Synchronization of brain activity associated with eye contact: comparison of face-to-face and online communication

Abstract

Online meetings have become increasingly prevalent, especially during the coronavirus disease 2019 pandemic. Although they offer convenience and effectiveness in various contexts, there is a pertinent question about whether they truly replicate the richness of in-person communication. This study delves into the distinctions between online and face-to-face interactions, with a particular focus on the synchronization of brain activity. Previous research has indicated a connection between synchronization and the quality of communication. Therefore, our hypothesis posits that face-to-face interactions lead to greater brain synchronization compared to online interactions, which often lack certain social cues. To investigate this, we conducted a study using functional near-infrared spectroscopy hyperscanning during an eye-contact task involving 28 male participants organized into 14 pairs. We assessed brain signal synchronization using wavelet coherence analysis. After comparing face-to-face and online conditions, our findings revealed significantly higher synchronization in face-to-face scenarios, particularly within the right temporoparietal region. These results align with the outcomes of other hyperscanning studies and suggest that face-to-face communication elicits a higher level of brain synchronization compared with online communication. In the future, this approach holds promise for evaluating the effectiveness of online meeting tools in achieving a more authentic virtual communication experience.

Introduction

The rapid surge in the utilization of online meetings, exemplified by platforms like Zoom, in the wake of the coronavirus disease 2019 pandemic has revolutionized the landscapes of education and the workplace^1,2. In education, students now partake in virtual classes through online platforms, offering increased flexibility and seamless information exchange. Similarly, businesses employ online communication tools to facilitate remote work and virtual meetings, ensuring teams remain connected and productive regardless of geographical constraints. Consequently, online meetings have ushered in a new era of flexibility in terms of location and scheduling, with discussions and decision-making conducted through these platforms garnering global attention³.

Despite the advantages of online tools in supporting and enhancing communication, they also present unique challenges. Screen-based interactions are perceived to potentially impede the full expression and interpretation of nonverbal cues, a critical component of smooth face-to-face communication^3,4. Nonverbal cues, encompassing facial expressions, gazes, and gestures, contribute additional layers of meaning and context to verbal communication. Among these cues, imitation behavior and eye contact hold particular significance in fostering empathy⁵. Imitation has been shown to enhance interpersonal likability and promote social behavior^6–8, and studies comparing nonverbal behaviors with and without eye contact have demonstrated that eye contact increases empathic recognition⁹. Given these considerations, the extent to which conventional online communication can replicate the empathic communication of face-to-face interactions remains unclear³.

Given the current research, revealing the differences between face-to-face and online communication from more biological perspectives can be of increasing importance. If discomfort or even unconscious feelings regarding online communication were represented in measurable, physiological signals, these parameters could provide valuable insights regarding the limitations of currently used online communication tools. This knowledge also might result in new evaluation methods for online interactions, potentially driving the development of novel online technologies that closely emulate natural face-to-face communication.

In this pursuit, the present study focuses on the synchronization of brain activity as a means to comprehend the physiological underpinnings of qualitative differences in communication. Departing from conventional experimental paradigms that examine the brain activity of individual participants during social interactions, hyperscanning facilitates the exploration of neural interconnections between multiple participants engaged in interaction¹⁰. The concept of hyperscanning, a technique that simultaneously records the brain activity of two or more individuals engaged in information exchange, emerged in the mid-1960s, initially employing electroencephalography to study synchronization¹¹. Recent years have witnessed the use of functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS), both capable of measuring physiological signals influenced by brain hemodynamics, contributing to our understanding of the physiological mechanisms underlying human communication^12,13. For example, an fMRI study has suggested that brain activity synchronizes when individuals share intentions through eye contact¹⁴. Other research has indicated that the synchrony of brain activity during eye contact is lower in individuals with autism spectrum disorder (ASD), who often struggle with making eye contact, than in healthy individuals¹⁵. Additionally, studies have shown brain activity and blinking synchronization between pairs of individuals engaged in a joint attention task, wherein visual attention was shared through eye gaze¹⁶. Similarly, fNIRS studies conducted in natural settings have demonstrated synchronization facilitated by eye contact^17,18. These studies have consistently highlighted that synchronization is more pronounced during in-person, face-to-face interactions than in conditions involving non-face-to-face interactions, such as maintaining eye contact with a static partner or a pre-recorded video partner. Furthermore, research comparing eye contact synchronization between individuals with ASD and healthy participants has reiterated these findings, with synchronization being higher among healthy participants, both in fMRI and fNIRS studies¹⁹.

