Distinct fNIRS Inter-Brain Coupling Patterns for Cooperation versus Competition in a Tennis Game

Abstract

Cooperation and competition represent two fundamental modes of social interaction, yet their underlying neural mechanisms remain incompletely understood. Functional near-infrared spectroscopy hyperscanning, enabling simultaneous measurement of hemodynamic activity across individuals, offers unique insights into the neural substrates underlying naturalistic interactions. Using this technique, we investigated cross-channel inter-brain coupling (IBC) between interacting individuals during cooperative and competitive play in a motion-sensing tennis game. Compared to resting-state and solo gameplay with observation, both conditions elicit significantly enhanced not only IBC between the dyads’ sensorimotor regions, but also cross-regional coupling between one participant’s sensorimotor cortex and the other’s dorsolateral prefrontal cortex (DLPFC) as well as temporoparietal junction, suggesting the contribution of high-order cognition networks to the observed IBC. Notably, competitive interactions produce stronger cross-regional IBC between DLPFC and sensorimotor regions than cooperative ones, implying an intensified demand for cognitive control during competition. Conversely, cooperation enhances neural coupling between team-mates within their prefrontal cortices, which could reflect shared goal representations. Behavioural cooperation performance is negatively correlated with the DLPFC–sensorimotor IBC. These spatially distinct patterns of condition-dependent neural coupling advance our understanding of the neural underpinnings of naturalistic social interactions.

Introduction

Cooperation and competition are fundamental and distinct forms of social interaction (Decety et al. 2004, Wittmann et al. 2016). Cooperation involves individuals sharing a common goal, whereas competition occurs when their goals conflict (Deutsch 1949, Vonk 1998). They are two key concepts to understanding social interaction (Kingsbury and Hong 2020). Despite a growing body of studies on their neural substrates, the specific roles of different brain regions in supporting these behaviours, especially in naturalistic settings, remain to be fully elucidated.

Several neural systems important for social cognition have been discovered in single-brain studies. First, the mirror neuron system, encompassing superior and inferior parietal lobules and premotor cortex, is activated when people observe or imitate others’ actions (Rizzolatti and Craighero 2004, Iacoboni 2009, Molenberghs et al. 2009). Second, the mentalizing system, including the medial prefrontal cortex and temporoparietal junction (TPJ), is involved in “theory of mind” tasks, that is inferring other agent’s intentions and thoughts (Decety et al. 2004, Xi et al. 2011, Kennedy and Adolphs 2012, Yeh et al. 2015). Third, executive-control regions, most notably the dorsolateral prefrontal cortex (DLPFC), underlie multiple cognitive processes such as goal maintenance, working memory, and response inhibition (Miller and Cohen 2001, Friedman and Robbins 2022). Social interaction, particularly in dynamic and reciprocal contexts, requires involvement of all these systems for encoding actions, inferring intentions, and coordinating information to orchestrate one’s own responses (Redcay and Schilbach 2019, Hamilton 2021, Barnby et al. 2023).

Over the past two decades, hyperscanning, the simultaneous measurement of brain activity from multiple individuals, has dramatically advanced our understanding of social interaction’s neural foundation (Montague et al. 2002, Babiloni and Astolfi 2014, Hamilton 2021). Hyperscanning operates with a multi-brain framework, investigating neural signatures of social interaction by examining the coupled neural systems of participants (Redcay and Schilbach 2019, Kingsbury and Hong 2020). Most hyperscanning research has focused on IBC during cooperative and competitive conditions (Babiloni et al. 2012, Cui et al. 2012, Liu et al. 2016, Balconi and Vanutelli 2017, Liu et al. 2018, Wang et al. 2018, Kingsbury et al. 2019, Czeszumski et al. 2020, Gugnowska et al. 2022, Zhao et al. 2024). IBC has emerged as a novel neural signature characterizing social interactions (Redcay and Schilbach 2019, Czeszumski et al. 2020). For instance, electroencephalograph (EEG) hyperscanning studies have revealed increased IBC in various frequency bands, mainly alpha and theta bands, during social interaction (Babiloni and Astolfi 2014, Mu et al. 2016, Liu et al. 2021).

Alternatively, functional near-infrared spectroscopy (fNIRS) hyperscanning is also commonly used in studying neural correlates of social interaction (Nam et al. 2020). Similar to functional magnetic resonance imaging, fNIRS signals reflect hemodynamic changes in localized brain regions. Moreover, portable fNIRS equipment allows interaction in the same room, and fNIRS is less sensitive to motion artefacts than EEG (Lloyd-Fox et al. 2010, Pinti et al. 2015), making it suitable for simultaneous social interaction studies. In one early work, Cui et al. (2012) developed a collaborative keypress task where two participants tried to press a single key simultaneously. Results revealed higher IBC in the superior frontal gyrus during cooperation. Multiple paradigms have since been used to investigate IBC during different forms of cooperation and competition, such as turn-based computation game, joint music performance, and problem-solving discussion (Liu and Pelowski 2014, Osaka et al. 2015, Liu et al. 2017, Lu and Hao 2019). Cooperative behaviour correlates with enhanced IBC in frontal and temporoparietal regions, overlapping substantially with neural systems of social interaction identified in single-brain studies (Mu et al. 2018, Czeszumski et al. 2020, Kingsbury and Hong 2020). In contrast, competition correlates with heightened IBC in the TPJ and the dorsolateral prefrontal cortex (Kayhan et al. 2022, Zhang et al. 2024), which are also neural signatures of cooperation identified across multiple hyperscanning studies (Xue et al. 2018, Lu and Hao, 2019, Lu et al. 2019a, 2019b, Li et al. 2020, Sun et al. 2020). How these regions differently engage in cooperation and competition remains to be elucidated.

