A multi-brain mechanism for observational threat learning

doi:10.21203/rs.3.rs-2215515/v1

Download PDF

Article

A multi-brain mechanism for observational threat learning

https://doi.org/10.21203/rs.3.rs-2215515/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Survival and adaptation in environments require swift and efficacious learning about what is dangerous. Across species, much of such threat learning is acquired socially, for example, through the observation of others’ (“demonstrators’”) defensive behaviors. However, the specific mechanisms responsible for the integration of information flowing between demonstrators and observers remain largely unknown. We addressed this dearth of knowledge by sequentially performing magnetoencephalography (MEG) imaging in demonstrator-observer dyads: a set of stimuli were first shown to a demonstrator whose defensive responses were filmed and later presented to an observer, with neuronal activity recorded from both individuals. Observers exhibited successful learning, as revealed by physiological responses and computational modeling. Sensor- and source-level results consistently demonstrated brain-to-brain coupling (BtBC) within demonstrator-observer dyads. Strikingly, BtBC in the fronto-limbic circuit (including insula, ventromedial and dorsolateral prefrontal cortex) predicted ensuing learning outcomes (i.e., conditioned responses). A machine learning algorithm revealed that the predictive power of BtBC on learning was magnified when a threat was imminent to the demonstrator. BtBC depended on how observers perceived their social status relative to the demonstrator, and was likely to be driven by shared attention and emotion, as bolstered by dyadic pupillary coupling. Taken together, our study describes a multi-brain mechanism for social threat learning, involving BtBC, which reflects social relationships and predicts adaptive learned behaviors.

Biological sciences/Neuroscience/Social neuroscience/Empathy

Biological sciences/Neuroscience/Social behaviour

observational threat learning

vicarious fear

brain-to-brain coupling (BtBC)

Magnetoencephalography (MEG)

social status

Across many species, successful adaptation to the environment requires learning from conspecifics about what should be avoided and approached ^1,2. For example, an observer can quickly learn to associate a neutral stimulus with threat by observing another individual’s (a “demonstrator’s”) defensive responses to the stimulus ^3–5. This form of observational or vicarious threat learning has been documented in many non-human species ^6–10 and across the human life-span, from infancy ^11,12 to adolescence ^13,14 and adulthood ^{4, 15–18}, serving as an important experimental model to understand social transfer of value information that can be generalized more broadly.

Several key mechanistic processes subserving such observational threat learning has been captured by research in social, behavioral and affective neuroscience. For instance, rodents and human studies converge on the conclusion that the amygdala, anterior insula (aINS), and anterior cingulate cortex (ACC) are involved in social threat learning ^19–21. Most of this research has, however, focused on neural mechanisms in observers, leaving it unexamined how neural systems coordinate across demonstrators and observers engaged in learning ²². Recent efforts leveraging functional near-infrared spectroscopy (fNIRS) and functional MRI (fMRI) have revealed brain-to-brain coupling (BtBC, i.e., the neural systems of two brains that are temporarily coupled ²³) between individuals during skill^24,25 and reward learning ²⁶. But due to limitations in temporal resolution, previous neuroimaging studies have not been able to specify the temporal unfolding of BtBC during observational threat learning. BtBC represents a proxy for correlated neural activity across the brains of two individuals ^{23, 27–29}, and emerges when e.g. dyads’ attentions or emotions are shared ^30–32. Unveiling the temporal dynamics of BtBC will help us understand the real-time information transfer process during social learning, and eventually aid the development of new strategies to treat learned maladaptive behaviors.

In this study, we aimed at examining the specific multi-brain mechanisms underlying successful observational threat learning. For this purpose, we used magnetoencephalography (MEG) imaging to record neuronal activity in observers learning the threat value of stimuli through watching demonstrators’ aversive reactions to the same stimuli. Demonstrators’ brain activity was pre-recorded, allowing for computations of BtBC across demonstrator-observer dyads. Compared to other approaches (e.g., simultaneous imaging ^22,33), this MEG-based sequential imaging approach allowed for a balanced trade-off between ecological validity and experimental control ³⁴ in the study of information transfer in dyads. Importantly, MEG shows superior utility in its temporal precision, and enabled us to capture the unfolding of time-locked neural responses in two individuals during social learning localized with high spatial resolution to the individual brain anatomy ³⁴.

Because the relative status between individuals is important for how they interact and exchange information ^35–37, we expected relative status to impact the learning process in our paradigm. According to the “low-status benevolence” hypothesis ³⁸, observers are expected to show stronger empathy (a core constituent of observational threat learning ^19,41,42) with, pay more attention to, and consequently learn better from, low-status (vs. high-status) demonstrators. In support of this conjecture, social status has been reported to modulate neural responses of empathy for someone receiving painful stimulation ³⁸. More specifically, activation was observed in regions associated with empathy: the anterior insula and anterior medial cingulate cortex in response to low-status targets, whereas activation in these regions was attenuated in response to high-status targets. Moreover, people tend to inhibit emotional sharing or even express schadenfreude towards misfortune high-status individuals ^39,40. Alternatively, according to the “high-status skilfulness” hypothesis ^43–46, observers should, conversely, pay more attention to, and learn better from, high-status (vs. low-status) demonstrators, as these demonstrators are assumed to be more skilful and reliable. This hypothesis is supported by research demonstrating that high status have a great social influence and therefore attract more attentional resources compared to those of relatively low status ^37,47. Naïve observers might trust a skilled (vs. unskilled) demonstrator more to accurately and reliably show adaptive behaviors to threat (e.g., facial expressions and body movements) that reflect the underlying contingency between reaction and consequence, impacting upon the efficiency of observation by learners ^43,48,49.

The goal of the current study was twofold. First, related to our primary aim, we examined the neural mechanisms (here, BtBC) involved in the transfer of threat information between demonstrators and observers, and how these mechanisms predict successful learning. Given that BtBC can remarkably reflect successful information transfer between individuals ^50,51, we expected an enhancement of BtBC during observational acquisition of threat would predict learning outcome (i.e., conditioned responses). In accordance with previous research on observational learning, we expected an enhanced BtBC in brain regions associated with action observation (such as motor areas ⁵²), reward processing (such as ventromedial prefrontal cortex, vmPFC ¹⁸, anterior insula, aINS ³, and dorsolateral prefrontal cortex, DLPFC ¹⁶), and social perception and cognition (such as superior temporal sulcus, STS ^3,5), and primarily engage brain oscillations within low-frequency bands (such as delta/theta bands, 1 to 8 Hz) that have been implicated in aversive learning ^53–57.

Second, we tested whether social status would bias observational learning and the associated BtBC. Prior research has shown that a higher level of BtBC is associated with better learning ^{22,24, 58–61}. To the extent that the “low-status benevolence” hypothesis holds true, we expected stronger learning-relevant BtBC (expected to characterize threat information transfer from demonstrators to observers) when observers believed they were watching a low-status demonstrator compared to watching high-status demonstrator. Alternatively, if the “high-status skilfulness” hypothesis holds true, we expected stronger learning-relevant BtBC when observers believed they were watching a high-status demonstrator compared to watching a low-status demonstrator.