Building upon this body of research, our study employs brain activity synchrony associated with eye contact as a metric for communication. Eye contact represents the most fundamental form of nonverbal information in communication and is widely recognized for its role in initiating and facilitating interpersonal connections^20–22. Brain synchronization facilitated by eye contact also plays a crucial role in early human development²³ and can unveil the nature of relationships between individuals²⁴. Conventional fNIRS studies have compared recorded and live conditions and have suggested that the richer the nonverbal information, the greater the synchrony in the temporal regions of the brain, such as the superior marginal gyrus and angular gyrus at the temporoparietal junction (TPJ)^17–19. Moreover, a recent study demonstrated significant increases in cross-brain synchronization within the somatosensory association cortices for the real face-to-face condition (FC) compared to those in the cross-brain synchronization for the online condition (OC), suggesting that the detection of dynamic social cues such as facial micromovements is reduced in online communication²⁵.

This study primarily aims to replicate findings that showed that compared to online communication, face-to-face communication enhances cross-brain synchrony between individuals. Moreover, this study seeks to investigate in greater detail at the channel level the cortical regions that exhibit synchronization. Although previous fNIRS studies have focused on specific regions of interest to highlight differences in cross-brain synchrony between experimental conditions^17,19,25, the spatial characteristics of cortical regions with increased synchronization remain insufficiently explored. Therefore, this study aims to advance our understanding of cross-brain synchrony by providing a more detailed characterization of the spatial patterns of cortical regions where synchronization is enhanced during face-to-face communication compared to that during online communication.

Methods

Participants

Twenty-eight male participants, organized into 14 pairs, with a mean age of 20.00 ± 0.35 years, took part in this interactive hyperscanning experiment utilizing fNIRS. These individuals were university students belonging to the same sports team and were familiar with one another. Ethical approval for this study was granted by the Biotechnology Research Ethics Committee of the Shibaura Institute of Technology, and informed consent was obtained from all participants before the commencement of the experiment. The experimental protocol was performed in accordance with the “Guidelines for ethics-related problems with noninvasive research on human brain function” established by the Japan Neuroscience Society. Furthermore, the participants depicted in Figs. 1 and 2 provided consent for the publication of their pictures.

Fig. 1.

Experimental setup and protocol. (a) Experimental setup for the face-to-face condition (FC). Participants were seated face-to-face at a distance of 140 cm from their partners. The remaining target (yellow cross mark) was positioned 10° to the left of the partner’s gaze. Written permission has been obtained from the participants for the publication of their pictures. (b) Experimental setup for the online condition (OC). A partition was placed between the participants, and the monitor showing the partner’s face was positioned 70 cm from the participants. The size of the partner’s face, height of the gaze, and angle of the rest of the target were adjusted using a camera to match the visual angle in the FC. Written permission has been obtained from the participants for the publication of their pictures. (c) Experimental protocol. A single block consisting of “Task” and “Rest” was repeated six times. During the “Task” period, the participant alternately looked at their partner’s eyes and rest target by intervals of 3 s, whereas they looked at the rest target for 15 s during the “Rest” period. The task design was adapted from previous studies^17–19.

Fig. 2.

