2.4.1.

Coherence coupling analysis

The time-dependent coherence of the fNIRS and systemic physiology time series was averaged over the two conditions. Thus, for each frequency band we obtained two coherence values $c_{SnEC}$ and $c_{SnEO}$, where Sn is one of the 29 possible WTC signals and indexes EC and EO indicate the eyes-closed and eyes-open conditions, respectively [Fig. 3(e)]. The aligned rank transformation analysis of variance (ART ANOVA),⁶¹ a nonparametric ANOVA, was performed to assess any difference between the two conditions [Fig. 3(g)]. In addition to the condition, also the group factor, consisting of two levels (familiar and unfamiliar pair) and the respective interaction (condition × group) were evaluated. Finally, a false discovery rate (FDR) correction was applied to the $p$-values to correct the multiple comparison situation. This analysis was conducted in $R$.

As a control analysis, we calculated the WTC for each type of signal for 2400 random pairs with replacement. The WTC values obtained by this procedure were then averaged over the two conditions and the four frequency bands. For each type of signal and each frequency band, we compared the distribution of true WTC values with that obtained from the randomly shuffled pairs by performing the Wilcoxon rank-sum test. This analysis was performed to assess that the functional hyperconnectivity between pairs occurred because of social interaction.

2.4.2.

Correlated-coherence coupling analysis

The correlation between each time-dependent WTC signal and the remaining 28 time-dependent WTC signals was evaluated [Fig. 3(h)]. The Spearman correlation ($r$) and the significance level ($p$-value) were calculated, generating a $28\times 28$ matrix of WTC time-series correlation coefficients for each individual pair, each condition (eyes-closed and eye contact) and each frequency band [Fig. 3(i)].

Then, a group average was performed by calculating the median of all the individual correlation matrices, obtaining a single correlation matrix for each condition and frequency band, as illustrated in Figs. 4 and 5. For the group average, only correlation coefficients were used that were statistically significantly different from 0 ($p<0.05$) to ensure that the group average was based only on meaningful physiological changes from the individual measurements.

Fig. 4

Group averages of the correlation matrices comparing the eyes-closed and eyes-open conditions in the VLF band (top left and top right, respectively) and LF1 band (bottom left and bottom right, respectively). In the two frequency bands, a marked matrix value indicates a significant difference and a higher median of the distribution than the other condition of the respective frequency band.

Fig. 5

Group averages of the correlation matrices comparing the eyes-closed and eyes-open conditions in the LF2 band (top left and top right, respectively) and HR frequency band (bottom left and bottom right, respectively). In the two frequency bands, a marked matrix value index indicates a significant difference and a higher median of the distribution than the other condition of the respective frequency band.

To investigate whether the correlation distribution of each pair of time series during the eye-contact condition was statistically different from the one during the eyes-closed condition, in each frequency range the Kolmogorov–Smirnov test was performed [Fig. 3(j)]. The significant results are visualized in Figs. 6Fig. 7–8 separated into four bands (VLF, LF1, LF2, and HR) and three groups: fNIRS, systemic physiological and fNIRS/systemic physiological data.

Fig. 6

Probability density functions of the distributions of each correlation coefficient of the 24 pairs in the VLF band separated into three groups: fNIRS, physiological, and fNIRS-physiological data. The probability density functions relative to the eye-contact condition are shown in red and those relative to the eyes-closed condition in blue. In the above figure are reported only the significant differences as a result of the Kolmogorov–Smirnov test. An asterisk indicates that the $p$-values are significant after the FDR correction.

Fig. 7

Probability density functions of the distributions of each correlation coefficient of the 24 pairs in the LF1 band separated into three groups: fNIRS, physiological, and fNIRS-physiological data. The probability density functions relative to the eye-contact condition are shown in red and those relative to the eyes-closed condition in blue. In the above figure are reported only the significant differences as a result of the Kolmogorov–Smirnov test. An asterisk indicates that the $p$-values are significant after the FDR correction.