To advance a better understanding of the functional meaning of BtBC during social learning, a machine learning algorithm was employed to provide a more precise computation for predictive power of BtBC on observational learning outcome across time (as compared to a conventional correlational analysis based on averaged BtBC). Gaze fixation and pupil dilation pattern as major types of adaptive behavior were used to test shared attention and emotion ^30,31 as sources for BtBC (i.e., whether BtBC was mechanistically driven by shared attentional effort and emotional processing between demonstrator and observer).

To investigate the multi-brain mechanism for observational learning, we developed a sequential MEG imaging protocol ³⁴. First, we filmed demonstrators and recorded MEG while participants were performing a Pavlovian threat conditioning task (N = 3, Pavlovian learning), where one conditioned stimulus, CS+, but not the other, CS-, was probabilistically (62.5%) followed by electrical nerve stimulation (i.e., unconditioned stimulus, US). The demonstrators were regular, naïve participants with naturalistic reactions (i.e., not actors; see an example video online). All demonstrators showed successful learning during acquisition in the Pavlovian learning (Fig. S1). Second, we showed these video recordings to naïve observers (N = 60, observational learning) during MEG scanning (Figs. 1A-B).

Before observational learning, social status was manipulated via performance payoffs in a math competition task ⁶²: half of the observers were instructed to believe they were observing a higher-status (HS) demonstrator (N = 30), and the other half, a lower-status (LS) demonstrator (N = 30, Fig. 1C). The influence of status manipulation on subsequent observational learning could be contaminated by task familiarity, which motivated us to have the tasks dissimilar (i.e., math competition vs. threat learning). Having a task that is not directly related also provides an even stronger test of the hypotheses that exogenously induced social status would affect learning. Similar to previous experiments ^62,63, this manipulation successfully induced feelings of social rank (see Methods). After observational learning, observers performed a direct test in the absence of the demonstrator to evaluate their learning outcomes [i.e., CS+/CS- differentiation on skin conductance response (SCR) and pupillometry]. Demonstrator and observer visits were on separate days (see Fig. 1D for the experimental protocol and Methods for more details). Each demonstrator was paired with about 20 observers, among which 10 observers were randomly assigned to the LS group and the other 10 observers to the HS group.

2.1. Successful observational learning in the sequential MEG imaging paradigm

A linear mixed-effects model tested and verified successful learning in naïve observers: observers showed significantly higher SCR amplitude for CS + compared to CS- trials (β = 0.39, SE = 0.07, t = 5.97, P < 0.001; Fig. 2A, left panel). A similar pattern was observed for CS+/CS- differentiation on pupil response (β = 0.45, SE = 0.09, t = 4.85, P < 0.001; Fig. 2A, right panel). No significant main effect of social status or interaction were detected (all t < 0.66, all P > 0.51).

To further uncover the computational basis for observational threat learning, we then used a simple Rescorla-Wagner reinforcement learning model (winning model after model comparison, see Methods) to quantify the SCR patterns ⁶⁴. We focused on the model-derived trial-by-trial initial value (V₀), which accounted for initial shock prediction during direct test, and the learning rate (α), which elucidated the weight of reward prediction error in value (shock) update during observational learning. Modeling results on initial value revealed main effects of CS type (β = 0.14, SE = 0.002, t = 73.93, P < 0.001) and social status (β = 0.03, SE = 0.01, t = 4.27, P < 0.001), as well as their interaction (β = -0.04, SE = 0.003, t = -15.76, P < 0.001), indicating a significant larger V₀ for “CS + _LS > CS-_LS” compared to “CS + _HS > CS-_HS” (Fig. 2B, left panel). The learning rate during observation learning showed main effects of CS type (β = 0.10, SE = 0.005, t = 19.76, P < 0.001) and social status (β = -0.08, SE = 0.01, t = -9.34, P < 0.001), and importantly, the interaction between them (β = -0.20, SE = 0.01, t = -27.90, P < 0.001). Post-hoc comparisons indicated that learning rate was significantly higher for CS + than CS- in the LS group, whereas this pattern was reversed in the HS group (Fig. 2B, right panel). Taken together, our physiological results demonstrated that naïve observers consistently acquired threat information irrespective of social status. Computational modeling further uncovered that status did, however, bias the initial value during direct test, as well as the learning rate for the extent of value (shock) update during observational learning.

2.2. Observational learning decreased widespread low-frequency neural response in observers

To investigate neural activity supporting observational threat learning, we first performed a sensor-level time-frequency analysis on observers’ MEG data (a parallel analysis on demonstrators’ brain data can be found in Fig. S2, albeit no statistical comparisons being applied due to the limited sample size of demonstrators). The non-parametric cluster-based permutation tests showed a significant difference between CS + vs. CS- (P = 0.001), informed by a negative cluster with the frequency ranging from 1–8 Hz and time interval ranging from 0.6–5.5 s post stimulus onset at the fronto-temporo-parietal channels (Fig. 3A-B). Source-level power mapping, using dynamic imaging of coherent sources (DICS), indicated that the CS effect (i.e., the 1–8 Hz response from 0.6–5.5 s) involved a decreased power in the ventromedial prefrontal cortex (vmPFC), occipital cortex (OCC), dorsolateral prefrontal cortex (DLPFC), Postcentral gyrus (PoG), and posterior superior temporal sulcus (pSTS) in low-frequency bands (including delta and theta bands, Fig. 3C).

In sum, our results describe a neural signature of observational learning that includes low-frequency neural oscillations. Similar low-frequency activity has been demonstrated in aversive learning and empathy ^53–57. Time-frequency analyses did not reveal any significant clusters for the main effect of social status or the interaction (all P > 0.13).

2.3. Shared neural responses between demonstrators and observers

We next examined whether shared neural responses between demonstrators and observers could reflect social status and predict learning outcomes. We used the well-established Circular Correlation Coefficient (CCorr; see Methods) to measure shared neural responses (i.e., BtBC) between demonstrators and observers. We focused on low frequencies of interest associated with observational threat learning (as detected by the time-frequency analyses in the previous section) divided into the canonical delta (1–3 Hz) and theta bands (4–8 Hz).

In the first step of the analysis, we compared BtBC during observational learning with BtBC during visual baseline. This analysis filtered out channel combinations (i.e., demonstrator channel paired with observer channel) where BtBC might simply emerge due to common visual inputs and/or environment. We identified 6281 (Fig. 4A) and 4330 (Fig. S3A) channel combinations where BtBC during observational learning significantly exceeded that during visual baseline in the delta and theta bands, respectively (all P_FDR < 0.05).

