Social synchronization of brain activity increases during eye-contact

Eye contact is a key social cue processed by regions in the social brain (e.g., medial prefrontal cortex, superior temporal sulcus), yet most prior work examines single brains. Hyperscanning studies indicate increased inter-brain synchrony during interaction and potential directionality in leader–follower contexts. This study asks: RQ1) How does eye contact affect inter- and intra-brain synchronization (frequency, magnitude, direction)? RQ2) What are the hyperbrain network characteristics during eye contact? RQ3) What is their functional relevance (friends vs. strangers)? RQ4) Is synchronization directed according to spontaneously emerging leader–follower roles? The authors hypothesized that eye contact would elicit greater inter- than intra-brain synchronization, that friends would show stronger inter-brain connectivity than strangers, and that directionality would flow from leader to follower.

Prior hyperscanning EEG/fNIRS studies show increased inter-brain synchrony during social interactions, including during live eye contact, often in alpha, beta, and gamma bands. Directed inter-brain effects have been linked to leader–follower dynamics in joint action, music, and parent–infant interactions. Computational models suggest leader–follower success emerges from increased inter-unit and decreased intra-unit coupling. Graph-theoretical analyses of hyperbrain networks reveal differences under cooperative vs uncooperative contexts (e.g., reduced inter-brain links and higher modularity for uncooperative dyads). However, studies rarely integrate intra- and inter-brain dynamics, isolate eye contact from other behaviors, or assess directionality and network topology as a function of social relationship or emergent roles.

Participants: 56 adult dyads (112 participants), 27 pairs of friends and 29 pairs of strangers (mean ages ~20.3–20.5 years), neurologically healthy. Ethics approvals obtained; participants compensated. Design and tasks: Hyperscanning EEG and dual eye-tracking while seated face-to-face using synchronized desktops. Eye-contact task: participants reproduced the duration of an auditory tone (1.5 s or 2.5 s) by maintaining mutual eye contact for the estimated duration after the tone; between-tone phases involved fixating a chinrest sticker. Control task: no mutual eye contact—partners alternated looking at eyes vs chinrest so that they never made eye contact while reproducing the duration. The analysis window focused on the reproduction period when both participants performed the estimation. Behavioral measures: Time reproduction durations were recorded to verify task engagement and partner influence. Leadership index computed per dyad based on which participant consistently broke eye contact first; median split created strong vs. weak leadership groups. EEG acquisition and preprocessing: Two Starstim 20 devices, 18 electrodes (10–20 layout: Fp1/2, Fz, Cz, Pz, Oz, F3/4/7/8, C3/4, P3/4/7/8, T7/8), re-referenced to linked earlobes, band-pass 5–45 Hz. Poor channels interpolated. Epoching aligned to onset of the common reproduction period; trials shorter than window excluded. Eye blinks identified via Fp1/Fp2 ±70 μV with ±40 ms exclusion. Eye tracking at 60 Hz used for event markers. One pair excluded for technical issues; five pairs for insufficient valid trials (final N=100 participants; 50 dyads) for EEG. Connectivity measures: Undirected phase synchronization via corrected imaginary PLV (ciPLV), robust to volume conduction; directed synchronization via phase slope index (PSI), capturing non-instantaneous, directed interactions. Frequencies analyzed: theta (4–8 Hz), alpha (8–12 Hz), lower beta (13–20 Hz), upper beta (21–30 Hz), gamma (30–45 Hz). Time window 0.5–2.0 s post reproduction onset to minimize eye-movement contamination. Statistics for synchronization: Nonparametric cluster permutation tests (network-based statistics) compared eye-contact vs control for inter- and intra-brain connections across frequency bands. Clusters formed from suprathreshold edges (|t|>2) into strongly connected components; 5000 permutations, two-tailed α=0.05, retaining max cluster t-statistic per permutation. Control analyses: Shuffled inter-dyad pairing to assess common input confounds for inter-brain effects (1000 shuffled datasets). Gamma power (absolute/relative) compared between conditions to rule out power differences. Network analysis: For each dyad, z-scored connectivity matrices (eye-contact vs control) computed separately for inter- and intra-brain edges, thresholded at Z≥1 to construct unweighted and weighted graphs. Global measures: network strength (mean z), density, global/local efficiency, modularity, assortativity, and rich-club coefficient (normalized against randomized graphs). ROI-level analyses grouped electrodes into 8 ROIs: RF (F4, F8, Fp2), RP (P8, P4), LF (Fp1, F7, F3), LP (P3, P7), RCT (C4, T8), LCT (C3, T7), MF (Fz, Cz), MP (Pz, Oz). Directed network analyses paralleled undirected analyses for PSI, considering edge direction and examining leader-to-follower vs follower-to-leader connections, plus leadership strength effects.

