Reaction Time
The results of RTs revealed neither a significant main effect of learning phase (β = -1.87, SE = 1.05, t418 = -1.79, P = .075) nor a significant main effect of context (β = 7.58, SE = 13.11, t80 = -0.58, P = .565). However, the two-way interaction effects in learning phase × context were significant (β = -3.50, SE = 1.28, t418 = -2.73, P = .007). This interaction demonstrated that the slope of RT decrease (i.e., the behavioral learning rate) was steeper for the joint context compared to the baseline context (see Fig. 2A). These results indicate that, in comparison to the baseline context, the joint context exhibited a facilitation effect on sequence learning.
Furthermore, two models were built to account for the behavioral performances in the modulating phase. The results of the quadratic model revealed a significant main effect of context (β = -68.02, SE = 21.13, t229 = -3.33, P = .001), a two-way interaction between context and phase (β = 66.66, SE = 20.37, t164 = 3.27, P = .001), and, notably, a two-way interaction between context and phase^2 (β = -15.72, SE = 5.04, t164 = -3.12, P = .002). However, the results of the linear model revealed neither a significant main effect nor an interaction (ts < 1.34, Ps > .182). Furthermore, model comparison supported the quadratic model (AIC = 2441.3, BIC = 247), which had a significantly better fit than the linear model (AIC = 2458.2, BIC = 2486.5, χ2 = 18.98, P < .001; see Fig. 2A).
For the interference effect, the results revealed neither a significant main effect of phase (β = 6.36, SE = 4.78, t82 = 1.33, P = .187), nor a significant main effect of context (β = 2.42, SE = 11.93, t73 = 0.20, P = .840). However, the two-way interaction effects in phase × context were significant (β = -19.50, SE = 5.85, t82 = -3.33, P = .001). A follow-up simple effect analysis showed that the RT was significantly larger in Block 7 (M = 425 ms, SE = 6.95) than in Block 6 (M = 412 ms, SE = 6.95) for the joint context (β = 13.14, SE = 3.38, t = 3.89, P < .001), but that there were no significant differences in RT observed between Block 6 (M = 429 ms, SE = 9.70) and Block 7 (M = 423 ms, SE = 9.70) in the baseline context (β = -6.36, SE = 4.78, t = -1.33, P = .187).
For the recovery effect, the results revealed neither a significant main effect of phase (β = -0.34, SE = 4.79, t82 = -0.07, P = .944) nor a significant main effect of context (β = 2.42, SE = 12.03, t73 = 0.20, P = .841). However, the two-way interaction effects of phase × context were significant (β = -11.94, SE = 5.87, t82 = -2.04, P = .045). The follow-up simple effect analysis showed that the RT was significantly larger in Block 7 (M = 425 ms, SE = 6.95) than in Block 8 (M = 413 ms, SE = 6.95) for the joint context (β = 12.28, SE = 3.39, t = 3.63, P < .001), but no significant differences in RT were observed between Block 7 (M = 423 ms, SE = 9.83) and Block 8 (M = 423 ms, SE = 9.83) in the baseline context (β = 0.34, SE = 4.79, t = 0.07, P = .944). Altogether, these results demonstrate an enhanced learning rate in the joint context arising from the facilitative influence of acquiring the underlying statistical regularities.
P3
The trial type similarity analysis on P3 revealed a significant main effect of trial type (β = -2.95, SE = 0.33, t82 = -8.95, P < .001), with self-relevant trials exhibiting a significantly higher P3 amplitude than self-irrelevant trials. However, there were no significant main effects of context (β = 0.18, SE = 0.38, t120 = 0.48, p = .631) nor interactions between trial type and context (β = -0.57, SE = 0.40, t82 = -1.40, P = .164, see Supplementary Fig. 3C). These findings suggest that participants had distinct neural activity between trial types, and therefore the subsequently analyses were conducted separately.
Regarding self-relevant trials (see Fig. 2B & 2C), the analysis of the learning phase on P3 amplitude revealed neither a significant main effect of phase (β = -0.06, SE = 0.04, t418 = -1.44, P = .151) nor a significant main effect of context (β = 0.48, SE = 0.50, t95 = 0.96, P = .341). However, the two-way interaction effects of phase × context were significant (β = -0.11, SE = 0.05, t418 = -2.09, P = .038). This was indicated by the slope of the P3 amplitude decrease, or the rate of the increased P3 amplitude, being steeper in the joint context compared to the baseline context. Furthermore, the results of the quadratic model revealed a significant main effect of context (β = -33.62, SE = 12.22, t164 = -2.75, P = .007), the two-way interaction between context and phase (β = 9.50, SE = 3.52, t164 = 2.69, P = .008), and, notably, the two-way interaction between context and phase^2 (β = -0.66, SE = 0.25, t164 = -2.62, P = .010). However, the results of the linear model revealed neither a significant main effect nor an interaction effect (ts < 1.88, Ps > .062). Furthermore, the model comparison results revealed that the quadratic model (AIC = 900.9, BIC = 932.6) exhibit a significant fit compared to the linear model (AIC = 912.2, BIC = 936.9, χ2 = 15.30, P < .001) for the modulating phase data.
For the interference effect in the self-relevant trials, the results revealed neither a significant main effect of phase (β =-0.17, SE = 0.26, t82 = -0.68, P = .499) nor a significant main effect of context (β = -0.31, SE = 0.52, t82 = -0.60, P = .552). However, the two-way interaction effects of phase × context were significant (β = 0.94, SE = 0.31, t82 = 3.01, P = .004). The follow-up simple effect analysis showed that the P3 amplitude was significantly larger in the Block 7 (M = 3.81, SE = 0.31) than in Block 6 (M = 3.04, SE = 031) for the joint context (β = 0.77, SE = 0.18, t = 4.25, P < .001), but no significant differences in RT were observed between Block 6 (M = 3.35, SE = 0.41) and Block 7 (M = 3.18, SE = 0.41) in the baseline context (β = 0.17, SE = 0.26, t = 0.68, P = .499). However, there were no significant main effects nor interaction effects seen for the recovery effect in the self-relevant trials (ts < -1.37, Ps > .173).
