applied through 32 Ohm Parrot ZIK® earphones at a volume of about 80 dB. The visual stimulation was from high intensity red LEDs in a 4×4 arrangement placed approximately one meter from the subject being stimulated. The subject kept his/her eyes closed because the light could easily be detected through the eyelids.

Procedure

The members of each pair were placed in two separate rooms approximately five meters from each other. Each room was sound- and light-proof, so as to block out any and all external sensory information.

Between these two rooms was a control room with two computers connected to the Emotiv® headsets and from which the research assistant controlled the sensory stimulation program and each partner’s EEG recording (see Figure S1 in Supplementary material). The software program in use ensured that the signals coming from the two EEG headsets were recorded simultaneously (to within 8 ms).

The partner designated as “Sender” was given the following instructions: “When you are ready, relax and be prepared to receive a visual and auditory stimulus which you will send to your partner. To assist your mental connection with him/her, concentrate on his/her photo before starting the experiment. Your only task is to mentally transmit what you will perceive, while limiting your body movements to prevent interference with your EEG activity. You will perceive 128 stimulations of 1 second each, separated by pauses of random length lasting 4 to 6 seconds in order to avoid predictable rhythms. The experiment will last about 10 minutes.”

The partner designated “Receiver” was given the following instructions: “When you are ready, relax and be prepared to receive the stimuli sent from your partner. To assist your mental connection with him/her, you will see a facial photo of him/her before starting the experiment. Your task is to mentally connect with him/her and try to perceive the stimulus he/she is receiving, while keeping your body still to prevent interference with your EEG activity. The experiment will last about 10 minutes.”

Once the quality of the EEG signals was confirmed, and with the consent of the subjects, the research assistant began running the experiment’s program. To prevent either subject from predicting when the first stimulus would be given, it was preceded by a period of silence of random duration from 2 to 3 minutes.

At the end of the experiment, after a period of rest, in m ost cases (if subjects agreed and time allowed) the role of each subject was reversed.

All together data from 25 pairs of subjects was collected over three days. The raw data are available at http://dx.doi.org/10.6084/m9.figshare.1466876 which include details of pairings.

Bias control

To avoid any experimenter’s effect, the research assistant who managed the software for the data acquisition were blind to the exact start of the stimulation sequence, given the randomization of the duration of the first pre-stimulation period as described above.

The reduction of the risk of any conventional communication between the pair of participants, was guaranteed by the sensory isolation of the two rooms were they were placed as already described. The only remaining possibility was to speak aloud each other, but this event could clearly be noticed by the research assistant.

Data analysis

Collection of the evoked potential was initially conducted by filtering the signals in the 1 – 12 Hz band followed by normalization (see software code of Appendix 1 at http://dx.doi.org/10.1101/022046), hence employing the traditional averaging method of time- and phase-locked epochs. The typical result of an evoked potential obtained from Senders can be seen in the graph in Figure 1.

Figure 1. Typical example of an evoked potential obtained from processing a Sender’s signals.

The graph is an average of 128 stimuli and 14 EEG channels. Usually two peaks are seen, a negative and a positive one, about 250 to 300 ms after the stimulus begins, and a minor peak at about 250 ms after the stimulus ceases.

The average Event-Related Potential (ERP) of all 25 files from Senders and Receivers was also calculated. To get the total sum of evoked potentials from all subjects while avoiding ERP different latency period problems, each subject’s individual evoked potential powers were added up. The resulting graph is shown in Figure 2.

Figure 2. Results of overall average of ERP power: top graph is that of Senders, bottom shows that of Receivers.

Figure 2 clearly shows an ERP in the Senders, but nothing of interest in Receivers.

Following this negative result we began using an original method created by one of our co-authors (WG), and described in detail in Tressoldi et al. (unpublished; pre-print proof available at http://biorxiv.org/content/early/2015/07/06/022046), which was found more resistant to jitter and interferences compared to the traditional averaging method. Furthermore, as described below and in more detail in the original paper, this new processing method is far less prone to unwanted effects of EEG artifact because as the Pearson Correlation depends only on signal phase and not to amplitude.

This method is based on the Pearson correlation between segments of data of fixed length L, as shown in Figure 3.

Figure 3. The window of length L runs along the traces left by two EEG channels.

The corresponding Pearson Correlation is calculated and recorded on the R(x) array.

As an example, the Emotiv® EEG Neuroheadset provides NC = 14 EEG channels and a sample frequency of 128/s; the stimulus is 1 second duration and an epoch’s length is 3 seconds, equal to 384 samples. In this case it becomes possible to calculate the R(x) array in a number of combinations of pairs equal to: Nt = NC_*(NC - 1)/2 = 91. The result can be written using a new array, R(I, X), in which I = 1... 91 and X = 1... 384 are the calculated values. The stimulus is administered at the same time as sample no. 128 and ceases after one second, with sample no. 256. The next processing step involves the average of R(I, X) over all the given stimuli, the result designated as R’(I, X).