Recent neuroscience research increasingly adopts naturalistic paradigms to study social interactions (Singer 2012, Liu et al. 2016, Liu et al. 2021, Hamilton and Holler 2023). Rather than isolating specific cognitive processes, these studies engage a broader spectrum of cognitions underlying social interaction. Naturalistic paradigms closely approximate real-world scenarios, potentially eliciting mental processes absent in more constrained experimental designs (Shamay-Tsoory and Mendelsohn 2019, Kihlstrom 2021). For example, naturalistic interactions usually involve unpredictable temporal dynamics and heightened emotional engagement (Schilbach et al. 2013, Balconi and Vanutelli 2017). Previous research has investigated IBC during naturalistic scenarios involving interpersonal communication, including group discussions (Jiang et al. 2012, 2015, Xue et al. 2018, Lu and Hao 2019), singing together (Osaka et al. 2015), and social games like Jenga and Tangram (Liu et al. 2016, Fishburn et al. 2018, Li et al. 2021a). Results from these paradigms align with previous findings while highlighting increased involvement of regions within the mentalizing and cognitive control systems. However, the inter-brain neural underpinning of non-verbal naturalistic interaction remains largely unexplored.

Non-verbal social interactions are common in daily life, such as musical performances and athletic competitions. They have been studied using paradigms involving joint musical performance (Lindenberger et al. 2009, Babiloni et al. 2012, Müller et al. 2013, Gugnowska et al. 2022), collaborative flight simulation (Astolfi et al. 2011, Toppi et al. 2016), and multiplayer motion-sensing sports gaming (Liu et al. 2021). Most of these studies utilized EEG hyperscanning methodology, revealing electrophysiological signatures of non-verbal naturalistic interactions. However, the precise localization of these neural signals remains uncertain due to the EEG’s limited spatial resolution.

In this study, we adapted a naturalistic paradigm using a motion-sensing tennis game to create an immersive environment simulating real tennis matches with constrained body movements. To investigate IBC during social interaction, we employed fNIRS hyperscanning to simultaneously measure brain activities from participant dyads. Our paradigm incorporates both cooperative and competitive conditions within the same game, allowing direct comparisons between these fundamental social contexts. This design builds upon existing literature that has identified neural signatures specific to cooperative or competitive interactions (Cui et al. 2012, Liu et al. 2015, 2016, 2017, 2021, Špiláková et al. 2020). We hypothesized the involvement of previously identified regions such as frontal and temporoparietal areas while expecting selective neural engagement differences between cooperative and competitive interactions, highlighting their unique cognitive mechanisms.

Methods

Participants

Eighty-eight male participants (age = 22.9 ± 3.1 years) from Tsinghua University were randomly paired, forming 44 dyads, all confirmed as mutual strangers. Eight dyads were excluded due to technical issues, leaving 72 participants (36 dyads, age = 23.0 ± 3.0 years) for analysis. The sample size was determined to reach 36 dyads based on a priori power analysis using G*Power to achieve a statistical power of 0.90 for an effect size of f = 0.25 at α = 0.05. All participants were right-handed, with normal or corrected-to-normal vision, and had prior digital game experience. Male participants were recruited to control gender differences in IBC revealed by previous studies (Baker et al. 2016, Li et al. 2021a). All participants provided informed consent and received monetary compensation. The study was approved by Ethics Committee of Tsinghua University and conducted in accordance with the Declaration of Helsinki.

Paradigm and procedure

Each dyad played Mario Tennis Aces (Nintendo, Japan) on a Nintendo Switch console (Fig. 1a and b) using the same two avatars (Mario and Luigi) to control for perceptual differences. Participants were first introduced to the game operations and practiced until proficient. The scoring rules mirrored real-world tennis. Participants then played multiple doubles games with Artificial intelligence (AI) players under three conditions (Fig. 1c), that is Cooperation (participants teamed up against two AI players), Competition (participants played against each other with AI teammates), and Solo/Observation (one participant played with three AI players while the other observed). As most participants had never played this game before, all AI’s proficiency levels were set to ‘Novice’ to ensure accessibility and control for difficulty level across sessions. Conditions were pseudorandomized across dyads. Pre-experiment resting-state data (3 mineyes open, 3 min eyes closed) were recorded as baseline.

Figure 1.

Experiment setup. (a) Mario Tennis Aces game screenshot. (b) Experimental scene during gameplay (with permission from the participants). (c) Experiment procedure. (d) fNIRS channel layout.

The game was displayed on a 55-inch TV in front of the participants. Participants stood still after adjusting their position. The position of the avatars was controlled by the computer and participants were instructed to move only their right hand. Participants swing the Joy-Con (game controller) in their right hand at the correct timing with proper strength and direction to catch a ball. They continued playing best-of-three matches until the minimum duration of the condition (10 min) was fulfilled. If less than 2 min remained at the end of a match, a single set was played. After each condition, participants filled out a short questionnaire regarding perceived cooperation and competition levels, and trust in the other participant, using 7-point Likert scales.