Fig. 8

Probability density functions of the distributions of each correlation coefficient of the 24 pairs in (a) the LF1 and (b) HR band separated into three groups: fNIRS, physiological, and fNIRS-physiological data. The probability density functions relative to the eye-contact condition are shown in red and those relative to the eyes-closed condition in blue. In the above figure are reported only the significant differences as a result of the Kolmogorov–Smirnov test.

In addition, the difference in the correlation distribution between familiar pairs and unfamiliar pairs groups was assessed. First, the difference of the correlation matrix values evaluated during the eye-contact condition and that during the eyes-closed condition for each of the four bands was computed. Next, a group average was performed by determining the median split for the familiar pairs and unfamiliar pairs as depicted in Figs. S1 and S2 in the Supplemental Materials. The difference between the distributions of the two groups was finally evaluated by Kolmogorov–Smirnov test in each frequency range.

3.1.1.

Differences between eyes closed and open

The fNIRS results of the ART ANOVA are summarized in Table S1 in the Supplemental Materials, and the systemic physiology results are summarized in Table S2 in the Supplemental Materials.

Significant results related to the difference between the two conditions are illustrated in Fig. 9.

Fig. 9

Violin plots of the distributions of the WTC values over the two conditions separated into three frequency bands (LF1, VLF, and HR) and two groups: fNIRS and systemic physiology signals. In the above figure are reported only the significant differences as a result of the ART ANOVA. Double magenta asterisks refer to statistically significant differences between the two conditions even after FDR correction.

After performing an FDR correction, results showed a significant difference between the two conditions in [tHb] (left + right) (LF1 band, $p=0.001$) (Table S1 in the Supplemental Materials), in Temp (left) and HR (VLF band, $p=0.010$ and $p=0.003$, respectively) (Table S2 in the Supplemental Materials). While both [tHb] (left + right) and HR showed a higher coherence in eye-contact condition compared with eyes-closed condition, Temp (left) presented an opposite effect (eye-contact coherence < eyes-closed coherence) (Fig. 9).

Additional significant coherence occurred without FDR correction where coherence during eye contact appeared higher in fNIRS signals in the VLF band, $[{\mathrm{O}}_2\mathrm{Hb}]$ (left) ($p=0.012$), $[{\mathrm{O}}_2\mathrm{Hb}]$ (right) ($p=0.043$), $[{\mathrm{O}}_2\mathrm{Hb}]$ (left + right) ($p=0.031$), and in the LF1 band, $[{\mathrm{O}}_2\mathrm{Hb}]$ (right, raw) ($p=0.017$), $[{\mathrm{O}}_2\mathrm{Hb}]$ (left + right) ($p=0.016$), [tHb] (right) ($p=0.004$) as summarized in Table S1 in the Supplemental Materials and illustrated in Fig. 9. No difference was observed between the corrected fNIRS signals and those without applying the short channel regression.

Regarding the systemic physiology signals, EDA (left), in the HR band, and HR, in the LF1 band, showed a statistically higher coherence in the eye-contact condition compared with the other one ($p=0.004$ and $p=0.011$, respectively), while pairs exhibited a higher correlation in Temp (left) and Temp (right) signals when they were not looking at each other (LF1 band, $p=0.010$ and $p=0.041$, respectively) as listed in Table S2 in the Supplemental Materials and depicted in Fig. 9.

It is important to note that the outliers in the violin plots shown in Figs. 9Fig. 10–11 do not impact the results since ART ANOVA is based on ranks and therefore not affected by extreme values.

Fig. 10

Violin plots of the WTC values distributions over the two groups (familiar and unfamiliar pair) grouped into fNIRS and systemic physiology signals and separated into (a) LF1 and VLF bands and (b) LF2 and HR bands. In the above figures are reported only the significant differences as a result of the aligned ranks transformation ANOVA. Double magenta asterisks refer to statistically significant differences between the two conditions even after FDR correction.

Fig. 11

Violin plots of the WTC values distributions over the two groups (familiar and unfamiliar pair) grouped into fNIRS and systemic physiology signals and separated into (a) LF1 and VLF bands and (b) LF2 and HR bands. In the above figures are reported only the significant differences as a result of the aligned ranks transformation ANOVA.