In the second step, we performed a series of linear mixed-effects models on channel combinations that survived in the first step. For the delta band, the analysis revealed a series of main effects of CS type (2959 channel combinations, all P_FDR < 0.05) and social status (8 channel combinations, all P_uncorrected < 0.001; Fig. 4B), and notably, their interaction (13 channel combinations, all P_uncorrected < 0.001; Fig. 4C, left panel). Further analyses on channel combinations which showed interaction effects revealed that CS + _LS elicited significantly stronger mean BtBC relative to CS-_LS (CS type × social status: β = -0.61, SE = 0.13, t = -4.83, P < 0.001; Fig. 4C, right panel). For the theta band, we found a series of main effects of CS type (630 channel combinations, all P_FDR < 0.05), indicating that mean BtBC was significantly larger for CS + than CS- trials (β = 0.16, SE = 0.02, t = 9.85, P < 0.001; Fig. S3B). No other effects were observed. As a control, we conducted parallel analyses in the alpha (9–12 Hz) and beta bands (13–30 Hz), neither band of which was among the a-priori (low-frequency) bands. No significant results were detected when we compared BtBC across CS type and social status, after multiple comparison corrections in these bands. In the remainder of the paper, we henceforth restricted our analyses to the low-frequency bands.

2.4. Shared neural responses selectively predicted observational learning

To further examine the interaction effect in the delta band and to test the relationships of BtBC patterns for each social status condition with learning, we conducted a nonnegative matrix factorization (NMF) analysis on BtBC (Fig. 5A). NMF clustered BtBC describing the unique demonstrator-observer brain network during observational learning ³². Previous research has shown that responses to CS + during observational learning can successfully predict the expression of learning at a subsequent time, as indicated by converging strands of physiological ^41,65,66 and neural evidence ⁶⁷. In line with these findings, we found that responses to the CS + _LS condition in NMF-derived cluster 1 correlated positively with differential SCR (r = 0.44, P = 0.02) and differential learning rate (r = 0.47, P = 0.01), indicating that an increase in BtBC as represented by cluster 1 predicted better learning (Figs. 5B-C). Clusters 2 and cluster 3 derived from NMF were not correlated with differential SCR (all |r| < 0.10, P > 0.63) or differential learning rate (all |r| < 0.04, P > 0.85). In the CS + _HS condition, clusters 1–3 were not correlated with differential SCR (all |r| < 0.25, P > 0.21) or differential learning rate (all |r| < 0.18, P > 0.36).

2.5. Source-level shared neural responses associated with social status predicted learning

To examine the neural substrates of the multi-brain mechanism underlying observational threat learning, we next sought to uncover shared neural responses in brain areas associated with learning and social status. We determined our regions of interest (ROI) based on DICS power mapping in this study and previous work ³. Following this, we selected ROIs, including insula (INS), anterior cingulate cortex (ACC), vmPFC, DLPFC, PoG, pSTS, and OCC, in both hemispheres. MEG time series in source space over these ROIs were extracted and submitted to the BtBC analyses.

We conducted the BtBC analyses in the delta band. For channel combinations where BtBC during observational learning was significantly stronger than that during visual baseline (Fig. 6A), linear mixed-effects models were conducted for each possible channel combination and revealed a series of main effects of CS type (56 ROI combinations, all P_FDR < 0.05; Fig. 6B, left panel), indicating that there was significantly stronger mean BtBC in CS + compared to CS- trials (β = 0.13, SE = 0.02, t = 5.46, P < 0.001; Fig. 6B, middle panel). A planned contrast on mean BtBC revealed a marginally significant difference between “CS + _HS > CS-_HS” and “CS + _LS > CS-_LS” (independent t-test, t = 1.98, P = 0.052; Fig. 6B, right panel). No main effect of social status or interactions were detected after FDR correction on either channel combination. Corresponding analyses were conducted also in the theta band (see Fig. S4), with no significant results involving social status or CS type.

We then carried out a series of Pearson correlations to examine in which ROIs BtBC significantly predicted observers’ learning outcome. In the CS + _LS condition, we observed that BtBC at lINS_rINS (i.e., demonstrator’s left insula and observer’s right insula), lvmPFC_rINS, rvmPFC_rINS, and lDLPFC_rINS during observational learning consistently predicted differential pupil responses in observers in the subsequent direct test (all r > 0.56, P_FDR < 0.05; Fig. 6C). No significant relationships between BtBC and differential SCR were detected (all P_FDR > 0.63). In the CS + _HS condition, no significant correlations between BtBC and learning were found after FDR correction (all P_FDR > 0.35). Furthermore, BtBC at lDLPFC_rINS also correlated with observer’s self-reported empathic concern (measured by Interpersonal Reactivity Index⁶⁸) in the LS group (r = 0.41, P = 0.03), but not in the HS group (r = -0.01, P = 0.95).

2.6. Shared attention and emotion as likely sources for BtBC predicting learning outcome

Having established that BtBC in the fronto-limbic circuit can predict learning outcome (as measured by differential pupil response, see the previous section), we further investigate how early that learning outcome at direct test could be decoded from source-level BtBC at observational learning. To this end, we conducted time-varying support vector regressions (SVR). Indeed, both empirical ^69,70 and theoretical ⁷¹ accounts of threat learning have described its temporal dependency in relationship to predictive events. Therefore, capitalizing on the uniquely high temporal resolution of MEG, rather than averaging BtBC across time, we repeatedly predicted trial-averaged learning outcome (i.e., pupil response) in the direct phase based on cumulative BtBC in the CS + _LS condition for each time point in the observational learning phase. This analysis would inform whether BtBC could predict learning outcome shortly after CS onset or when a threat is imminent (i.e., close to the US). We observed that the prediction performance was improving and reached significance starting at about 4.7 s after CS onset, immediately preceding the US (note that a US was expected to come 5.5 s post CS onset; all P_FDR < 0.05, Fig. 7A). The prediction accuracy of the model was expressed by the Pearson correlation coefficient between the actual and predicted values ^72,73. The mean absolute errors (MAEs) were also reported. Better fit of the prediction model can be characterized by a higher value of correlation coefficient and a lower value of MAE. Our results showed that, the correlation coefficients between predicted and actual values were larger than 0.49, with MAEs smaller than 0.28. We performed a parallel analysis in the CS + _HS condition. No significant predictions were found after FDR correction (all P_FDR > 0.16). These findings indicate that BtBC data are able to predict learning outcome for the LS condition (but not HS) when a threat is imminent to the demonstrator.