A high-resolution camera recorded game footage to code the starting and ending time points of each match. To align the timing of the videos and neural signals of the dyad, three beeps were played and recorded by the camera with three triggers sent simultaneously to the fNIRS devices before the game started.

Several personality traits of the participants related to cooperative and/or competitive dispositions were collected using scales before the experiments, including the Basic Empathy Scale (Jolliffe and Farrington 2006), the Assessment of Achievement Motives (Nygård and Gjesme 1973), the Cooperative/Competitive Strategy Scale (Simmons et al. 1988), the Interpersonal Trust Scale (Rotter 1967, Ding and Peng 2020), the Brief Aggression Questionnaire (Webster et al. 2014), and the Big Five Inventory (John et al. 1991).

Functional near-infrared spectroscopy recording and data preprocessing

During the experiment, the fNIRS signals of both participants were recorded simultaneously using two identical portable NIRS systems (NirSmart, Danyang Huichuang Medical Equipment Co., Ltd, China) with a sampling frequency of 11 Hz. The concentration changes of brain oxygenated haemoglobin (HbO) and deoxy haemoglobin (HbR) were detected using near-infrared light of two different wavelengths (730 nm and 850 nm). The experiment uses 24 light sources and 16 detectors to form 61 effective channels. The average distance between the source and the detector is 3 cm, with reference to the international 10/20 system for positioning. Based on previous studies, the brain areas mainly observed in this study are the prefrontal cortex, the inferior parietal lobule, the superior temporal gyrus (Cui et al. 2012, Liu et al. 2015, 2017, Liu et al. 2016), as well as extensive areas of the somatosensory and motor cortex (see Fig. 1d). With these regions covered, the remaining five channels were placed exclusively over the left hemisphere targeting the somatosensory association cortex, considering the primary involvement of right-hand movements in our experiment. Detailed channel locations are available at osf.io/bh2m5.

The preprocessing of the fNIRS signals was conducted in two steps with the MATLAB toolkit HomER2 (Huppert et al. 2009), following previous fNIRS IBC studies (Li et al. 2021b, 2023). First, we used the targeted principal component analysis (tPCA; function hmrMotionCorrectPCArecurse, STDEVthresh = 30, AMPthresh = 0.5) to remove motion-related artefacts. This approach was to detect motion principal components at targeted epochs and project them out (Yücel et al. 2014). Second, we further identified remaining channel-wise motion-related artefacts (function hmrMotionArtifactByChannel, STDEVthresh = 30, AMPthresh = 0.5) and reduced them using a cubic spline interpolation method (function hmrMotionCorrectSpline; Scholkmann et al. 2010). Across participants, an average of 3.2% of fNIRS data during the entire tennis game (SD = 4.3%, range: <0.1%–17.5%) were marked as motion artefacts and corrected. Across channels, an average of 3.2% (SD = 4.6%, range: 0.3%–26.5%) were corrected. The preprocessed data were then visually inspected. No participants or channels were excluded due to excessive motion artefacts. Representative examples of fNIRS time series data before and after preprocessing are presented in Fig. 2. The preprocessed HbO signals were used as indicators of functional brain activations in subsequent analyses (Cui et al. 2012, Li et al. 2021b, 2023).

Figure 2.

Representative fNIRS time series data before and after preprocessing from different participants and channels.

Data analysis

Behavioural analysis

Subjective ratings of cooperation, competition and trust from both participants were averaged to form dyad-level ratings. The ratings following the cooperation and competition conditions were compared to validate experiment manipulations.

We explored the relationship between subjective feelings and objective performance using Spearman’s rank correlations. Behavioural cooperation and competition performance were assessed by calculating score ratios between human and AI teams during cooperation, and between the two human–AI teams during competition.

Inter-brain coupling analysis

IBC between participants was calculated as below (Li et al. 2021b, 2023). First, the wavelet transform coherence (WTC) approach was used to measure the similarity of neural signals between two participants during a given period (Cui et al. 2012, Jiang et al. 2015, Baker et al. 2016, Liu et al. 2016, Kayhan et al. 2022), which produced inter-brain similarity for each time point across various frequency bins (Grinsted et al. 2004). The non-analytic Morlet wavelet was used for WTC calculations. Following previous studies, WTC results above 0.7 Hz or below 0.01 Hz were excluded from further analyses to avoid higher-frequency physiological noise and low-frequency fluctuations respectively (Guijt et al. 2007, Tong et al. 2011, Liu et al. 2019, Li et al. 2021b, 2023). For the cooperation, competition and solo/observation conditions, WTC was calculated for each match. For the resting-state data, WTC was calculated separately for the eyes-open and eyes-closed periods. WTC was performed for all possible 3721 channel combinations across 61 × 61 channels. To ensure reliable results for all time points, fNIRS data were extracted with a 100-s buffer before and after each period of interest. WTC during these extraneous segments were then removed after calculations. For each period, IBC was averaged across time and Fisher-z transformed. The coupling values from all periods within the same condition were then further averaged. Finally, since two participants in a dyad engaged symmetrically, the coupling values of corresponding channel combinations (e.g. channel combinations 1-40 and 40-1) were averaged. The above computation yielded IBC of 1891 channel combinations × 75 frequency bins (0.01–0.7 Hz) in each condition.