3.1.2.

Impact of the type of relationship

The pairs who took part in our study belonged to two different groups according to their type of relationship: the first group included siblings and couples (spouses and boy/girlfriends) and the second pairs did not know each other or to a lesser extent than the other group.

After performing the FDR correction, our results showed that the coupling in the familiar pairs was higher than in the other group in the LF1 band for $[{\mathrm{O}}_2\mathrm{Hb}]$ (right, raw) ($p<0.001$), [tHb] (right, raw) ($p=0.004$), HR ($p=0.004$), and in VLF band for [tHb] (left) ($p<0.001$). On the contrary, in the case of ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ in the LF1 band, the correlation was found to be higher in unfamiliar pairs ($p=0.004$). These results are reported in Tables S1 and S2 in the Supplemental Materials and are visualized in Fig. 10.

Without FDR correction, other significant differences in the coherence were noted. In all significant fNIRS results, the familiar pairs exhibited a higher hyperconnectivity. These fNIRS signals are $[{\mathrm{O}}_2\mathrm{Hb}]$ (right), [tHb] (right), and [tHb] (left + right) in the VLF band ($p=0.014$, $p=0.021$, and $p=0.006$, respectively), $[{\mathrm{O}}_2\mathrm{Hb}]$ (left + right, raw) and [tHb] (left + right) in the LF1 band ($p=0.016$ and $p=0.014$, respectively), and $[{\mathrm{O}}_2\mathrm{Hb}]$ (right, raw) in the LF2 band ($p=0.006$). The same effect (coherence familiar pairs > coherence unfamiliar pairs) was evident in DBP and HR (VLF band, $p=0.014$ and $p=0.008$, respectively) and in Temp (left) (HR band, $p=0.015$). The synchrony between the unfamiliar pairs was higher than the other one for ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ (LF2 band, $p=0.019$; VLF band, $p=0.022$), EDA (right) (HR band, $p=0.045$), and SBP (HR band, $p=0.026$).

3.1.3.

Relation between the experimental conditions and the two groups

The statistically significant results without applying the FDR correction considering the two factors interaction (condition and group) is visualized in Fig. 11. In all the signals, the familiar pair group always shows a higher coherence during the eye contact compared with the eyes-closed condition coherence. In addition, the difference between eye-contact and eyes-closed conditions in the familiar pairs is statistically higher than the difference between the eye-contact and eyes-closed conditions in the unfamiliar pairs. No significant effect was found after performing the FDR correction. The posthoc statistics of the significant results are available in Table S3 in the Supplemental Materials.

3.1.4.

Control analysis

The fNIRS results of the control analysis are summarized in Table S4 in the Supplemental Materials, while the systemic physiology results are listed in Table S5 in the Supplemental Materials. After performing the FDR correction, our results revealed that the coherence in the real pairs were statistically higher than the spurious coherence during the eye-contact condition in many fNIRS signals: [HHb] (left), $[{\mathrm{O}}_2\mathrm{Hb}]$ (right), $[{\mathrm{O}}_2\mathrm{Hb}]$ (left + right), [tHb] (right), and [tHb] (left + right) in the LF1, LF2, and VLF band ($p<0.001$); [HHb] (left, raw) in the LF1 band ($p<0.001$), in the LF2 band ($p=0.016$), in the VLF band ($p=0.013$); [HHb] (right) in LF1 and VLF ($p=0.001$); [HHb] (left + right) in LF1 and VLF ($p<0.001$), in LF2 ($p=0.009$); [HHb] (left + right, raw) in VLF ($p<0.001$); $[{\mathrm{O}}_2\mathrm{Hb}]$ (left) in LF1 and VLF ($p<0.001$), in LF2 ($p=0.013$); [tHb] (left) in LF1 ($p<0.001$), in LF2 ($p=0.001$) and in VLF ($p=0.002$). The same effect (coherence in the real pairs > spurious coherence) was found during the eyes-closed condition: [HHb] (left) and [HHb] (left + right) in LF1 band ($p<0.001$); [HHb] (left + right) in ($p=0.015$); $[{\mathrm{O}}_2\mathrm{Hb}]$ (left + right) in LF2 band ($p=0.004$). The results confirm that coherence values in real pairs are statistically higher than those of random pairs in many fNIRS signals.