Probe positions during functional near-infrared spectroscopy (fNIRS) measurements. The estimated fNIRS measurement positions on the brain and the appearance of the probe cap worn by the author (R.S.) are shown at the top and bottom, respectively. The red and blue circles indicate the locations of the light source and detection probes, respectively, and the area between them represents the measurement point. The probabilistic registration method was used to estimate the locations of the measurement channels in the stereotaxic brain coordinate system or the Montreal Neurological Institute (MNI) space. This figure features a photo provided by and used with the permission of the author (R.S.).

Experimental procedure

Following the completion of a brief questionnaire to obtain basic information such as sex, age, and health status, the participants were seated in a chair, facing each other, within a sound-attenuated chamber. During this experiment, all pairs of participants engaged in the same eye-contact task used in previous studies^17–19,25 under two distinct conditions: direct FC and OC. Both conditions were conducted on the same day, with a minimum of 10 min between them, and the order was counterbalanced across pairs.

In the FC, participants performed the eye-contact task while seated at a distance of 140 cm from each other, with their gaze alternating between their partner’s eyes and a rest target (Fig. 1a). The rest target (a yellow cross) was positioned 10° within each participant’s peripheral vision, to the right of the inferred line of sight between them.

For the OC, we utilized the online meeting platform, Zoom (Zoom Video Communications, Inc.). Participants engaged in the same eye-contact task, with their partner’s image displayed on a 27-inch liquid crystal display monitor (GW2780, BenQ Japan Co., Ltd.; Fig. 1b). To prevent mutual visibility, a partition was placed between the participants, and each participant viewed their partner’s image on a display positioned 70 cm in front of them. Camera (C505, Logicool Co., Ltd.) adjustments were made to ensure that factors such as the partner’s facial size, gaze height, and rest target angle of the OC matched those of the FC.

Eye contact task

The experimental procedure consisted of alternating “Task” and “Rest” phases, forming a single block, repeated six times. Specifically, participants alternated their gaze between their partner’s eyes and a rest target at 3-s intervals during the Task phase. Following the Task period, they maintained their gaze on the rest target for 15 s during the Rest period. This 30-s sequence was repeated six times within one session (Fig. 1c). Participants were instructed to minimize head movement, avoid verbal communication, make eye contact with their partner (not the camera lens), and maintain facial expressions as neutral as possible. Prior to commencing the experiment, participants were directed to await an auditory cue signaling them to focus on the rest target. Auditory cues were incorporated into the Task sequence, and participants adjusted their gaze accordingly.

fNIRS data acquisition

We employed two fNIRS systems (ETG-4000, Hitachi Medical Corporation, Japan) to simultaneously measure hemodynamic signals in both participants. During the experiment, participants wore fNIRS probe caps on both sides of their temporal lobes. A 3 × 5 optode probe set (comprising eight emitters and seven detector probes with a 30 mm separation) was utilized, resulting in 22 measurement positions (channels) in each hemisphere, yielding a total of 44 channels for each participant (Fig. 2). The probe placement was determined to encompass the TPJ, a region implicated in social interactions, including joint actions and nonverbal communication^17–19. Following the international 10–20 system, channel 21 (ch21) was positioned at T4, and channel 42 (ch42) was positioned at T3.

A probabilistic registration method was employed to estimate the coordinates of the measurement channels in the Montreal Neurological Institute (MNI) space^26,27. To facilitate this estimation, data were collected on the three-dimensional (3D) coordinates of the 44 channel locations before the experiment. These data had been previously acquired from seven male volunteers (aged 21–24 years) using a 3D-magnetic space digitizer (3D probe positioning unit for NIRS, EZT-DM101, Hitachi Medical Corporation, Japan). These volunteers were not included again as participants in the current study, and we assumed that the mean position of each channel would closely align if the race, age, and sex of the participants in both groups were matched. Based on this assumption, potential spatial registration errors were estimated to be approximately 4.7–7.0 mm, as reported in the previous study²⁷. We calculated the mean MNI coordinates of the seven prior participants and the estimated anatomical regions (Table 1) for visualization and interpretation of the results.

Table 1.

Estimated location of each near-infrared spectroscopy channel on a normalized brain image.