For the main analyses, we compared the IBC among cooperation, competition, and resting-state conditions. The cooperation versus competition contrast aims at discovering the unique IBC patterns associated with the two distinct interaction forms while controlling for comparable perceptual environment and motor activity level. The resting-state condition is chosen as the baseline as it provides a task-free baseline, which could highlight IBC inherent in general social interaction. To test the effect of different conditions, one-way repeated measures ANOVAs (rmANOVA) were conducted for the IBC of each channel combination and frequency bin for all dyads, with conditions as the within-subject factor. A nonparametric cluster-based permutation method was used to address the multiple comparison problem, which is well-suited for neural data analysis as it accounts for the spectral continuity of real statistical effects (Maris and Oostenveld 2007, Li et al. 2021b, 2023). Clusters were identified as contiguous frequency bins with uncorrected P < .05 within each channel combination, with the maximum F value among the frequency bins as the cluster-level statistic. Neighbouring channel positions were not considered in the identification of clusters to avoid potential confounding from global physiological noise. A null distribution of cluster-level statistics was generated by shuffling the condition labels of IBC (n = 1000 permutations) and obtaining the maximum cluster-level statistic per permutation. Cluster P-values were computed by comparing real statistics to the null distribution, with clusters deemed significant at P < .05. In the post-hoc analyses, IBC within each significant cluster was first averaged across frequency bins and then compared pairwise using Fisher’s Least Significant Difference test for each condition.

We also conducted control analyses for the significant clusters to evaluate whether the IBC observed during cooperation and competition exceeded that of the solo/observation condition, using one-tailed paired-sample t-tests. It should be noted that the task asymmetry under this condition, due to the different roles of playing and observation, could result in a lowered IBC level. Indeed, our initial analysis indicates that IBC during solo/observation was not necessarily higher than during resting-state (see Fig. S1).

Behavioural relevance analyses

Clusters showing significant differences in IBC between cooperation and competition were further subject to correlational analyses with behavioural metrics, including the subjective ratings after cooperation and competition, the objective performance, and personality traits. For cross-participant comparisons, the percentage increase in IBC relative to resting-state baseline during cooperation and competition was calculated. Conditional means (cooperation and competition) and differences (cooperation minus competition) of IBC were then averaged across clusters showing similar directional effects (i.e. clusters with higher IBC during competition than cooperation, and vice versa) as neural indices.

Single-brain control analyses

To explore the possible unique value of IBC, single-brain analyses were conducted as control. Specifically, single-brain activation and intra-brain connectivity were calculated for channels corresponding to the channel combinations exhibiting significantly different IBC between cooperation and competition.

For all relevant channels, single-brain activations were defined as standardized HbO signals, which were converted into z-scores by the middle 90 s of resting-state data. Participants’ average activations during cooperation and competition were then compared using paired sample t-tests.

Intra-brain connectivity was calculated using a procedure similar to that for IBC but applied to neural signals from channel combinations within the same participant. rmANOVAs were applied to compare intra-brain connectivity across cooperation, competition, and resting-state.

Results

Behavioural results

All scales showed at least acceptable reliability (Cronbach’s αs > 0.70), except for the Interpersonal Trust Scale (α = 0.51), which was excluded for further data analysis.

Participants reported significantly higher subjective feelings of cooperation for the cooperation condition (mean = 5.68, SD = 0.99) compared to the competition condition (mean = 2.08, SD = 0.96; t(35) = 15.49, P < .001) and significantly higher subjective feelings of competition for the competition condition (mean = 6.00, SD = 0.53) in contrast to the cooperation condition (mean = 1.99, SD = 0.93; t(35) = 20.73, P < .001). Participants also reported significantly higher subjective feelings of trust in their partner after cooperation (mean = 5.40, SD = 0.86) than competition (mean = 4.22, SD = 1.05; t(35) = 6.54, P < .001). The subjective feelings of cooperation and trust for the cooperation condition were positively correlated (r(35) = 0.36, P = .029).

In the cooperation condition, the score ratio between the participant side and the AI side was smaller than one (median = 0.62, interquartile range: Q1 = 0.47, Q3 = 0.93). The subjective feelings of cooperation and trust of participants were not significantly correlated with the score ratios (cooperation: Spearman’s ρ = 0.20, P = .238; trust: Spearman’s ρ = 0.09, P = .619). Similarly, feelings of competition showed no significant relationship with the score ratios of the winning to losing sides (Spearman’s ρ < 0.01, P = .990; score ratio median = 1.40, Q1 = 1.17, Q3 = 1.57).

Inter-Brain coupling related to cooperation and competition

Non-parametric permutation tests identified 65 clusters showing significant differences in IBC among cooperation, competition, and resting-state conditions (rmANOVA partial: mean = 0.36, SD = 0.03, range: 0.33–0.45). These significant clusters comprise different channel combinations, with varying frequency bins, which can be grouped into 0.01–0.014 Hz, 0.05–0.1 Hz, and 0.35–0.7 Hz (Fig. 3a and b). Across all significant clusters, IBC is significantly higher during both cooperation and competition compared to the resting-state (cooperation versus resting-state: ps < .014; competition versus resting-state: ps < .012, LSD corrected). Moreover, post-hoc analyses identified three distinct patterns among significant clusters, which diverged when comparing IBC between cooperation and competition. The first pattern shows no difference of IBC between conditions, the second implies higher IBC during competition, and the third demonstrates higher IBC during cooperation (Fig. 3c).