Regarding systemic physiology signals, there was a significant effect (real coherence > spurious coherence) in HR and EDA (left and right) but only before performing FDR correction.

3.2.1.

Correlations between fNIRS and systemic physiological signals

The median correlation matrices relative to the eyes-closed and eyes-open conditions are shown in Fig. 4 (for the VLF and LF1 frequency bands) and Fig. 5 (for the LF2 and HR frequency bands). From the figures it is obvious that a strong correlation exists between all signals in the VLF band, and a higher correlation between the fNIRS signals in the other three frequency bands compared with the systemic physiology variables and fNIRS-systemic physiology variables.

The significant results obtained after performing the Kolmogorov–Smirnov test are grouped into three categories (fNIRS, systemic physiology and fNIRS/systemic physiology) and are illustrated in Fig. 6 (VLF band), Fig. 7 (LF1 band), and Fig. 8 (LF2 and HR bands). Statistically significant results after performing an FDR correction were marked with an asterisk.

The probability density functions of the distributions relative to the eye-contact condition are shown in red and those relative to the eyes-closed condition in blue.

The analysis framework enables us to investigate the simultaneous correlation of two time-dependent coherence vectors [${\mathrm{WTCS}}_1(t)$ and ${\mathrm{WTCS}}_2(t)$, where $S_1$ and $S_2$ represent two biosignals] between two individuals in each frequency band. A high correlation between the two coherence vectors can be indicated either as a positive time correlation between $S_1$ and $S_2$ time series or as a negative one. This method provides a synchrony indication between two pairs of signals. The correlated-coherence coupling analysis was conducted in MATLAB^®.

The correlation between fNIRS and systemic physiological time series was also assessed with respect to the familiar and familiar pairs. The median correlation matrices relative to the pair groups are shown in Fig. S1 (for the VLF and LF1 frequency bands) and Fig. S2 (for the LF2 and HR frequency bands) in the Supplemental Materials. No statistically significant difference was found between the two groups after performing FDR correction.

4.1.1.

Differences between eyes closed and open

Several studies have explored the importance of visual contact in social cognition so far, e.g., the importance of social attention.^67–69 Within this context, a number of studies were performed using static images or schematic stimuli.^70–72 Recent studies focused on face-to-face gaze rather than traditional stimulations by photo/video display.

Our study results show that pairs exhibit statistically higher coherence in physiological and fNIRS signals during face-to-face eye gaze than when their eyes were closed. Although external factors such as light coming into participants’ eyes during the eye-contact phase might cause an effect on the brain and body activity, coupling between the subjects occurred at different times when they made eye contact (as also shown in Fig. 2). Therefore, we believe that the main effect is due to the visual interaction of the subjects. A strong hyperconnectivity in [tHb] in the LF1 band of the whole prefrontal cortex was found. Since for this signal the coherence in true pairs (during the eye contact) was found to be statistically higher than the coherence in random pairs, we can conclude that it is associated with social interaction. Moreover, interbody synchrony of the HR during eye contact was observed. In the VLF band, a significant and strong coherence was observed during visual communication in the skin temperature of the left wrist and the HR. In line with our results, such synchronization in HR has already been found between mother and child during face-to-face interaction.⁷³ However, since interbody synchrony in true pairs is not statistically higher than the random pairs (but only before performing the FDR correction), we cannot conclude that this effect is caused by eye contact.

Furthermore, a significant coherence during the phase of eye contact between the subjects was found in the skin temperature from both the wrists (in the LF1 band) and in EDA measured on the left wrist (in the HR band) that was not significant after the FDR correction. Concerning EDA, a similar result has been found by Jarick and Bencic⁷⁴ where dyads showed an increase in skin conductance while making eye contact compared with nondirect eye contact. To the best of our knowledge, skin temperature is a parameter has never been investigated in a hyperscanning study before.