Right hemisphere						Left hemisphere
ch	MNI coordinates			Anatomical region	%	ch	MNI coordinates			Anatomical region	%
	X	Y	Z				X	Y	Z
1	36	− 72	54	Parietal_Sup	53	23	− 51	− 1	55	Precentral	73
				Angular	38					Postcentral	24
				Parietal_Inf	6					Frontal_Mid	3
				Occipital_Sup	3	24	− 55	− 29	56	Postcentral	62
2	54	− 53	55	Parietal_Inf	87					Parietal_Inf	38
				Angular	11	25	− 51	− 55	56	Parietal_Inf	95
				Parietal_Sup	2					Angular	5
3	59	− 29	55	Parietal_Inf	35	26	− 36	− 74	53	Angular	36
				SupraMarginal	33					Parietal_Sup	32
				Postcentral	26					Parietal_Inf	32
				Parietal_Sup	6	27	− 54	13	40	Precentral	59
4	56	− 2	53	Precentral	57					Frontal_Mid	30
				Frontal_Mid	37					Frontal_Inf_Oper	9
				Postcentral	6					Frontal_Inf_Tri	2
5	33	− 85	41	Occipital_Sup	54	28	− 62	− 17	45	Postcentral	45
				Occipital_Mid	40					SupraMarginal	41
				Angular	4					Parietal_Inf	14
				Parietal_Sup	2	29	− 62	− 45	46	Parietal_Inf	75
6	52	− 68	46	Angular	93					SupraMarginal	25
				Parietal_Inf	7	30	− 51	− 70	45	Angular	88
7	64	− 44	46	SupraMarginal	53					Occipital_Mid	9
				Parietal_Inf	45					Parietal_Inf	3
				Angular	2	31	− 33	− 86	40	Occipital_Mid	73
8	65	− 18	45	SupraMarginal	51					Parietal_Inf	16
				Postcentral	49					Occipital_Sup	11
9	59	10	40	Precentral	77					Parietal_Sup	0
				Frontal_Inf_Oper	14	32	− 65	− 4	29	Postcentral	87
				Frontal_Mid	4					Precentral	13
				Postcentral	4	33	− 68	− 34	32	SupraMarginal	99
10	46	− 81	32	Occipital_Mid	81					Temporal_Sup	1
				Angular	19	34	− 61	− 61	32	Angular	78
11	61	− 59	34	Angular	81					SupraMarginal	17
				Parietal_Inf	18					Parietal_Inf	5
				Occipital_Mid	1	35	− 46	− 83	30	Occipital_Mid	81
12	70	− 34	34	SupraMarginal	100					Angular	19
13	68	− 5	31	Postcentral	92	36	− 62	10	11	Frontal_Inf_Oper	65
				Precentral	4					Rolandic_Oper	25
				SupraMarginal	4					Frontal_Inf_Tri	5
14	37	− 93	15	Occipital_Mid	92					Postcentral	4
				Occipital_Sup	8					Precentral	2
15	55	− 74	19	Temporal_Mid	63	37	− 68	− 21	14	Temporal_Sup	43
				Occipital_Mid	37					SupraMarginal	23
16	68	− 49	19	Temporal_Sup	55					Postcentral	22
				Temporal_Mid	30					Temporal_Mid	12
				SupraMarginal	9	38	− 67	− 51	17	Temporal_Sup	49
				Angular	5					Temporal_Mid	36
17	71	− 23	16	Temporal_Sup	66					SupraMarginal	15
				SupraMarginal	28					Angular	1
				Postcentral	3	39	− 54	− 76	16	Occipital_Mid	55
				Rolandic_Oper	2					Temporal_Mid	40
18	66	7	15	Rolandic_Oper	41					Angular	6
				Precentral	27	40	− 38	− 94	13	Occipital_Mid	100
				Postcentral	17	41	− 67	− 8	− 6	Temporal_Mid	78
				Frontal_Inf_Oper	13					Temporal_Sup	23
				Temporal_Sup	1	42	− 71	− 38	− 2	Temporal_Mid	100
19	48	− 86	3	Occipital_Mid	74	43	− 62	− 65	1	Temporal_Mid	73
				Occipital_Inf	26					Temporal_Inf	18
20	63	− 63	4	Temporal_Mid	85					Occipital_Mid	5
				Temporal_Inf	15					Occipital_Inf	5
21	73	− 38	2	Temporal_Mid	84	44	− 47	− 88	0	Occipital_Mid	90
				Temporal_Sup	16					Occipital_Inf	10
				Temporal_Inf	0
22	71	− 11	− 2	Temporal_Sup	75
				Temporal_Mid	25