Figure 3.

Results of IBC across frequency bins and channel combinations. (a) Heat map of –log₁₀(rmANOVA P-values) from for all frequency bins within channel combinations of significant clusters. White dashed lines indicate frequency ranges of significant clusters, that is 0.01–0.014 Hz, 0.05–0.1 Hz, and 0.35–0.7 Hz. (b) Heat maps of –log₁₀(minimum rmANOVA P-value) for each channel combination. Left: all channel combinations; Right: significant clusters only. (c) Topographical representation of channel combinations in significant clusters No. 1–5 denote five clusters with significant conditional IBC differences. Comp: competition; Coop: cooperation; RS: resting-state.

Most of the significant clusters (60 out of the 65 clusters) implicate no difference in IBC between cooperation and competition and were extensively located within the sensory and motor cortices. Clusters exhibiting significant conditional differences include channel combinations between the mentalizing and cognitive control neural areas, that is the dorsolateral prefrontal cortex DLPFC; Brodmann’s area (BA 9 and 46) or temporoparietal junction (TPJ; BA 40), in one participant and the sensory and motor cortices of the other. Detailed locations of the significant clusters are available at osf.io/bh2m5. We conducted a control analysis to compare the observed IBC during cooperation and competition with that during solo gameplay and observation. Across all 60 clusters, IBC during cooperation and competition was significantly higher compared to that during solo/observation condition (cooperation versus solo/observation: ps < .003 in 59 clusters, P = .037 in one cluster; competition versus solo/observation: ps < .001 in 59 clusters, P = .006 in one cluster).

Four significant clusters show significantly higher IBC during competition compared to cooperation (ps = .020, .014, .005, and .041, LSD corrected; Cohen’s d = 0.41, 0.43, 0.50, and 0.35). All of these clusters (labeled as Clusters No. 1–4; see Fig. 3c for detailed location) indicate IBC between the bilateral DLPFC of one participant and the sensorimotor cortices of the other, with three clusters (Clusters No. 1–3) consisting of the BA46 part of bilateral DLPFC (channels 3 and 8) and Cluster No. 4 located on the more posterior part of DLPFC (BA9) and close to the premotor cortex. For the control analysis, across all four clusters, IBC was significantly higher in both the cooperation and competition conditions compared to the solo/observation condition (cooperation versus solo/observation: ps < .001, <.001, = .009, and < .001; competition versus solo/observation: all ps < .001).

Moreover, one cluster (Cluster No. 5) demonstrates higher IBC during cooperation in contrast to competition (P = .043, LSD corrected; Cohen’s d = 0.35) and is located within the prefrontal cortex, specifically between the frontal eye fields (BA 8) and the frontopolar area (BA 10). Figure 4 illustrates the IBC of all participant dyads at these five clusters. Further control analyses revealed that, for Cluster No. 5, IBC was significantly higher during cooperation compared to the solo/observation condition (P = .031), whereas no IBC difference was observed between competition and solo/observation (P = .528).

Figure 4.

IBCs of all participants’ dyads during cooperation and competition relative to resting-state. Coop: cooperation; Comp: Competition. Clusters No.1-4 exhibit higher IBC in competition than cooperation, and Cluster No.5 exhibits higher IBC in cooperation than competition.

Behavioural relevance of IBC

The five clusters with significant conditional IBC differences were further analysed for potential behavioural relevance.

First, the IBC of two clusters (Clusters No. 2 and No. 4) during cooperation is negatively correlated with cooperation scores, that is the score ratios between the human and AI sides (Cluster No. 2: r(35) = −0.43, uncorrected P = .007; Cluster No. 4: r(35) = −0.34, uncorrected P = .040), with one cluster (Cluster No. 2) survive FDR correction (corrected P = .038 and .099; Fig. 5). Both clusters show higher IBC for competition compared to cooperation in post-hoc analyses. We further calculated the average IBC during cooperation for all clusters with the same pattern (i.e. Clusters No. 1–4), and it is also negatively correlated to cooperation scores (r(35) = −0.45, P = .006). The cluster with higher IBC for cooperation than competition (Cluster No. 5) does not show a significant correlation with cooperation scores (r(35) = −0.26, P = .117). We also explored the correlation between competition scores (score ratios of winning to losing sides during competition) and IBC during competition. No cluster shows significant correlation with competition performance (ps > .366, rs = .06, .07, .08, .16 and .11 for Clusters No. 1–5, respectively).

Figure 5.

Correlations between the IBC during cooperation and cooperation scores. Neural coupling was quantified as the ratio between IBC observed during cooperation and baseline coupling measured during resting-state tasks.

We analysed correlations between participants’ personality traits and their differential IBC index (cooperation minus competition; see Table 1). No individual cluster exhibits significant correlation with any personality traits after applying FDR correction (see Table S1 for detailed results). After averaging across Clusters No. 1–4, the differential IBC shows a marginal positive correlation with participants’ Openness scores (r(35) =0.43, FDR-corrected P = .083).