In addition, pairwise synchrony in the eye-contact condition compared with the eyes-closed condition was also significantly greater but only without FDR correction in $[{\mathrm{O}}_2\mathrm{Hb}]$ throughout the whole prefrontal cortex (in the VLF band) and in [tHb] in the right prefrontal cortex (in the LF1 band). In line with our results, other studies involving the brain activity of a single subject demonstrated a difference between resting-state brain activity and brain activity when the subject was observing.^75–89

A few fNIRS hyperscanning studies were conducted so far that investigated the difference in functional hyperconnectivity between the eyes open and eyes closed condition. Osaka et al.⁹⁰ demonstrated a higher brain-to-brain coherence during a cooperative humming task during eye contact compare to no eye contact. Hirsch et al.⁹¹ found in the left frontal region a higher hyperconnectivity during the eye-to-eye contact compared with the eye-to-picture gaze condition.

Consistently with previous studies, the results of our study indicate that visual contact is a salient component and plays a key role in social interaction during which high hyperconnectivity was observed.

4.1.2.

Relation between the experimental conditions and the two groups

The familiar group exhibited a stronger body-to-body coupling with regard to HR (in the LF1 band) and brain-to-brain coupling in terms of [tHb] of the right prefrontal region (in the VLF band) and the raw $[{\mathrm{O}}_2\mathrm{Hb}]$ of the right prefrontal region (in the LF2 band) compared with the unfamiliar pairs. On the other hand, the unfamiliar pairs showed a stronger body-to-body coupling in terms of ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ (in the LF1 band) compared with the familiar pairs. These findings are unique.

A higher body-to-body coupling trend in the familiar group also emerged between systemic physiology variables such as EDA and skin temperature but only without FDR correction.

Slovák et al.⁹² explored the connection between EDA and emotional engagement during a conversation task. According to their findings, the degree of synchrony between pairs depends on the strength of the empathic connection between the subjects.

The familiar group exhibited a stronger synchrony trend during eye contact than the other group in terms of skin temperature (in the HR band), $[{\mathrm{O}}_2\mathrm{Hb}]$, and [tHb]. The lack of a statistical significance may be due to the small sample size and large variability within each group.

4.2.1.

Higher coherence during eye contact between physiological data

We found a significantly higher coherence between skin temperature (from both wrists) and EDA in the VLF band when the subjects had eye contact compared to the period when they had their eyes closed even after the FDR correction. EDA changes depend on the quantity of sweat secreted by eccrine sweat glands controlled by the sympathetic nervous system (SNS). If the sympathetic branch of the autonomic nervous system (ANS) is highly activated, the sweat glands activity increases, leading to higher skin conductance, and vice versa. Previous studies have shown that eye contact leads to significantly higher skin conductance compared with nonmutual gaze.^93–97 The relationship between EDA, skin temperature, and mutual gaze has never before been investigated. Khan et al.⁹⁸ reported results similar to ours, noting a significant positive correlation between these two variables when students performed an engineering exam. The higher correlation of two pairs EDA time series during eye contact could be attributable to an increased emotional response either as a result of affection or due to a sense of unease. This increased the EDA and the skin conductance. On the other hand, this could trigger cutaneous vasoconstriction, which results in decreased skin blood flow inducing a drop in skin temperature. In fact, skin temperature is directly correlated with peripheral blood flow.⁹⁹ Based on this, recent studies have shown that direct eye contact between two people causes a stress reaction, which manifests itself as an increased SNS activity.¹⁰⁰ As a result, the high correlation during the eyes open condition between EDA and Temp coherence vectors may be attributable to a negative time correlation between the two signals of the individual pairs.

The same pattern was observed between the skin temperature of the left wrist and ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ in the VLF band as well as between the skin temperature of the right wrist and ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ in the LF1 band. An increase in ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ coherence between pairs may theoretically arise from three sources: simultaneous decrease (or increase) in alveolar ventilation, increase (or decrease) in ${\mathrm{CO}}_2$ production, and the quantity of ${\mathrm{CO}}_2$ that contributes to the respiratory capillary circulation.¹⁰¹ Respiratory reactions induced by eye contact produce changes in ventilation that are reflected by alterations in ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$. In fact, in most people, emotional arousal is linked to reduced ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ levels.¹⁰² This slight hypocapnia occurs when alveolar ventilation is greater than that required to eliminate metabolically formed ${\mathrm{CO}}_2$.¹⁰³ Since the arousal affects both parameters simultaneously, this leads to a high coherence.