To better understand the functional meaning of the BtBC for the prediction of learning outcome in the LS group, we carried out two complementary analyses. First, to ascertain how the observers allocated their attentional focus, we parsed the fixation proportion (i.e., fixation time at each area of interest normalized to the total fixation time) during observational learning. A linear mixed-effects model demonstrated that the observers pre-dominantly paid attention to the area including the demonstrator’s right hand, which was attached to the stimulator that administered the electric stimulation during CS + vs. CS- trials (β = 0.48, SE = 0.05, t = 10.52, P < 0.001; Fig. 7B), in the time window of 4.7–5.5 s post CS onset. Other areas of interest did not show this attentional bias (Face: β = -0.04, SE = 0.02, t = -1.87, P = 0.07; Stimulus: β = -0.003, SE = 0.006, t = -0.43, P = 0.67; Fig. 7B). Second, we computed eye-to-eye coupling (a corollary of shared attention ⁷⁴ and emotion ³¹) between demonstrator and observer in the same time window for both CS + and CS- trials. Here, eye-to-eye coupling was defined based on the similarity between demonstrator’s and observer’s pupil dilation that is thought to typically track attentional effort ⁷⁵ and sensitive to emotional peaks ³¹. Notably, eye-to-eye coupling was significantly stronger in the CS + compared to CS- trials (β = 0.54, SE = 0.15, t = 3.83, P < 0.001; Fig. 7C). These exploratory analyses indicate a higher level of shared attentional effort and emotional response between observers and demonstrators on CS + vs. CS- trials when observers believed they were watching a LS demonstrator.

The present study used a multi-brain MEG-imaging approach to understand the neural mechanisms underlying learning through observation, and its dependency on social status. We showed that following the observation of a demonstrator’s defensive threat responses, naïve observers successfully acquired, and later expressed, learning (indexed by mean SCR and pupillometry) in the absence of the demonstrator. Although the impact of social status was not visible in these average-based indices of learning, it was manifested in two important computational properties: initial value and learning rate. On the neural level, we found that CS + compared to CS- conditions downregulated widespread delta and theta oscillatory responses in the observers when watching the demonstrator. Importantly, interbrain analysis identified significantly increased low-frequency BtBC during observational acquisition of threat compared to the visual baseline for the demonstrator-observer dyads, suggesting that the alignment of neural processes could not be explained by shared environment or stimulus inputs. Such BtBC in the delta band was greater for CS + vs. CS- trials in the LS group but not in the HS group. Strikingly, the social status-dependent BtBC was also predictive of observers’ ensuing learning outcome (as indicated by SCR and pupillometry) and the predictive power of BtBC was remarkably amplified after about 4.7 s relative to CS onset (immediately preceding the US). Further analysis on gaze fixation and eye-to-eye coupling revealed that BtBC predicting learning outcome was likely to be driven by shared attention and emotion. In the following, we discuss our five key findings.

First, our peripheral physiological results validated successful threat learning (CS + > CS-) in naïve observers in a novel sequential MEG imaging paradigm. Compared to conventional video-based social learning paradigm ⁵, the use of naïve demonstrators in our paradigm enhanced ecology validity without compromising experimental rigor ³⁴. Moreover, the inclusion of social status, added to the realism by introducing an important social variable present in some forms in most social encounters. Our results showed that social status biased the learning rate during observational acquisition and the initial value in subsequent direct test. These observations suggest that perceived social relationships might affect the fine-grained dynamic update of threat value predictions as well as acquired prior expectation toward threat. We also observed that the HS and LS groups showed opposite patterns in terms of learning rate, indicating that in the LS group, observers showed faster learning for the CS + than CS- trials, whereas in the HS group, observers showed slower but steadier learning ⁷⁶. These findings could be interpreted as observers empathize quicker and thus learn from the LS demonstrator and might suggest that observers learn from HS demonstrators through a different strategy, possibly through slowly establishing trust of the demonstrator.

Second, observational threat learning downregulated low-frequency brain oscillatory activity. Specifically, we found decreased oscillatory power for CS + vs. CS- trials in the delta (1–3 Hz) and theta (4–8 Hz) frequency bands. We highlight two possible mechanisms through which low-frequency activity might play a crucial role in observational threat learning. First, the local and distal communications of threat-related oscillatory activity might shape social learning. This account is supported by previous evidence suggesting that the low-frequency oscillations represent a mechanism for the coordination of cell assemblies that synchronizes spiking activity and gates interregional information transfer of aversive responses, as demonstrated in both human ^55,57 and nonhuman animal models ⁵⁶. Here, we extend the previous proposal by arguing that synchronization of oscillatory activity contributes to information transfer across brains ²². A second possible account for low-frequency responses to social threat learning would suggest that prefrontal low-frequency oscillatory power reduction might reflect the modulation of cardiac vagal control that mediates heart rate variability ⁷⁷, which is known to affect emotional regulation brain networks ⁷⁸. This perspective is consistent with other accounts suggesting that neural processes associated with emotional processing are involved during observational learning to regulate cognition and adaptive behaviors ^79,80. Future research should focus on examining the exact functional meanings of oscillatory power in low-frequency bands during social threat learning.

Our third main findings was the significantly enhanced low-frequency BtBC at centro-frontal channels, which echoed findings in previous studies ^81,82. The fact that BtBC was identified for observational learning vs. visual baseline allowed us to exclude the possibility that BtBC could simply reflect shared environment or stimulus inputs between demonstrator and observer ^22,83. On the source level, observers’ right INS coupled to demonstrators’ left INS, bilateral vmPFC, as well as right DLPFC (all within a fronto-limbic circuit) in a way that predicted learning outcome. In addition, BtBC involving observers’ rINS correlated with their self-reported empathic concern, which assesses “other-oriented” feelings of sympathy and concern for unfortunate others ⁶⁸. These were only observed in the delta band (but not in any other band) and in the LS (but not HS) group. INS is an important hub of aversive learning network linked to empathic value of pain in observers ³. This involvement bolsters the “low-status benevolence” hypothesis positing that observers might be more empathic towards low-status demonstrators. The rationale behind “low-status benevolence” might be underpinned by the knowledge that others are superior (than oneself) often violates positive self-concept and aggravates negative feelings due to upward social comparison ³⁹; in contrast, knowing others are inferior is associated with subjective sensitivity to other’s pain and affective sharing ³⁸. Affective sharing has been reported as a key aspect supporting observational threat learning, as argued in several previous work ^41,42. It is reasonable to infer that the LS (vs. HS) group was involved more in the process of affective sharing, leading to stronger learning (as indexed by computational learning rate and initial value). Concerning the functional meaning of vmPFC, previous evidence from learning and decision-making has shown that vmPFC encodes direct valuation ^84,85. A recent patient study also demonstrated a causal role of vmPFC in direct threat conditioning, suggesting that vmPFC is required to generate a sustained conditioned response when anticipating the US during threat learning in humans ⁸⁶. Our results demonstrate the role of vmPFC in threat learning into a social context. These findings echo previous neuroimaging studies in humans showing shared neuronal patterns of this region in direct and observational learning, suggesting that vmPFC not only represents individuals’ own valuation of social information in learning ^85,87 but also processes observational outcome prediction errors ¹⁶. Like vmPFC, DLPFC activity was commonly involved in direct and observational learning, and typically correlated with expected action prediction errors ¹⁶. In support of this, the DLPFC has been implicated in selection of action ⁸⁸, where DLPFC activity paralleled with uncertainly about which action to select ⁸⁹. In social situations, DLPFC was thought to be an important driver of emotion regulation ⁹⁰. A meta-analysis of fMRI studies of emotional regulation found that increased BOLD signal in DLPFC was associated with strategies that aimed at downregulating negative emotions ^90,91. These findings suggest that an integrated neural system that arises from demonstrator and observer, involving the brain’s affective, reward, and social hubs, supports observational learning of threat.