Table 1.

Pearson correlation coefficients and uncorrected P-values (in parentheses) of neural coupling and personality traits.

Cluster	Personality Neural index	O	C	E	A	N	Emp	Agg	Coop	Comp	AMS
No. 1–4 averaged (Comp > Coop)	Diff	0.43 (0.008)	−0.07 (0.698)	0.02 (0.927)	−0.29 (0.091)	0.25 (0.134)	0.05 (0.779)	0.07 (0.665)	−0.04 (0.809)	0.27 (0.116)	0.18 (0.285)
	Coop	0.13 (0.449)	0.16 (0.364)	−0.04 (0.824)	−0.35 (0.034)	−0.06 (0.727)	−0.05 (0.754)	0.14 (0.431)	−0.29 (0.082)	0.05 (0.791)	−0.01 (0.944)
	Comp	−0.32 (0.059)	0.21 (0.225)	−0.05 (0.769)	−0.03 (0.854)	−0.31 (0.067)	−0.10 (0.573)	0.05 (0.785)	−0.22 (0.193)	−0.23 (0.184)	−0.19 (0.256)
No. 5 (Coop > Comp)	Diff	0.16 (0.339)	−0.03 (0.884)	0.06 (0.717)	−0.14 (0.401)	0.15 (0.389)	0.02 (0.901)	0.18 (0.300)	0.01 (0.951)	−0.01 (0.965)	0.11 (0.527)
	Coop	0.20 (0.244)	0.03 (0.866)	0.20 (0.252)	0.03 (0.869)	−0.09 (0.609)	0.06 (0.738)	−0.07 (0.702)	−0.09 (0.619)	0.00 (0.992)	0.23 (0.184)
	Comp	0.07 (0.682)	0.07 (0.683)	0.19 (0.266)	0.22 (0.207)	−0.30 (0.074)	0.05 (0.761)	−0.31 (0.068)	−0.13 (0.451)	0.01 (0.967)	0.18 (0.306)

All neural coupling was quantified as the ratio between coupling observed during cooperation/competition and baseline coupling measured during resting-state tasks.

Diff: differential neural coupling (cooperation minus competition); Coop: neural coupling during cooperation. Comp: neural coupling during competition. O, C, E, A, N: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism in the Big Five Inventory; Emp: Basic Empathy Scale; Agg: Brief Aggression Questionnaire; Coop/Comp: Cooperative/Competitive Strategy Scale; AMS: Assessment of Achievement Motives.

The IBC during cooperation is not significantly correlated with participants’ subjective ratings of cooperation (all uncorrected ps > .23) or trust (all uncorrected ps > .45) for any of the clusters. Likewise, IBC during competition exhibits no significant correlation with participants’ subjective ratings of competition (all uncorrected p > .23).

Single-brain activation

All channels corresponding to the channel combinations in clusters with significant conditional IBC differences were analysed for single-brain activation. No channel shows significantly different brain activation between cooperation and competition (all ps > .137).

Intra-brain connectivity

For all channel combinations in clusters with significant conditional IBC differences (Clusters No. 1–5), the intra-brain connectivity during both cooperation and competition is significantly higher than resting-state (cooperation versus resting-state: ps < .001, < .001, <.001, = .020, and <.001; competition vs resting-state: ps < .001, < .001, <.001, = .023, and <.001). However, no significant conditional difference between cooperation and competition was found for intra-brain connectivity (ps = .787, .095, .922, .923, and .543; see Fig. 6).

Figure 6.

Intra-brain connectivity of all participants’ dyads during cooperation and competition relative to the resting-state. Coop: cooperation; Comp: Competition. Clusters No.1-4 exhibit higher IBC in competition than cooperation, and Cluster No.5 exhibits higher IBC in cooperation than competition.

Discussion

We investigated the neural substrates underlying cooperation and competition using a motion-sensing tennis game. By comparing cross-channel IBC across different conditions, we identified multiple brain regions exhibiting enhanced neural coupling at the sensorimotor cortices during social interaction relative to the resting-state baseline. Comparison of cooperation and competition further revealed spatially distinct IBC patterns under the two conditions, and correlation between IBC and task performance was discovered. Together, these findings provide new insights into the inter-brain neural mechanisms of cooperative and competitive social interactions.