Significant coherence was found in the VLF and LF1 band between the skin temperature (of both wrists) and cardiovascular variables such as HR, PP, SBP, SpO₂ and MAP. Such cardiovascular variations may be interpreted as an SNS response related to the pair’s interaction.¹⁰⁴

In addition, a significant symmetry in both wrists between EDAs as well as skin temperatures was found in the VLF band. In the LF1 band instead, coherence in ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ was found to be correlated to the coherence in PP (also after the FDR correction) and SBP.

In fact, during eye contact, the cardiovascular variables are affected by the ANS. Both branches of the ANS (parasympathetic nervous system and SNS) are of significant importance in controlling cardiovascular dynamics, including HR, MAP regulation, and respiration,^105,106 but they have contrary effects on the functions they regulate. In case of possible discomfort, due to the mutual gaze of the pairs, the SNS is more activated, causing an increase in MAP and, at the same time, in skin conductance.

Several studies have investigated and found an asymmetry between the right and left sides of the human body related to the measurement of skin temperature and EDA.^107–110 These results could explain the reason why some variables in our study are correlated only with EDA and/or skin temperatures on the right wrist and not on the left and vice versa. Besides, this relationship was found to be stronger on the right wrist in accordance with Demirel et al.¹⁰⁷ According to Demirel et al., such asymmetry may be associated with the peripheral immune asymmetry or asymmetric lymph node distribution.

In our results, skin temperature plays a predominant role. In fact, correlations with several cardiovascular and EDA parameters have been found, and most importantly, the correlation during eye contact has been found to be more significant than that period when the subject had their eyes closed.

4.2.2.

Higher coherence during eye contact between physiological and fNIRS data

Several significant temporal relationships between fNIRS and systemic physiological signals were identified, mainly in the LF1 (see Fig. 7). This is an indicator that the fNIRS signal was influenced by changes in systemic physiology.^{42,46,111,112} The fNIRS signals measured therefore represent a weighted sum of physiological changes due to neurovascular coupling as well as due to different physiological causes (e.g. changes in systemic physiology).

The strong coherence in all frequency bands between fNIRS signals and EDA is the most prominent finding. This is in line with previous publications that focused on investigating a relationship between brain hemodynamics and oxygenation with EDA on the intrapersonal level.^62–65 This relationship was found to be dependent on the type of task the participants were involved, as discussed by Holper et al.⁶⁶

Another remarkable finding is the higher coherence in the signal phase during eye contact between skin temperature and fNIRS signals, especially in the LF1 frequency band. Concerning the skin temperature, we found an interesting pattern: While the correlation between physiological signals and skin temperature was higher in the right arm in the LF1 and VLF bands, the correlation between fNIRS signals and skin temperature in the LF1 band was predominant in the left wrist. This is a unique finding in this study.

In addition, a significant correlation between the fNIRS variable and the ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ was obtained, only for the right prefrontal region in the LF1 and LF2 bands.

Our results suggest the presence of interconnection between cerebral hemodynamic variables and SNS activity. Although such relationships have been reported previously, we are the first to identify them in investigating the long-term effects of eye contact in the context of SPA-hyperscanning.

4.2.3.

Higher coherence during eye contact between fNIRS data

The strongest correlation during eye contact of fNIRS occurred in the LF1 and LF2 bands. No evident pattern has arisen apart from a greater symmetry between the left and right region in $[{\mathrm{O}}_2\mathrm{Hb}]$ as well as in [tHb] in the LF1 band during the eye contact period compared with the period with eyes closed.

4.2.4.

Higher coherence during eyes-closed condition

A lower coherence was observed in some signals in all four frequency bands during the eyes’ contact than when eye contact was blocked. We have found no explanation for this phenomenon.