FIGURE 2. The schematic procedure of converting an fMRI dataset into a format ready for coherence analysis. (a) Player A, B, C, and D take turns interacting as pairs at the same time. Player D is outside the scanner. “Feedback” is defined as the time in the second stage of a trial when the actual choices of the pairs are made, to the end of the 2-s period in each trial. (b) A single deconvolutional GLM is created, with 9 stick-like extensions for each of the two regressors (“target_trial” and “other_trials”). This method allows a least-square separation of the target trial, from which the whole brain beta series are independently estimated subsequently for the 9 TRs following the feedback-initiated event; (c)Concatenating all the 9 TRs from each trial, resulting in 324 volumes [9 TRs x 6 trials (each run for each pair) x 6 runs] for each pair, as the periodic data format ready for coherence analysis. (d)The rTPJ is chosen as the seed ROI of the current study for the interpersonal seed-brain coherence analysis. The ROI beta series are extracted from the previously concatenated beta series and applied to the whole brain of the other subject. (e) The regions coupled with the rTPJ at the target frequency bin, the 36th (1/18 s = 0.0556 Hz), are separately mapped.
Various subgroups, indices and pairing combinations for analyses
As an alternative method to characterize individual behavioral characteristics, k-means cluster analyses, a module in JASP with the Silhouette optimization algorithm (Rousseeuw, 1987), was further implemented, yielding 3 subgroups:
  1. reciprocal: taking turns for the bigger reward, n = 49;
  2. dominant: choosing the bigger reward in 2 consecutive trials, n = 14;
  3. submissive: choosing the smaller reward in back-to-back trials, n = 15.
The descriptive statistics of each subgroup were shown in the Supplementary Table S1. These subgroup labels were instrumental in the brain-behavior connectivity analyses.
Two kinds of behavioral indices were devised: individual-wise and pair/dyad-wise. For the individual-based behavioral indices, each player interacted with the other three players and had 6 consecutive interactions with one player in a run, 6 runs in total; out of 6 runs, there were 35 times of alternation because the first interaction/trial was not included (35 times/trials x 3 other players = 105 trials). Three kinds of trial could be derived out of the total 105 trial pairs for each participant:
a) successful trials: Either Player A or B won the bigger reward.b) failed trials: Both Player A and B chose the same (bigger or smaller) reward, ending up both winning zero.c) successful trials with reciprocity: Player A won the bigger reward, and then B won the bigger reward in the next trial, or vice versa;d) successful trials without reciprocity: Either Player A or B won the bigger reward for two trials in a row;
Three individual-based indices were calculated based on these three kinds of trial pairs:
  1. the reciprocity index: defined as the number of successful trials with the winner of the bigger reward alternated (e.g., AB or BA), divided by the 105 calculable trials;
  2. the greed index: defined by the number of successful trials without reciprocity, such as one (e.g., AA or CC) winning consecutive bigger rewards;
  3. the total reward: defined as the sum of money each participant earned in the whole experimental session.
The 2nd set of, or pair-based, indices are:
pairwise success rate: defined as the pairs successfully winning the reward, regardless of which player won the bigger/smaller amount;
pairwise reciprocity rate: defined as the pair taking turns in getting the bigger reward.
Details of pairing for coherence analyses are as follows:
  1. Within-group Interacting Pairs (WIPs) . The WIPs are pairs within the same group and are interacting at the same time. For example, when Player A is interacting with Player B (i.e., A&B) and Player C is interacting with Player D (i.e., C&D) in Block 1, A&B and C&D are two interacting pairs. For a given experimental group (3 scanned + 1 behavioral), there are 3 players X 2 pairings X 26 groups =156 combinations.
  2. Within-group Non-interacting Pairs (WNPs) . To use the A&B/C&D pairings above for example, A&C is the non-interacting pair (for Player D is outside the fMRI, thus with no neural data). For a given experimental group, there are 3 players X 4 pairings X 26 groups = 312 combinations.
  3. Between-group Permuted Pairs (BPPs) . BPPs are pairings of players from different groups, e.g., A from Group 1 to B/C from Group 2, etc. (3 players X 50 pairings X 26 groups = 3,900 combinations).
Independent two-sample t-tests were carried out for various ROIs on the three pairings (WIPs, WNPs, and BPPs) to determine whether they were different distinct populations.
RESULTS
Behavioral results
In the individual level, the reciprocity index, the mean percentage of successful trials with the winner of the bigger reward alternated was 56.47± 17.8 % (the reciprocity index), the mean percentage of successful trials without reciprocity was 4.80± 4.18 % (the greed index), and the average sum of money each participant earned in the whole experimental session was $290.1± 44.1 NTD.