For self-irrelevant trials, there were no significant learning effects (ts < 1.90, P > .058) nor modulating effects (ts < 1.18, Ps > .240 for the linear model; ts < 0.88, Ps > .381 for the quadratic model) on the P3 amplitude. These findings indicate that neural activity during sequence learning, as indexed by P3, exhibits enhanced dynamics specifically for self-relevant regularities in the joint context.
Intra-brain Functional Connectivity (FC)
For the theta band, as FC results revealed neither a main effect of trial type nor of the two-way interaction of trial type and context (ts < 1.33, Ps > .186, see Supplementary Fig. 3D), a series of LMMs were performed for all electrode pairings using the averaged PLVs. The results revealed significant main effects of the phase (9 channel combinations, all PFDR < .05) and phase × context interactions (7 channel combinations, all PFDR < .05; see Fig. 2D). Further analyses of the channel combinations showing interaction effects revealed that the slopes of the intra-brain PLV decrease, or of the intra-brain PLV learning rate, were steeper in the joint context as compared to the baseline context (β = -2.85×10− 3, SE = 1.30×10− 3, t418 = -2.19, P = .029; see Fig. 2D). However, the results revealed only a significant effect of phase^2 (β = 0.01, SE = 0.01, t164 = 2.16, P = .032) and of phase (β = -0.15, SE = 0.07, t164 = -2.16, P = .032) during the modulating phase. We did not observe other significant effects during the modulating phase (ts < 0.14, Ps > .887 for the linear model; ts < 0.46, P > .637 for the quadratic model).
Brain-to-Brain Coupling Analysis
In the first step, a total of 3,019 channel combinations were identified where the PLV significantly exceeded the averaged surrogate PLVs in the theta and alpha bands (all PFDR < .05). Subsequently, in the second step, a series of LMMs were conducted on the channel combinations that remained after the first step (see Fig. 3A).
For the theta band, analysis results revealed significant main effects of phase and, notably, their interactions in 31 channel combinations (see Fig. 3B), all PFDR < .05. Further analysis on the channel combinations showed interaction effects demonstrated by the slope of BtBC increase, or of BtBC learning rate, being steeper in the joint context as compared to the baseline context (phase × context: β = -6.69×10− 3, SE = 9.14×10− 4, t278 = -7.31, P < .001). We then investigated the impact of statistical regularities on self–other integrations based on the aforementioned identified electrode pairings. The results of the quadratic model revealed a significant main effect of context (β = 0.73, SE = 0.21, t108 = 3.43, P = .001), the two-way interaction between context and phase (β = -0.20, SE = 0.06, t108 = 3.30, P = .001) and, notably, the two-way interaction between Context and Phase^2 (β = 0.01, SE = 0.01, t108 = 3.23, P = 0.002). However, the results of the linear model revealed a significant main effect of context (β = 0.04, SE = 0.02, t125 = 2.32, P = .022). Furthermore, the model comparison results revealed that the quadratic model (AIC = -860.5, BIC = -835.5) exhibited a significant fit compared to linear model (AIC = -854.2, BIC = -835.4, χ2 = 10.36, P = .006) for the modulating phase data (see Fig. 3C).
Corroborating with RT analysis, we also performed two LMMs for interference and recovery effects on PLV values. For the interference effect, where the statistical regularity shifted from presence to absence, the LMM included context (i.e., Joint vs. Baseline) and phase (Block 6 vs. Block 7) as fixed effect, along with their interaction. The results revealed a main effect of Phase (β = 0.01, SE = 0.003, t54 = 2.93, P = .005), a significant main effect of context (β = 0.13, SE = 0.03, t57 = 4.24, P < .001), and notably, two-way interaction effects of phase × context (β = -0.02, SE = 0.01, t54 = -3.92, P < .001). The follow-up simple effect analysis showed that the PLV values were significantly larger in Block 6 (M = 0.109, SE = 0.004) than in Block 7 (M = 0.100, SE = 0.004) for the joint context (β = 0.01, SE = 0.003, t = 2.62, P = .011), but the PLV values were significant fewer in Block 6 (M = 0.087, SE = 0.004) than in Block 7 (M = 0.097, SE = 0.004) for the baseline context (β = -0.01, SE = 0.003, t = -2.93, P = .005). However, there were no significant main effects or interaction effects for the recovery effect (ts < -1.82, Ps > .074).
To further elucidate the significance of electrode pair interactions within the theta band, we conducted a Non-Negative Matrix Factorization (NMF) analysis for each context. The results showed that the two-cluster solution (see Fig. 3E and G) exhibited the best fit in both joint (see Fig. 3D) and baseline contexts. Furthermore, we found that the rate of increased BtBC within NMF-derived Cluster 1 correlated negatively with the behavioral learning rate (r = 0.39, P = .043), indicating that an increase in BtBC as represented by Cluster 1 predicted a better statistical learning performance (see Fig. 3F). Meanwhile, the rate of increased BtBC within NMF-derived Cluster 2 correlated positively with the rate of increased intra-brain functional connectivity (r = 0.44, P = .021; see Fig. 3H), indicating that an increased BtBC within Cluster 2 could predict increased intra-brain functional connectivity.