Fourth, BtBC was predictive of learning rate and ensuing learning outcomes. In particular, a time-varying support vector regression model revealed that BtBC predicted learning outcome after several seconds relative to the CS onset (immediately preceding the US). These effects were detectable in the LS group only, suggesting that social status relationship is a key factor shaping the alignment of neural processes across demonstrator and observer. Such neural alignment impacts on the real-time information transfer and update of threat value, and eventually on the learning outcome ²⁴. The predictive power of BtBC was augmented when a threat was imminent to the demonstrator (close to the timing at which the shock was delivered), suggesting that BtBC supporting social learning of threat value requires a buffering time (rather than immediately after the onset of CSs) to allow the observers integrate endogenous and exogenous cues. The late emergence (related to CS-onset) of this effect mimics a typical finding in research on Pavlovian fear conditioning where the first 4-seconds are thought of as an immediate orienting response to a CS, whereas the seconds just before the US onset are considered to depend on more complex cognitive computations, including processing of properties of the CS, US, and their relationship (e.g. expectancy learning and sensory integration) ^92–94. In our paradigm, it is possible that this temporal interval also contains integration of information about the demonstrator, such as their internal affective states (empathy) and relative social status. One alternative explanation for the mechanism of BtBC is that BtBC could reflect the alignment between imagery (simulation) of finger movement in observers and execution of involuntary movement in demonstrators, as these two processes can elicit similar response in the motor system ⁹⁵. We ruled out this possibility because (1) we did not observe BtBC in the left motor cortices, and (2) we observed that BtBC can predict learning outcomes, reflecting that BtBC is relevant for information transfer beyond sensorimotor alignment.

Finally, our fifth main finding is that the observed BtBC seemed to be driven by shared attention and emotion between the observer and demonstrator. We tested this possibility by parsing observers’ gaze fixation and demonstrator-observer pupillary coupling. We noticed that when the US was approaching, observers primarily paid attention to demonstrator’s right hand where the shock was administered, which is an area that most demonstrators reported that they would pay major attention to during US imminence as well. This suggests that when a threat was imminent to the demonstrator, both demonstrators and observers focused on the upcoming shock delivery, creating an experimental situation favoring the emergence of BtBC. We tested this possibility by parsing pupillary coupling within the dyads. Pupillary dilations reflect the dynamics of conscious attention and emotion; hence spontaneous coupling of pupil dilation patterns has been widely used as a metric of shared processing of attention and emotion ³¹. Previous studies have shown that collective pupillary coupling reached its crest during the emotional peaks of a narrative ³¹, and was positively correlated with engagement during conversation ⁷⁴. Consistent with BtBC patterns, we found CS + vs. CS- trials elicited greater eye-to-eye coupling in the LS group during observational learning, suggesting shared attention and emotion as likely sources for BtBC ³⁰.

In summary, our study investigated the neural mechanism of social threat learning from an interpersonal neuroscience perspective ²². Leveraging a sequential MEG imaging approach, we recorded neuronal activity from demonstrator-observer dyads during the acquisition of threat value. We found that neuronal activity in the low-frequency bands coupled across the demonstrator and observer in a way that reflects social status relationships and predicts learning outcomes. Besides providing neuro-physiological evidence supporting that BtBC is a possible neural marker for social threat learning, our work opens unprecedented opportunities for future research and clinical practices to better understand and treat threat- and anxiety-related disorders as well as maladaptive fears.

4.1. Ethics statement

The study was conducted following the Declaration of Helsinki, with written informed consent obtained from each volunteer before data collection. The study procedure was approved by the Ethics Review Authority in Sweden (2020 − 00101). Each participant was reimbursed with 350 SEK (about $40) for their participation.

4.2. Participants

Sixty adults (37 females, aged 27 ± 5.07) were recruited as observers and three adults (2 females, aged 28 ± 2.89) as demonstrators. Each demonstrator was randomly paired with twenty observers, forming 60 demonstrator-observer dyads. All participants were healthy, right-handed, and were recruited by advertisements at Karolinska Institutet and surrounding local communities. Data from five observers were dropped out due to insufficient signal quality (N = 3) or scanner malfunction during MEG recording (N = 2), hence 55 dyads contributed to legit data in the final analyses. No statistical methods were used to pre-determine the sample size but our sample size was set to match those reported in previous publications ^3,96.

4.3. Study procedure

The demonstrators and observers visited the laboratory separately. Demonstrators' visit entailed (i) a Pavlovian learning session and (ii) a direct test session, whereas observers' visit entailed three sessions: (i) social status inducing, (ii) observational learning, and (iii) direct test (Fig. 1). Before the task, participants were attached to SCR and shock electrodes, underwent a personalized shock calibration procedure, and were seated in the MEG scanner (see details in the next sections). The experiment was controlled using Presentation software (NeuroBehavioral Systems, Albany California, USA). At the very beginning of the experiment, both demonstrator and observer were asked to watch two white geometric shapes (a parallelogram and a rhombus) on a black background without any additional information, which served as the visual baseline.

Demonstrator sessions. The formal experiment in demonstrators included two blocks of Pavlovian learning. During the acquisition, two different white geometries served as CSs (a triangle and a rectangle, different from shapes in the visual baseline; Fig. 1B). Each CS lasted for 6 s and was presented 16 times, out of which 10 presentations of the CS+ (62.5%) were paired with the US. The US following the CS + was a 200-microsiemens (µS) electrical nerve stimulation to the demonstrator (see also the calibration procedure below), which was delivered 5.5 s after the onset of CS presentation. The CS- was never co-terminated with a shock. During the inter-trial-interval (ITI), a fixation cross appeared on the center of the screen for 8–12 s. The whole procedure was repeated as a second block, with different CSs (a circle and a pentagon). That is, each block features unique stimulus images, in order to exclude the possibility of carry-over of learning between blocks. The whole procedure of demonstrator sessions was filmed using an MEG-compatible video camera. The shooting angle was fixed across participants. Videos recorded during the Pavlovian learning session in demonstrators were extracted using Adobe Premiere Pro CS6. These videos were later shown to naïve observers during the observational learning session.