Mean and standard deviation of Montreal Neurological Institute (MNI) coordinates across participants and corresponding anatomical regions with their probabilities (%) are shown for each channel (ch).

Data analysis

Preprocessing

The fNIRS data underwent preprocessing using the Platform for Optical Topography Analysis Tools, a MATLAB-based plug-in analysis software²⁸. Optical data from two different wavelengths (approximately 690 and 830 nm) were employed for each NIRS channel to generate time series data for oxygenated hemoglobin (oxy-Hb) and deoxygenated hemoglobin (deoxy-Hb) signals, based on the modified Beer–Lambert law^29,30. These signals were represented as mM·mm, reflecting the product of variations in hemoglobin concentration (in mM) and changes in optical path length (in mm) within the active region (effective optical path length).

Raw fNIRS signals often contain artifacts arising from body movement and skin blood flow^31–33, which may differ from actual hemodynamics in the cerebral cortex³³. To address this, we applied a signal separation method to isolate functional hemodynamics in the cerebral cortex³⁴. This method is rooted in the theory positing a negative linear relationship between oxy- and deoxy-Hb signals, ensuring that both Hb signals yield identical statistical outcomes. Consequently, only the oxy-Hb signal was employed for subsequent analyses in this study.

Wavelet coherence

To evaluate the synchrony of brain activity over time between participants in a pair during the experimental task, we employed wavelet transform coherence analysis which allows for the simultaneous analysis of time and frequency components and the evaluation of correlations at localized time intervals^{17,18,25,35–37}. The Wavelet Toolbox in MATLAB 2021b was used to calculate the coherence of the participants in a pair for each channel; coherence per unit time was computed for each trial in the frequency range of 0.025–0.4 Hz and an average was obtained across six trials per unit time (30 s). Subsequently, the one-trial coherence was divided into six 5-s periods labeled as P1, P2, P3, P4, P5, and P6. In the subsequent statistical analysis, coherence between the FC and OC was compared using two-tailed t-tests. After detecting the periods and channels with significant differences between the FC and OC (p < 0.05, uncorrected), we used a permutation method to confirm the significance level for the correction of multiple comparisons. In this process, we performed the same analysis for pairs randomly selected in different combinations from the actual pairs (shuffled pairs) 1000 times. Then, the t-values of the top 5% of 1000 shuffled pairs were used as the corrected threshold of significance (p < 0.05, corrected).

Results

The mean temporal coherence changes of all pairs are shown for individual channels in Fig. 3a. The red and blue lines represent the FC and OC, respectively. A noticeable trend emerged, with brain signal synchronization appearing higher in the FC than in the OC across several channels. To assess these differences, two-tailed t-tests were conducted for each channel at 5-s intervals within the 30-s timeframe, with significance criteria set to include at least one period with p < 0.05 (uncorrected) among the six periods. This analysis identified significant differences, where the FC exhibited higher synchronization than the OC, in five right hemispheric channels and one left hemispheric channel (Fig. 3b). The number of significant periods is color-coded, with yellow representing one significant period, orange indicating two, and red signifying three. Two channels, ch7 and ch12, displayed significance across three successive periods. Based on spatial registration analysis, ch7 was primarily located in the right supramarginal gyrus (including the inferior parietal lobule and angular gyrus), whereas ch12 was primarily situated in the right supramarginal gyrus. ch1, primarily in the right superior parietal lobule (encompassing the angular gyrus and inferior parietal lobule), and ch25, primarily in the left inferior parietal lobule (also including the angular gyrus), exhibited significance across two different periods. Lastly, ch2 (right inferior parietal lobule, including the angular gyrus and superior parietal lobule) and ch11 (right angular gyrus, including the inferior parietal lobule) each displayed significance for one period. Notably, the t-test indicated no significant difference in the case of OC > FC comparison (Fig. 3c).