The results of the pair-based indices indicate\sout averagely 76.4% (=27.5/ out of 36, or 6 in 6 runs) of the trials the pairs successfully won the reward (the pairwise success rate), and \soutaveragely 73.8% (=25.8/35 possible alternations from each dyad) the pair took turns in getting the bigger reward (pairwise reciprocity rate). The pairwise success rate was significantly correlated with the pairwise alternation rate (r = 0.921, p < 0.001), which suggests that the majority of players soon learned to coordinate with three others, without being given explicit instructions and being strangers throughout (i.e., participants were drawn from three different universities). These also reflect the effectiveness of the study design, which enabled the participants to follow their natural tendencies.
The homogeneity checks of tri-fMRI data
As a sanity check, the triad-fMRI data from the three sites (NCKU with GE MR750, NTU with Siemens Prisma, and NCCU with MAGNETOM Skyra), after the same preprocessing steps and coherence preparations \sout(Figure 2), were compared in the seed-rTPJ betas and the rTPJ-rTPJ coherence values (Supplementary Figure S1). The results show that these data were highly overlapping on the major dependent measures. In other words, despite different raw image values and scanning parameters from each vendor, the three sites’ fMRI data remained homogenous and comparable, ready for the interpersonal coherence analyses.
GLM contrasts and PPI analyses
As the first pass of most fMRI studies, the GLM contrast between the successful and failed reciprocal trials, on the 2nd feedback phase, were identified and compared. Here, only the stage of revealing the final result (i.e., Stage 2 Feedback) was chosen as the dyads shared this period to see whether the collaboration/coordination was (reciprocally) successful for reward. This feedback stage helped the dyads with trust-building for the following trials. Two additional reasons are as follows: a) collinearity and b) the extension and replication of our previous study (Wang et al., 2023). In our experiment design, the events of the parameters estimated were fixed. These highly correlated events (i.e., Stage 1 Decision is always followed by Stage 1 Feedback, then Stage 2 Decision, and finally Stage Feedback 2 Feedback) might result in being biased. Consequently, the GLMs were separate for each stage (only ‘Stage 2 feedback’ was adopted in the following analysis), instead of a single GLM run with all 4 onsets. Second, we adopted Stage 2 Feedback in our previous analysis for coherence. This study was aimed to extend and replicate our previous one; thus, the same period of time was employed for further analysis.
Figure 3a represents the brain regions with significant activations in the [Successful > Failed trials] contrast. As shown in the Supplementary Table S2, the contrast [Successful trials > Failed trials] in the second stage of feedback seven activated clusters are identified, including the right Precentral Gyrus (rPG), the right Posterior Cingulate, the right Caudate, the left Middle Temporal Gyrus, the left Middle Frontal Gyrus, the left Medial Frontal Gyrus, and the left Inferior Frontal Gyrus. To examine the brain-behavior relationship, we calculated the correlation between the three indices (i.e., greed, and total reward) and the [Successful trials > Failed trials] contrast. No ROI was found correlated.
In addition to GLM contrast analysis, the rTPJseed downloaded from Neurosynth.com was used as the seed region to examine the target regions where their functional connectivity was significantly stronger in the successful vs. failed reciprocity trials. In the second feedback phase, the bilateral Precentral Gyri, bilateral Inferior Frontal Gyrus, bilateral Insula, right Medial Frontal Gyrus, left Inferior Parietal Lobule, right Cingulate Gyrus, left Middle Frontal Gyrus, right Postcentral Gyrus and left Precuneus were identified as more negatively functionally connected with rTPJ in the successful trials than in the failed trials (see the Supplementary Table S3). The target regions were also mapped in Figure 3b. To examine the brain-behavior relationship, we calculated the correlation between the three indices (i.e., reciprocity, greed, and total reward) and the [Successful trials > Failed trials] contrast. Only the right Sub-gyral was found correlated with reciprocity (see the supplementary Table S4).
We further employed connectivity-based correlations with individual total reward. To investigate the underlying network differences with their individual total reward in the successful vs. in the failed trials, the rTPJseed connectivity was further examined. The results showed negative functional connectivity in the right postcentral gyrus, right Angular Gyrus, and right Inferior Parietal Lobule when the participants successfully reciprocated with their pairmates (see Figure 3c and the Supplementary Table S5).These results indicated that the lower the functional connectivities between the hub areas of different sub-networks, such as attention/memory and self-other awareness areas between rTPJ-rIPL and -rAG, the higher reproductive success.