Observer sessions. The formal experiment in observers was divided into a status-inducing phase, and two blocks of observational learning and direct test. In the status-inducing phase, we used a math competition task to manipulate observer's perceived social status (i.e., social rank positions) ⁶². Similar procedures to manipulate social status were initially established by Zink et al. ⁹⁷ and later developed by Hu and colleagues ⁶². These authors demonstrated that participants strongly engaged in the social status context with this manipulation. Observers were told that more than 50 volunteers had participated in the task, one of which would be randomly selected as the observer’s partner (i.e., the demonstrator). Observers were then given 10 s for each of six questions to indicate, by pressing the left or right button on a response box corresponding to the left or right expression, which expression had a greater value (Fig. 1C). They were told that performance in this math task would be evaluated by both accuracy and speed, which would be compared with their randomly paired partner. A performance payoff was finally shown based on their estimation performance relative to the demonstrator (in reality the outcome was fixed and manipulated by the experimenter). In the high-status (HS) group, the demonstrator was associated with five stars, whereas in the low-status (LS) group, the demonstrator obtained one star only; the observer's ranking was always in the middle with three stars. Half of observers (N = 27) believed they were observing a HS demonstrator, and the other half (N = 28), a LS demonstrator. An independent t-test confirmed that the manipulation of social status was successful in altering feelings of rank (superior/inferior, t = 12.26, P < 0.001), indicating that observers perceived themselves as higher in status after they saw their partner attain one star (7.11 ± 1.34) than five stars (2.44 ± 1.48). To clarify, the demonstrators were blind to any ranking information.

After that, the observers were told that they would view their partner undergo a learning task and they themselves would go through a similar task thereafter. They watched the video of the demonstrator’s responses to the CS-US pairings to learn which stimulus predicted shock (observational learning session), and then being exposed to the CS in the absence of the demonstrator (direct test session). The observational learning and direct test sessions were repeated as a second block. Videos recorded from each block of the Pavlovian learning sessions in demonstrators were played in the corresponding block of the observational learning sessions in observers (thus, Pavlovian learning in demonstrators and observational learning in observers shared the same experimental timing and trials). At the beginning of the second block, the observers were asked about how many stars they and their partner attained respectively in the math competition task, ensuring that they kept the social rank positions in their minds.

4.4. Electrical nerve stimulation

Electrical shocks, comprising of a single 200-µS pulse, were administrated using an SCG30 stimulator (DeMeTec GmbH, Germany), with felt tips for median nerve stimulation. Before the experiment, participants went through a standard workup procedure to adjust the shock level individually (mean ± SD, 25.32 ± 13.85 mA). This procedure allows each participant to select an intensity that may be perceived as uncomfortable or annoying, but not painful. In the paradigm, shocks will be probabilistically administered (62.5% of CS + trials) to the right wrist of the demonstrators, when the CS + appears. Albeit undergoing a similar stimulation setup, the observers won’t get any shocks during the experiment (except for the final CS + trial during the direct test). The shock electrodes were placed over the participant’s right wrist and adjusted to a position where receiving a shock could elicit a right thumb movement (Fig. 1A). The thumb movement of the demonstrator thus served as a critical cue for observers to acquire the CS-US contingency.

4.5. Data acquisitions

4.5.1. MEG imaging

A Neuromag TRIUX 306-channel MEG system, including 102 magnetometers and 102 pairs of planar gradiometers, was used for MEG data acquisition. Data were measured at a sampling rate of 1000 Hz, with an online bandpass filter (0.1–330 Hz). The MEG facility was placed inside a two-layer magnetically shielded room (model Ak3B, Vacuumschmelze GmbH), with internal active shielding to mitigate electromagnetic artifacts. Four head-position indicator (HPI) coils were attached to the participant's head to monitor head position during MEG recording. The locations for the HPI, anatomical landmarks (e.g., the nasion and the left and right tragus), and additional points (measuring participant’s head shape), were determined using a Polhemus Fastrak motion tracker before the measurements, which allowed a co-registration of the participant's anatomical MRI with the MEG data. Horizontal and vertical electrooculograms (EOG) were recorded with MEG imaging simultaneously.

4.5.2. Structural MRI

High-resolution T1-weighted 3D volume MRI data were acquired using a GE Discovery 3-T MR scanner (voxel size: 1.0 × 1.0 × 1.0 mm, field of view: 256 mm, repetition time: 2300 ms, echo time: 2.98 ms) from both demonstrators and observers. MRI data were processed with FreeSurfer (version 7) to get cortical reconstruction and volumetric segmentation ⁹⁸.

4.5.3. Eye tracking

For eye tracking, we used an EyeLink 1000 binocular tracker (SR Research, Canada), which measured both eyes binocular at a sampling rate of 500 Hz. Participants were seated at approximately 80 cm distance from the eye tracker attached to a projector screen, positioned at approximately 100 cm distance. All participants underwent a 5-point eye-tracking calibration procedure at the beginning of the experiment.

4.5.4. Skin conductance

We measured participants' electrodermal activity using a BIOPAC MP150 system as the skin conductance signal. Signals were output into the MEG system analog channels and sampled side-by-side with the MEG data at a rate of 1000 Hz. SCR electrodes were attached to participants' middle phalanges of the second and third fingers of the left hand (Fig. 1A).

4.6. Data analyses

4.6.1. Learning assessment

We assessed the learning performance based on conditioned response, which was operationalized as the differential (CS + minus CS-) SCR amplitude and/or differential pupil diameter. In addition, a computational modeling approach was applied to provide more mechanistic insights that helped quantifying the underlying processes of observational learning on the trial-by-trial basis. For statistical tests, given the fact that each demonstrator formed about 20 dyads with 20 distinct observers, we used linear mixed-effects modeling. Data were modelled by CS type (within-subject predictor: CS + vs. CS-), social status (between-subject predictor: high status vs. low status), and the interaction of these fixed effects. Random effects were estimated for observer and demonstrator identity, with the former being a nested factor within the latter [y ~ CS type * social status + (1|demonstrator/observer)]. Linear mixed-effects modeling was conducted using the lme4 package in R ⁹⁹. ANOVAs were conducted to detect the main and interaction effects, using the anova function in R. Significance tests on the parameters were performed using Satterthwaite P values implemented in the lmerTest package ^100,101. Post-hoc contrasts were carried out using the emmeans package ¹⁰².

SCR. The raw signals were low-pass filtered (1 Hz) and high-pass filtered (0.1 Hz) offline. The SCR was measured for each trial as the largest base-to-peak amplitude in µS from 0.5 to 4.5 s following each CS presentation. SCR scores were z-transformed to approach normal distribution ¹⁰³.

Pupillometry. Blinks, as revealed by the peak detection on the velocity of the pupil signal and the EyeLink software, were linearly interpolated. Following recent recommendations ^104,105, we estimated the effect of blinks and saccades on the pupil response via deconvolution, and omitted these components from the data using linear regression. The pupil time series were then bandpass filtered (0.01-10 Hz), z-transformed, and resampled to 100 Hz. Pupil response for each trial was estimated as the average pupil diameter in the time window of 0-5.5 s relative to stimulus onset, that was baseline corrected by subtracting the mean pupil diameter 0.5 s prior to the onset of the CS presentation.