Fig. 3.

Differences of wavelet coherence between the face-to-face condition (FC) and online condition (OC) (a) Comparison of temporal changes in coherence for all channels in both conditions. The mean time courses of coherence across participants are plotted for both the FC (red) and OC (blue). The shaded parts represent standard errors for each condition. Significant periods (p < 0.05, uncorrected) were identified by two-tailed t-tests performed at 5-s intervals over 30 s, and the number of significant periods is indicated in yellow for 1, orange for 2, and red for 3. (b) Channel map of higher coherence in the FC than in the OC. Similar to (a), the number of significant periods determined by two-tailed t-tests (p < 0.05, uncorrected) is indicated in black for 0, yellow for 1, orange for 2, and red for 3. Significant differences were found in five channels of the right hemisphere and in one channel of the left hemisphere. (c) Channel map of higher coherence in the OC than in the FC. Similar to (b), the number of significant periods was determined using two-tailed t-tests (p < 0.05) in the case of higher coherence in the OC than in the FC. No significant differences were found for any channel.

For channels that exhibited significantly higher coherence in the FC (ch1, ch2, ch7, ch11, and ch12 in the right hemisphere and ch25 in the left hemisphere), detailed graphs with enlarged information are presented in Fig. 4a,b). An asterisk denotes significant differences between the FC and OC during the respective periods. In ch1, significant differences between the FC and OC were observed in P4 (T = 2.40, p < 0.05) and P5 (T = 2.17, p < 0.05). ch2 exhibited significant differences between the FC and OC in P4 (T = 2.71, p < 0.05). The ch7 demonstrated significant differences between the FC and OC in P4 (T = 2.23, p < 0.05), P5 (T = 2.51, p < 0.05), and P6 (T = 2.35, p < 0.05). ch11 displayed significant differences between the FC and OC in P4 (T = 2.19, p < 0.05). In ch12, significant differences between the FC and OC were found in P4 (T = 2.44, p < 0.05), P5 (T = 3.01, p < 0.01), and P6 (T = 2.32, p < 0.05). ch25 exhibited significant differences between the FC and OC in P6 (T = 2.65, p < 0.05). To confirm the significance level by correcting for multiple comparisons, the actual t-values were compared with distributions of 1000 pseudo t-values derived from shuffled pairs (Fig. 4c). The bar graphs depict the mean t-values across these 1000 shuffled pairs, with error bars indicating the standard deviation. All actual t-values denoted by circles (○) were higher than the t-values of the top 5% of the 1000 shuffled pairs, denoted by diamonds (◊).

Fig. 4.

Detailed results of channels showing higher coherence in the face-to-face condition (FC). (a) Temporal changes in coherence for channels with significant differences (ch1, ch2, ch7, ch11, and ch12) in the right hemisphere (b) Temporal changes in coherence for channels with significant differences (ch25) in the left hemisphere For both (a,b), the lighter shades represent the standard error for each condition. The asterisk indicates a significant difference between the FC and the online condition (OC) in the time period. (c) Comparison of t-values between the coherence in actual partners and that in shuffled pairs for the significant periods of the significant channels. The simulated t-values were obtained from the t-test in 1000 shuffled pairs, and the mean values are shown in the bar graph with error bars of standard deviations for each. The diamond symbols (◊) denote the top 5% of the distribution obtained from the shuffled pairs, and circle symbols (○) represent the real t-values obtained from the actual partners.