Across multiple brain regions, we observed significantly higher IBC during both cooperative and competitive interactions compared to the resting-state and solo conditions. These results not only confirm the validity of the naturalistic tennis paradigm to simulate real-life interactions but also demonstrate the potential of cross-channel IBC analyses. Most of these coupling patterns show no significant differences between cooperation and competition. Shared neural substrates are expected as many fundamental social and cognitive processes are common to both modes of social interaction (Deutsch 1949, Balconi and Vanutelli 2017). Since both conditions involve participants focusing on a shared physical environment, which is a well-established driver of IBC (Hasson et al. 2004, 2012), within-sensorimotor IBC could emerge during social interaction according to this perceptual similarity account. Indeed, significant IBC clusters located in extensive sensorimotor areas across the dyad were observed. However, we also detected neural coupling between one participant’s DLPFC and TPJ areas and the other’s sensorimotor cortex during social interaction. This suggests that the observed IBC during cooperation and competition in this study involves not only shared perception but also higher-order cognitive functions supporting active engagement (Dikker et al. 2021), such as executive control as well as the mentalization of the partner’s action intentions and psychological states (Frith and Frith 2006, Schippers et al. 2009, Luyten and Fonagy 2015). The frequency of the observed DLPFC coupling is relatively high (0.35–0.7 Hz), but several lines of evidence suggest that it cannot be attributed to physiological noise such as cardiac activity. First, the peak frequency of this coupling was approximately 0.5 Hz, which is consistent with previous findings from social interaction studies (Holper et al. 2012, Nozawa et al. 2016) and significantly lower than typical cardiac frequencies centred around 1 Hz (Reddy et al. 2021). Given the physical exertion involved in this motion-sensing tennis game, actual heart rates were likely even higher. Second, the coupling was localized between the DLPFC and sensorimotor cortex, rather than showing widespread effects characteristic of extra-cerebral artefacts. Additionally, the frequency of meaningful neural synchrony is believed to align with the temporal dynamics of the task (Cui et al. 2012, Cheng et al. 2015, Osaka et al. 2015), and the observed IBC frequency is consistent with the timescale of tennis gameplay.

The contrast between competition and cooperation revealed stronger DLPFC–sensorimotor IBC between the dyad during competitive gameplay. Prior research has established the critical role of DLPFC in both cooperative and competitive interactions (Li et al. 2020, Lu et al. 2019a, 2019b, Lu and Hao 2019, Sun et al. 2020, Xue et al. 2018, Zhang et al. 2024). Our findings not only validate the involvement of DLPFC in both contexts but also reveal its enhanced engagement during competitive interactions. Given the established role of the DLPFC in executive function and cognitive control (Friedman and Robbins 2022, Miller and Cohen 2001), the observed DLPFC-related IBC likely reflects the heightened cognitive demands during cooperative and competitive gameplay. Participants were required to maintain overarching goals, dynamically adapt strategies, monitor the states of other agents, and inhibit autonomous yet suboptimal motor responses. Furthermore, previous research has demonstrated that the DLPFC represents action–outcome associations in actual, hypothetical, and social contexts (Abe and Lee 2011, Barraclough et al. 2004, Procyk and Goldman-­Rakic 2006, Weissman et al. 2008). Since significant IBC between one’s DLPFC and the other’s sensorimotor cortex was observed, a possible account could be that the DLPFC actively encodes and evaluates action–outcome associations not only for oneself but also for the interacting individual. This could enhance game engagement and facilitate one’s own action preparation by maintaining an internal model of the ongoing interaction. While the present inter-brain findings could imply DLPFC as a key integrative hub for cognitive control and social cognition (Hutcherson and Tusche 2022, Tusche and Hutcherson 2018, Weissman et al. 2008), further research is necessary to directly test this interpretation.

Importantly, DLPFC–sensorimotor coupling was significantly stronger during competition than cooperation. McDonald et al. (2020) discovered that DLPFC activity is selectively modulated by the degree to which an opponent’s actions influence one’s own outcomes during competitive gameplay. In light of this, and considering the DLPFC’s established role in integrating value signals into cognitive control (Dixon and Christoff 2014, Chu et al. 2023), one possible explanation for the observed IBC difference is that the utility of monitoring the other player’s behaviour is greater during competition than cooperation. During competition, an opponent’s actions and their outcomes have a more direct influence on one’s outcomes (e.g. the direction and placement of the opponent’s shot determine where a player needs to move to catch the ball), potentially leading participants to allocate greater attentional resources to the other player’s behaviour, which could in turn be reflected in stronger DLPFC–sensorimotor coupling. Conversely, the cooperative condition may have encouraged mutual trust, which reduces the motivation to closely monitor the partner. This interpretation aligns with the observed negative correlation between DLPFC–sensorimotor IBC and cooperation performance: successful cooperation may be associated with less intensive partner monitoring, and thus weaker coupling in this circuit. While these interpretations remain speculative, they highlight the potential for task context—particularly differences in interdependence and motivation—to shape the neural dynamics of social interaction. Taken together, our results suggest that the DLPFC may differentially contribute to processing partner-­related information across competitive and cooperative contexts, possibly modulated by the perceived utility of predicting others’ action–outcome associations.

The enhanced DLPFC coupling during competitive interactions appears to differ from previous findings in problem-solving paradigms, where stronger DLPFC coupling was noted during cooperation in contrast to competition (Lu et al. 2019). Furthermore, a negative correlation between DLPFC coupling and behavioural performance during cooperation was observed in this study, while positive correlation was reported in problem-solving contexts (Lu et al. 2019). These discrepancies may be attributed to the distinction that our study identified cross-regional IBC between one individual’s DLPFC and their partner’s sensorimotor regions, whereas Lu et al. (2019) investigated homologous DLPFC coupling. More importantly, while their problem-solving task relies predominantly on verbal communication, the current paradigm emphasizes non-verbal interaction. These findings collectively suggest potentially distinct patterns of neural coupling underlying verbal and non-verbal social interaction.