Computational modeling. The simple Rescorla-Wagner reinforcement learning model was used for a standard account of error-driven associative learning ¹⁰⁶. We defined s_t as the conditioned stimulus (CS + or CS-) on trial t, U_t as the US delivered (1 for US, 0 for no US), V_t(x) as value (i.e., shock) predictions. We estimated the punishment prediction error δ_t, which measured the difference between the expected and predicted value on trial t:

$${V}_{t+1}\left({s}_{t}\right)= {V}_{t}\left({s}_{t}\right)+\alpha {\delta }_{t}$$

$${\delta }_{t}= {U}_{t}- {V}_{t}\left({s}_{t}\right)$$

Here, initial values (V₀), was set to 0.5 for the observational learning session, but, importantly, was estimated as a free parameter for the direct test session to capture the carry over learning effect across sessions. For each trial, s_t was updated according to the prediction error. The learning rate $\alpha$ (0 < $\alpha$ < 1) for the value update was a constant free parameter for each participant.

We fitted the Rescorla-Wagner model to the SCR data ⁶⁴. We modeled the likelihood of SCR on each trial (SCR_t) as follows:

It represents a Gaussian distribution around a mean determined by the scaled value predicted by the model on a given trial t, plus a constant term. For completeness, we also tested a Rescorla-Wagner with V₀ being fixed at 0.5 for the direct test session, and a Pearce-Hall model ⁵⁴ (both with and without V₀ being fixed for the direct test session). The model described above was consistently the winning model across experimental conditions. Model estimation and model selection were performed using the hierarchical Bayesian approach with the statistical computing language Stan ¹⁰⁷.

4.6.2. Participant’s response to conditioned stimuli

The raw MEG data were first submitted to MaxFilter software which implemented a temporal signal space separation algorithm (buffer length = 10 s, cut-off correlation = 0.98) to remove external noise from MEG recordings ¹⁰⁸. Movement correction was performed by shifting the head position to a position based on the median of the continuous head position during the MEG recording. Further pre-processing was performed using the Fieldtrip toolbox ¹⁰⁹ in MATLAB R2019b (MathWorks Inc. Natick, MA). A discrete Fourier transform filter was applied to reduce line noise (for the 50 Hz, 100 Hz, and 150 Hz). Jump artefacts and artefacts from eye-blinks were identified and marked in the continuous MEG signals using Fieldtrip’s automatic artefact detection algorithms for MEG data and EOG data respectively. Artefacts were removed by zero-masking the detected data segments (0.11% ± 0.13% of the whole data). An independent component analysis ¹¹⁰ was then conducted. Components related to eye-blinks and heartbeats were identified based on their correlations with EOG or ECG and removed from the data (4.14 ± 8.62 components). The cleaned MEG time series were then chopped into epochs during the time window of -6 to 10 s relative to stimulus onset. Epochs were averaged per session, per condition, for each participant.

Non-average epochs were submitted to the time-frequency analysis to estimate participant’s response to stimuli. We calculated induced activity in delta (1–3 Hz), theta (4–8 Hz), alpha (9–12 Hz), and beta (13–30 Hz) frequency bands by time-frequency decomposition. A complex Morlet wavelet with a Gaussian kernel of 7-cycle width was used. This procedure was applied to frequencies ranging from 1 to 30 Hz in steps of 1 Hz, in the time window of -2 to 5.5 s relative to stimulus onset. Event-related power changes were estimated as the percentage change of power relative to the baseline (-2 to -0.2 s relative to stimulus onset). Oscillatory activities within different frequency bands (delta, theta, alpha, and beta) were extracted during the time window of 0-5.5 s post stimulus (i.e., CS) onset. This analysis was done for observers during the observational learning phase. In this study, we focused our analyses on MEG signals from magnetometers, which were reported to generate equivalent results as those from gradiometers after signal space separation ¹¹¹.

For statistical analysis of the response to stimuli, we did non-parametric cluster-based permutation tests¹¹² on the entire time-frequency window from 0-5.5 s and 1–30 Hz. First, a dependent-sample test (a step in the cluster-based permutation test) on CS + trials vs. CS- trials was conducted. Second, to investigate whether oscillatory activities were biased by social status, an independent test between the high-status (HS) and low-status (LS) conditions was performed. The interaction between CS type and social status was further tested by conducting a cluster-based permutation test comparing the differences between the contrast “CS + _HS > CS-_HS” and the contrast “CS + _LS > CS-_LS”. Samples were clustered in all tests in connected sets based on temporal, frequency, and channel (N > 2) adjacency exceeding alpha level (0.05). Cluster-level statistics were calculated by taking the sum of the t-values within the cluster with was calculated across randomly shuffling of condition labels. The permutations were performed 1000 times. Cluster statistics exceeding 0.95 percentile (two-tailed) of the permutation distribution are considered significant (alpha = 0.05).

We calculated dynamic imaging of coherent sources (DICS) to map out spectral power along the cortex and to find sources of the significant time-frequency responses. Time and frequency windows were matched with those on the sensor level. The DICS cortical power map was computed as the percentage change of power [i.e., (Power_CS+ - Power_CS−) / Power_CS−]. We morphed each individual brain to an average brain in FreeSurfer to get the group-level results.

4.6.3. Brain-to-brain coupling (BtBC)

Brain-to-brain coupling (BtBC) was estimated using Circular Correlation Coefficient (CCorr) ¹¹³. CCorr is a measure of the circular covariance of differences between the observed phase and the mean phase ¹¹⁴. It is defined as follows:

$${CCorr}_{x,y}= \frac{{\sum }_{t=1}^{N}\text{s}\text{i}\text{n}(x- \stackrel{-}{x})\text{s}\text{i}\text{n}(y- \stackrel{-}{y})}{\sqrt{{\sum }_{t=1}^{N}{sin}^{2}(x- \stackrel{-}{x}){sin}^{2}(y- \stackrel{-}{y})}}$$

In the formula, $\stackrel{-}{x}$ and $\stackrel{-}{y}$ are the mean phases for channels from the demonstrator and the observer in a pair, respectively. According to simulations, CCorr was much more robust to coincidental coupling, compared to e.g., phase-locking value, when analyzing BtBC data ¹¹⁴. CCorr was computed for the time window of 0-5.5 s post stimulus onset (during observational learning for observers, and during Pavlovian learning for demonstrators), and for each of frequencies that showed significant effect in time-frequency analysis. Phases for each specific narrow frequency bands were estimated using Hilbert transformation. CCorr values were then normalized by Fisher’s z transformation and converted into absolute values ^32,115. CCorr values ranged from 0 (totally decoupled) to 1 (perfectly coupled), and was calculated using the CircStat toolbox for MATLAB ¹¹⁶.