Discussion

This study examined significant differences between online and face-to-face communication, focusing on cross-brain synchronization, and found that face-to-face interactions (FC) led to higher coherence compared to online interactions (OC). This primary finding replicates basic findings of previous research^17,18,25,38 and reinforces the differences between face-to-face and online communication, as discussed below.

The primary novel finding of this study is the identification of enhanced coherence in the right TPJ during face-to-face communication. This finding extends our understanding beyond the most relevant previous study, which compared face-to-face and online communication conditions²⁵. The previous study reported increased coherence within the somatosensory association cortex during face-to-face communication compared to that during online communication but did not clearly distinguish between the left and right hemispheres. Our findings align with the results of earlier studies highlighting right hemispheric dominance in brain synchronization during eye contact tasks^16,17. This right hemisphere predominance may be attributed to the general association of social interactions with this hemisphere. In contrast to the left hemisphere’s predominant role in language processing^39,40, the right hemisphere primarily handles nonverbal information^41,42, encompassing the interpretation of emotions through facial expressions^43,44 and gesture production⁴⁵.

The regions exhibiting significant coherence differences were situated in the TPJ, comprising the angular gyrus, supramarginal gyrus, superior parietal lobule, and inferior parietal lobule. Notably, the prominent coherence difference between the FC and OC was detected in the TPJ, in line with prior research findings. The right TPJ has earned recognition as a “social brain region,” as it has been implicated in numerous neuroimaging studies focusing on social cognitive functions like theory of mind, empathy, and social attention⁴⁶. The act of deducing intentions and emotions from nonverbal cues is referred to as mentalizing^47,48. The TPJ exhibits increased activation during tasks explicitly requiring mentalizing, such as adopting another perspective and comprehending intentions^49,50. Moreover, the TPJ plays a pivotal role in interpreting where another individual is directing their gaze through observations of eye movements⁵¹. Thus, the TPJ plays an important role in interpersonal interactions⁵², which is supported by our findings. Furthermore, ch12, which exhibited coherence differences between the FC and OC across multiple periods, was located in close proximity to the supramarginal gyrus. The right supramarginal gyrus is associated with the theory of mind and empathy^53,54 and may play a significant role in social communication, including eye contact.

Several previous studies have reported heightened cross-brain coherence linked to eye contact^{14–19,24,25,38,55}. Coherence was reported to increase with a greater number of eye contacts⁵⁵ and decrease when fewer eye contacts were established, as observed in pairs involving individuals with ASD¹⁹. Given that the number of eye contacts was controlled to be consistent between both conditions in this study, factors contributing to the coherence difference warrant discussion. Previous fNIRS studies reported that synchronization is more pronounced in actual FCs than in conditions involving eye contact with a static picture¹⁸ or a pre-recorded video¹⁷ of the partner. Considering these prior findings, the distinction between 3D and two-dimensional images⁵⁶ could have contributed to the higher coherence in the FC compared to that in the OC in our study. Moreover, the most relevant previous study²⁵ compared face-to-face and online communication conditions using a multi-modal approach and discussed potential factors contributing to the increased coherence observed during face-to-face communication. The authors suggested that facial micromovements are more difficult to detect in an online environment, as evidenced by shorter average gaze dwell times on the partner’s face. They also highlighted differences in the line of sight to the partner’s eyes in the virtual environment, which may weaken the activation associated with interactive and social processing. Similarly, we propose that participants’ tendency to focus on their partner’s eyes rather than on the camera during online communication is a critical factor, potentially leading to the absence of genuine eye contact that is typically observed in face-to-face communication. This divergence in gaze could diminish both the quantity and quality of eye contact, thereby affecting coherence^15,55. Additionally, this phenomenon might be further exacerbated by delays inherent in online interactions, which can disrupt real-time information exchange and hinder the natural flow of communication³⁸.