Behavior and leadership: Participants reproduced durations appropriately; during eye contact they underestimated durations relative to control and partners' estimates were correlated, indicating interaction. In some dyads, one participant consistently broke eye contact first (leader), defining leader–follower roles. Undirected inter-brain synchronization: Significant gamma-band (30–45 Hz) cluster with higher inter-brain ciPLV during eye contact vs control (54 links; cluster t=132.94, t-critical=76.84, p=0.0148), predominantly right hemisphere and midline. Direct comparison within the cluster: t(49)=4.282, p<0.001, Cohen's d=0.606. Shuffled inter-dyad control showed low probability of observing the real cluster under common-input assumptions (p=0.0020; real cluster t=132.94 vs shuffled t-critical=79.02). Undirected intra-brain synchronization: Significant gamma-band intra-brain ciPLV cluster higher in eye contact vs control (cluster t=192.07, t-critical=87.94, p=0.0123), mainly right parietal to widespread regions. Within-cluster comparison: t(99)=3.676, p<0.001, d=0.368. No significant differences in gamma power between conditions. Undirected network comparisons (eye contact z-scored vs control): Inter-brain edges showed higher z-scores than intra-brain edges overall, t(99)=5.623, p<0.001, d=0.562. Mixed ANOVAs:

• Network strength: main effect edge type (inter>intra), F(1,98)=37.551, p<0.001, partial η²=0.277; main effect friendship, F(1,98)=9.567, p=0.003; interaction F(1,98)=7.792, p=0.006. Friends>strangers for inter-brain strength t(98)=3.029, p=0.003, d=0.608; no intra difference.
• Density: inter>intra, F(1,98)=28.532, p<0.001; friendship effect F(1,98)=6.627, p=0.012; interaction F(1,98)=6.627, p=0.012. Inter-brain density friends>strangers t(98)=2.955, p=0.004, d=0.593; no intra difference. Both groups showed inter>intra density.
• Efficiency: friends showed higher global efficiency t(48)=2.664, p=0.009, d=0.535 and local efficiency t(48)=2.192, p=0.031, d=0.440; no differences in modularity or assortativity.
• ROI effects: Midposterior (Pz/Oz) and midfrontal (Fz/Cz) had highest degree and efficiency; frontal (LF/RF) lowest. Friends>strangers main effects without ROI interactions, indicating similar topology but stronger networks in friends.
• Rich-club: Present in 43/50 dyads (86%); no difference in prevalence between friends (78.3%) and strangers (92.6%), χ²=2.119, p=0.145. Hubs most often at Pz, P4/P3, Cz/C4/C3; frontal regions least likely hubs. Directed inter-brain synchronization (PSI): Significant alpha-band (8–12 Hz) cluster during eye contact from leader to follower (cluster stat=43.35, t-critical=27.33, p=0.0278), with right hemisphere leader regions driving follower frontal and midline regions. PSI strength correlated with leadership strength during eye contact (r=0.508, p<0.001), not in control (r=−0.265, p=0.066). No cluster from follower to leader or in control>eye-contact. Directed intra-brain synchronization (PSI): Significant alpha-band cluster (eye contact>control) showing frontal-to-parietal/occipital driving, especially left/frontal driving right posterior (cluster stat=15.737, t-critical=10.157, p=0.033). No significant clusters in other bands. Directed network properties and leadership: Leaders exhibited a higher proportion of outgoing inter-brain edges than followers, with an interaction by leadership strength: F(1,94)=5.642, p=0.020; interaction F(1,94)=13.500, p<0.001; in strong-leadership dyads leaders>followers in outgoing edge proportion t(40)=2.946, p=0.005, d=0.909; not significant in weak-leadership dyads. Outgoing edge strength showed a three-way interaction (leaders vs followers × strong vs weak × outgoing vs incoming), F(1,94)=9.010, p=0.003; in strong-leadership dyads, leaders had stronger outgoing connections than followers t(40)=2.744, p=0.009, d=0.847; no differences for incoming connections. Global network properties (efficiency, modularity, assortativity) did not differ between strong vs weak leadership dyads. Leaders in strong-leadership dyads showed marginally higher node global efficiency (access to network), especially in midfrontal and right centrotemporal regions.