In contrast, IBC within prefrontal regions (specifically BA 8 and 10) was enhanced during cooperation compared to competition. This pattern aligns with previous findings demonstrating differential engagement of the prefrontal cortex during cooperative and competitive contexts (Decety et al. 2004, Cui et al. 2012, Tsoi et al. 2016). A popular explanation of cooperation-specific neural coupling is that shared goal representations between individuals are more closely aligned during cooperation. However, we did not find significant correlation between the aforementioned prefrontal IBC during cooperation and behavioural performance. Since cooperation in tennis doubles requires not only goal alignment but also proper task allocation for each player to handle balls where one has an advantage, the relationship between IBC and cooperative performance might be more complex. Interestingly, whereas all other clusters exhibited significantly greater coupling in both cooperative and competitive gameplay relative to solo/observation, coupling at this cluster during competition did not differ from the solo/observation condition. This pattern may point to neural processes that are selectively enhanced during cooperative interactions but remain relatively unchanged during competition or passive observation.

We further investigated individual differences in IBC during social interaction by examining its correlations with participants’ personality traits. Specifically, we explored whether these traits could account for the IBC difference observed between cooperation and competition. No significant correlation was identified between personality traits and neural coupling patterns after correcting for multiple comparisons. The ‘openness to experience’ trait shows a tendency toward a positive correlation with the IBC conditional difference, suggesting a possible role of personality in social interactions, although this relationship achieves only marginal significance after FDR correction. Given that openness to experience is associated with creative thinking (McCrae 1993, Connelly et al. 2014), this finding aligns with previous research demonstrating enhanced IBC at DLPFC during creative cooperation (Xue et al. 2018, Lu et al. 2019).

Several limitations of the current study should be taken into consideration in future studies. First, only two human players participated in the Tennis doubles in each experiment, which limits the ecological validity relative to real-world doubles tennis. Human players might introduce greater behavioural unpredictability and richer social–cognitive demands—such as dynamic role-switching, intention inference, and error monitoring—that may further modulate IBC patterns. Future studies involving all-human teams would be valuable for testing the generalizability of our findings and more comprehensively inspecting the IBC between team-mates and opponents. To support such investigations, expanding fNIRS coverage to include the ROIs of all four participants would be critical for capturing the full spectrum of intra- and inter-brain neural dynamics. Second, cooperative and competitive scenarios encompass diverse manifestations (Liu and Pelowski 2014). Whether the spatially distinct patterns reported in this study indicate task-­specific neural mechanisms or can be generalized to other social scenarios requires further validation with multiple paradigms. Also, specific elements of the experimental paradigm—such as AI’s proficiency level—may also influence both behavioural and neural outcomes. For example, cooperation performance and associated neural coupling might initially increase with moderate challenge, but drop if the AI becomes too difficult, potentially following an inverted-U relationship. Future studies could systematically manipulate these parameters to better understand their impact on social interaction and inter-brain dynamics. Third, the inclusion of only male dyads could limit the generalizability of our findings, highlighting the need for future studies to examine mixed-sex and female-only dyads. Significant behavioural and neural sex-related differences between same- and mixed-sex dyads have been observed in a cooperation task (Baker et al. 2016). Similarly, male-only dyads could exhibit distinct behavioural and neural patterns in competitive interactions (Benenson and Abadzi 2020, Yang et al. 2022). Future studies exploring this paradigm with diverse gender compositions would provide valuable insights into how gender dynamics influence neural coupling during social scenarios. Lastly, the absence of a consistent positive correlation between IBC and behavioural performance represents a limitation of the present work. Although our study was designed to contrast cooperation and competition at the condition level, future research should delve deeper into within-condition dynamics, examining, for example, how responsibility is distributed between collaborators and how trial-by-trial performance fluctuations relate to moment-to-moment changes in neural coupling.

Combined together, our analyses revealed systematic differences in the spatial distribution of IBC clusters corresponding to different conditions. Coupling patterns shared by cooperation and competition are predominantly localized to sensorimotor regions, along with cross-regional coupling between higher-order cognition (mentalizing and cognitive control) and sensorimotor networks. Elevated coupling during cooperation emerges within prefrontal regions. Enhanced coupling during competition occurs between the prefrontal cortex and sensorimotor regions, which is negatively correlated with behavioural cooperation performance.

Author contributions

Haoyu Zhang (Formal analysis [lead], Investigation [equal], Methodology [equal], Writing—original draft [equal], Writing—review & editing [equal]), Huashuo Liu (Conceptualization [equal], Investigation [equal], Methodology [equal], Writing—original draft [equal], Writing—review & editing [equal]), Zhuoran Li (Methodology [equal], Writing—review & editing [equal]), Dan Zhang (Conceptualization [equal], Funding acquisition [lead], Methodology [equal], Supervision [equal], Writing—review & editing [equal])

Supplementary data

Supplementary data is available at SCAN online.

Acknowledgements

We thank all participants for their time and Nina J. Chen for assisting with data collection.

Conflict of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Funding

This work was supported by the Major Science and Technology Program of Jiangsu Province (BG2024025), the National Social Science Foundation of China (Major Program: 24&ZD251), the National Natural Science Foundation of China (T2341003), the Graduate Education Innovation Grants of Tsinghua University (202504Z005) and the Education Innovation Grants of Tsinghua University (DX02_20).

Data availability

All codes and data necessary to reproduce the figures and tables in this article are available at osf.io/bh2m5. Complete data are available from corresponding authors upon reasonable request.