Regarding the time domain of coherence, we analyzed the frequency range of 0.025–0.4 Hz based on previous studies^17,18. Consequently, we consider that our results also primarily reflect components corresponding to periods between 15–25 s¹⁷ or 10–20 s¹⁸. These periods are slightly shorter than the Task period of 30 s, suggesting that the synchronizing components are involved in “sending, receiving, and interpreting rapidly streaming, socially relevant visual cues”¹⁸. Another possible interpretation was noted for the cross-brain coherence; the brain signals might synchronize because the two individuals processed the same information on the same time scale¹⁸. However, this interpretation is probably not supported for the following two reasons. First, similar coherences were not observed in shuffled pairs. If the coherence were simply due to the activation waveform caused by the same task using the same timing, the brain signals would synchronize more or less with any other participant. Second, the reported synchronized regions were not always consistent with the activation areas for the eye-contact task. For example, we found significant differences in task-related activation in two channels, ch11 in the right angular gyrus and ch27 in the left precentral area (see “Supplementary Material”). Although ch11 did overlap with a region where a higher coherence was found in the FC, ch27 in the left hemisphere did not. We also found no activation in the right supramarginal gyrus (ch7 and ch12) which was the center of coherence in our study. Furthermore, regions exhibiting significant differences in synchrony did not necessarily correspond to those with significant differences in activation, as observed in a previous study²⁵, suggesting that these may represent independent phenomena. Thus, we support the interpretation that the observed cross-brain synchronization reflects a mechanism for high-level perception of rapidly streaming and meaningful social cues. Focusing on the temporal evolution of coherence in significant channels, we also observed periods with significant differences in the coherences emerging after the Task phase. This delay can be partially attributed to the inherent lag of hemodynamic responses relative to neural activity. However, further investigation is needed to elucidate the detailed time course of coherence more comprehensively.

This study has a few limitations. First, the participants were exclusively male members of the same baseball team. This design differs from that of the previous study²⁵, which included a diverse group of participants comprising both men and women, as well as strangers and acquaintances. Although it remains unclear whether the social relationship within a pair affects cross-brain synchronization, the observed outcomes in this study may have been significantly influenced by the pre-existing close relationships among team members who regularly engage in the sports activity together^57,58. Second, there is room for signal quality improvement in the fNIRS data. Although we employed a signal separation analysis to eliminate systemic effects, including skin blood flow³⁴, future research may benefit from using measurement equipment with shorter-distance measurements for more reliable results⁵⁹. Third, the spatial registration has two potential sources of error. One is the manual error which may have occurred during the positioning of the probe holder. Our prior evaluations of manual errors estimated an average error of approximately 4.3 ± 3.0 mm⁶⁰. The other is any variation in head size, which might lead to larger errors in channels located farther from the T3 and T4 landmarks used for positioning. However, even with head size variations within the 55–60 cm range, theoretical displacement at the farthest channels is less than 8 mm. Consequently, the maximum combined spatial error is approximately 20 mm²⁷, which is still below the spatial resolution of the fNIRS system. This suggests that spatial errors are likely negligible in our study, especially since five consecutive channels exhibited significantly higher coherence during the FC compared to that during the OC. This finding indicates that coherence was distributed across a region larger than the area covered by a single channel.

In conclusion, our investigation into brain activity synchronization during eye contact tasks in the FC and OC revealed higher coherence in the FC, specifically within the right TPJ. These findings, obtained from a sample of individuals belonging to the same baseball team, may reflect real-world communication dynamics and provide valuable insights into the distinctiveness of face-to-face communication, which cannot be fully elucidated through fMRI studies. Additionally, this study underscores the utility of brain synchronization analysis for assessing the quality of online communication tools and contributes to the development of innovative communication technologies.

Author contributions

R.S. and H.S. designed the experiment, R.S. conducted the experiment, R. S. analyzed the data, and R.S. and H.S. wrote the paper. All authors reviewed the manuscript.

Data availability

The datasets generated during the current study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors have no relevant financial interests in this article and no potential conflicts of interest to disclose.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-84602-x.