Therefore, for each value of I = 1.... 91 a baseline is calculated, comprised of the average of R’(I, X) in the pre- and post- stimulus areas. Finally a new array is calculated, Sync(I, X), based on the result’s absolute value: Sync(I, X) = Abs(R’(I, X-Baseline)).

Then the average of all the Nt combinations gives us the final array, Sync1(X), which represents the total variations of the EEG correlations during a 3 second epoch, for all stimuli and all EEG channels. It is also possible to calculate a similar array, Sync2(C, X), for each channel, C.

To be extra certain, the analysis of experimental data was nonetheless conducted on longer epochs – up to 4 seconds – comprised of 1.5 seconds pre-stimulus, 1 second stimulus, and 1.5 seconds post-stimulus. To calculate the probability that the observed differences in baselines are due to chance, the experimental data were compared to those obtained with a simulation conducted using a bootstrap procedure with the following characteristics:

a) Signals from the desired frequency range (in our case 9–10 Hz in Receivers and 1–16 Hz in Senders) are filtered using a digital filter that leaves signal phases intact, according to the Discrete Fourier Transform (DFT) and its related inverse processing. The filtered files are then saved. We point out that for Senders the standard frequency range (1–16 Hz) was used because the ERP is usually generated in this range.

b) The same processing method (The GW6 method) is applied to these files, but choosing at random the point in which a stimulus is thought to be present. For each file, the same number of stimuli (128) are evaluated, as in the experimental tests. For each file at least 20 bootstrap calculations are made, eventually resulting in over 500 files. The average of these calculations constitutes the blue bootstrap curve in Figure 4 and Figure 5, which therefore represents the expected probability due to chance, to compare with the obtained experimental curves (red). This method appears valid in that it gives a virtually flat curve (blue), close to zero throughout.

Figure 4. Overall result obtained from Senders and filtering all 25 EEG files from 1 to 16Hz + normalization, followed by application of the GW6 method.

On the vertical axis are correlation values x100. The blue curve denotes the average of 500 bootstrap files.

Figure 5. Overall result of 25 Receivers obtained from filtering the EEG signals in a narrow band, from 9 to 10 Hz + normalization, followed by application of GW6 method.

The red curve is the average correlation and the blue is the bootstrap curve (average of 500 files), which represents the expected probability due to chance.

In Figure 4, a distinct peak (red curve) is seen which represents a correlation ERP at about 300 ms from the start of the stimulus, followed by a weaker peak at the end of the stimulus. This graph is similar to the one obtained from standard averaging in Figure 2, only expanded.

In Figure 5, with respect to the Receivers, there is an area that exceeds the normal chance expectation represented by the bootstrap curve. This area is highlighted in yellow and can be calculated as the difference with respect to the bootstrap curve.

A similar analysis was conducted by filtering signals within the 1 to 16 Hz band (as in the Senders) with statistically null results, and subsequently in the 8 to 16 Hz band, followed by the 8 to 12 Hz, but the best result was obtained in the 9 to 10 Hz range (see Table 1).

Statistical control

After having established an increase in cerebral correlation in Receivers coinciding with the remote stimulus given to Senders, it is necessary to determine the importance of this difference with respect to the statistical chance. To this end, instead of resorting to conventional statistical methods, often inapplicable to complex situations such as this, we used an emulation procedure of the Monte Carlo type consisting of the following steps:

a) Take the bootstrap files within the desired frequency band – in our case from 9 to 10 Hz (around 500 files), or from other bands. All of these files are the result of a GW6-type processing and are now the input data set for the Monte Carlo emulation.

b) 25 “fake” files are randomly chosen and their final average is calculated, in the same way as the average of the 25 “real” files.

c) The difference in area with respect to the average of all the 500 bootstrap curves is calculated, as in Figure 5. This difference may be either a negative or positive number.

d) Determine if this number is higher than that obtained from the real experimental files. If it is higher, a counter is incremented.

e) Start again from a) and repeat the cycle as required (we repeated the cycle 2000 times).

At the end, determine how many times out of 1000 a group of 25 bootstrap files randomly exceeds the value of the area found experimentally. The results give the probability of obtaining a surplus of area by chance, as in Figure 6. This procedure does not use any specific a priori statistical model, and is based solely on applying numerous emulations exactly as with real data. As shown in Table 1, in the 9 to 10 Hz band the result is significant to a level of around 2–3/1000, equivalent to P ≤ 0.003, and is even significant in the 8 to 12Hz band, with P ≤ 0.04. The almost Gaussian distribution of the values shows that the method is valid and agrees with normal statistics.

Figure 6. The graph shows the distribution of 2000 Monte Carlo emulations with respect to Receivers’ files filtered in the 9 to 10 Hz band.

The value of the area to be exceeded (see Table 1) is 0.3106 (red line). The distribution is approximately Gaussian and the specified value is exceeded by chance 3 out of 2000 times.