Eye contact selectively enhances inter-brain synchronization more than intra-brain synchronization, indicating that eye contact is an inherently social and communicative cue. Undirected synchronization increases in gamma band for both inter- and intra-brain links, while directed synchronization emerges in alpha band, consistent with roles of lower frequencies in top-down coordination and cross-frequency hierarchy (lower frequencies organizing higher). Eye contact hyperbrain networks display rich-club organization with hubs in midline and parietal regions, suggesting these nodes integrate within-brain and across-brain information during mutual gaze. Social relationship modulates network strength and density: friends show stronger and denser inter-brain connections and higher efficiency, though overall topology (e.g., modularity, assortativity) is similar. Leadership spontaneously observed in gaze behavior predicted directed inter-brain coupling in alpha from leader to follower, and leaders exhibited greater outgoing connection strength and network access in strong-leadership dyads. Together, findings support the interactive brain hypothesis: analyzing coupled brains yields insights not apparent from single-brain measures, and inter-brain dynamics may be more sensitive to social context than intra-brain dynamics.

This study demonstrates that mutual eye contact increases inter- and intra-brain synchronization, with stronger effects between brains than within, and reveals hyperbrain network features during eye contact. Undirected gamma-band synchrony and directed alpha-band interactions co-occur, suggesting coordinated multi-frequency dynamics linking internal processing with social coupling. The hyperbrain network shows rich-club hubs in midline and parietal regions that may integrate self- and other-related information. Inter-brain synchronization is stronger in friends than strangers, and directed coupling flows from leaders to followers in dyads with clear leadership. These results advance understanding of eye contact as a fundamental social signal shaping brain-to-brain coupling. Future research should: elucidate causal mechanisms (e.g., cross-frequency coupling across brains), increase ecological validity (naturalistic conversations, spontaneous eye contact), test generalization to other tasks (e.g., dance, dialogue), improve spatial resolution (source localization), and assess clinical relevance (e.g., autism) and temporal dynamics around initiation and breaking of eye contact.

• Ecological validity: Eye contact was experimentally imposed and coupled to a time-reproduction task; naturalistic, spontaneous eye contact may differ.
• Multiple comparisons: A range of network metrics were tested without stringent correction, raising potential Type I error risk, though a two-step approach mitigated this.
• Spatial resolution: EEG sensor-level analyses limit precise source localization; interpretations of cortical regions are inferential.
• Task specificity: Leadership effects were defined via gaze-breaking and may not generalize to other interaction types requiring continuous mutual adaptation.
• Power and data exclusions: Some dyads excluded due to insufficient valid trials; directed analyses limited to dyads with clear leadership.
• Potential shared input confounds addressed for inter-brain gamma via shuffling, but other unmeasured common factors may remain.