For statistical analyses, we performed linear mixed-effects modeling to account for the nested data structure for each combination of demonstrator-observer channels (102 × 102 = 10404 channel combinations in total). To reduce the number of statistical tests, we used the following two steps. First, we compared the task BtBC (BtBC during observational learning) with baseline BtBC (BtBC for visual baseline, during which participants were watching geometries without additional information) for each channel combination. We focused on channel combinations for which task BtBC was significantly larger than baseline BtBC. This step, following recent recommendations ^32,117, allowed us to filter out channel combinations with null effect and mitigate the concern that observed BtBC effects were simply due to similar environment or inputs. Only significant channel combinations survived from the first step were analyzed in the second step, in which BtBC was statistically compared across CS types and social status. FDR correction (P_FDR < 0.05) was applied to control for the multiple testing ¹¹⁸. We also reported channels exhibiting a significant effect without correction (P_uncorrected < 0.001). For visualization of BtBC that showed main or interaction effects, we used a custom script (dualheadnetplot.m) in MATLAB ¹¹⁹.

4.6.4. Clustered BtBC

In the next step, BtBC networks (associated with significant interaction effects from BtBC statistical analyses above) during CS + _HS and CS + _LS were respectively clustered using nonnegative matrix factorization (NMF) ¹²⁰. To reach stable results, we ran NMF 1000 times. Factorization rank (i.e., the number of the cluster, N = 3) was determined by taking the first value of rank for which the cophenetic coefficient starts decreasing, according to the “Brunet” version of NMF ¹²¹. The relationship between clustered BtBC and learning was tested using Pearson correlation analyses.

4.6.5. Source-level BtBC

To find the cortical source of BtBC, we then applied source reconstruction to the data. To this end, noise weighted minimum-norm estimates (MNE) were used for calculating BtBC in source space. We estimated the cortically constrained dynamic statistical parametric mapping (dSPM) ¹²² of the MEG data. The pre-processed data were imported into MNE-Python (version 0.23.4) ¹²³. Participant’s structural MRI was used to generate the head model and co-registered with the MEG data. The forward solution was estimated from a source space of 5124 evenly spaced points and a boundary-element model (BEM) volume conductor model derived from the subject’s anatomical MRI. The inverse solution was calculated based on the forward solution, with the noise covariance matrix estimated from 2 min of empty room data recorded before the experiment.

We derived MEG time series in source space from the dSPM source reconstructions, which were then exported to MATLAB for BtBC analysis. Regions of interest were defined based on anatomical labels in the Desikan-Killiany Atlas ¹²⁴. The ROI selections aimed to match the regions showing pronounced neural activity in DICS power mapping. Five regions in both hemispheres were initially included: vmPFC (atlas label: “medial orbitofrontal”), DLPFC (“rostral middle frontal”), pSTS (“superior temporal”), PoG (“postcentral”), and OCC (“lateral occipital”). Furthermore, considering the critical roles of INS (“insula”) and ACC (“caudal anterior cingulate”) in social threat learning ³, we additionally included both.

To probe the relationship between source-level BtBC and learning, we conducted a series of Pearson correlations on each ROI. For each group, FDR correction (P_FDR < 0.05) was applied to control for multiple comparisons¹¹⁸.

4.6.6. Predicting learning from BtBC

The time course of BtBC for CS + was averaged over the source-level ROIs that showed significant correlations with learning outcome. The cumulative BtBC, which showed more stable prediction accuracy compared to moment-to-moment BtBC data ^125,126, across the time was used as the predictor. Specifically, we repeatedly predicted learning in the direct test based on time-cumulative BtBC in the observational learning from 0.1 to 5.5 s (post stimulus onset) with a time increment of 0.1 s (the parameter of time increment has been tested in our previous work ¹²⁶). The cumulative BtBC at time point k was computed as a sum of BtBC at time points from 1 to k:

$${BtBC}_{cum}= {\sum }_{i=1}^{i=k}{BtBC}_{i}$$

The support vector regression (SVR) was used to predict learning outcome at each time point. The training dataset was trained by ε-SVR with the radial basis function (RBF). A leave-one-dyad-out cross-validation was used. Prediction performance of time-varying SVR was estimated by the Pearson correlation coefficient between predicted and actual learning outcome ^72,73, with a series of P-values FDR corrected ¹¹⁸. We also reported mean absolute errors (MAE) for evaluation of prediction performance. SVR were carried out using the libsvm toolbox (version 2.3) in MATLAB.

To provide a better understanding of cumulative BtBC predicting learning, we also parsed behavioral significance at time points that showed satisfactory prediction performance (P_FDR < 0.05). Two complementary analyses were conducted. First, we aimed at determining which part(s) of the video the observers were focusing on. Fixation behaviors were identified with the EyeLink software, and analyzed using custom scripts in MATLAB. Three rectangular areas of interest (AOI) were defined (Fig. 6): (i) the CS on the projector screen, (ii) the demonstrator’s face, and (iii) the demonstrator’s right hand entailing the electrical nerve stimulation. Fixation time at each AOI was normalized to the total fixation time at the screen. Second, we evaluated whether the observers built shared attention ^31,74 and emotion ³¹ with the demonstrators. To this end, we estimated eye-to-eye coupling by computing the CCorr (same as that for BtBC computations) between time series of pupil response between demonstrators and observers.

Acknowledgment

We would like to thank Tabea Eimer, Colin Jones, Claire Pleche, Andreas Henriksson, Anton Toftgård, and Daniel Sortebech for their assistance in data collection.

Funding

Funding is provided by a Consolidator Grant (2018-00877) from the Swedish Research Foundation to Andreas Olsson. MEG data for this study are collected at NatMEG, the National Facility for Magnetoencephalography (www.natmeg.se), Karolinska Institutet, Sweden. MRI data collection is supported by a grant to the Stockholm University Brain Imaging Centre (SU FV-5.1.2-1035-15). Structural MRI from three observers is retrieved from the IronAge study database, which is financially supported by the Swedish Research Council.

Author contributions

Y.P. and A.O. conceived the initial research idea. D.L. and A.O. supervised the project. Y.P. acquired the data. Y.P., M.C.V., and D.L. implemented the MEG setup. Y.P. performed data analyses, with inputs from M.C.V. on MEG data analyses and L.Z. on computational modeling. Y.P., M.C.V., L.Z., D.L., and A.O. interpreted the results and wrote the manuscript.

Competing interests

The authors declare no competing interests.

Data and code availability

All data needed to evaluate the conclusions in the paper are present in the paper, Supplementary Materials, and/or the OSF repository (https://osf.io/ueagm/). Further inquiries can be directed to the corresponding authors.

Boyd, R., Richerson, P. J. & Henrich, J. The cultural niche: Why social learning is essential for human adaptation. Proc. Natl. Acad. Sci. 108, 10918–10925 (2011).
Hoppitt, W. & Laland, K. N. Social Learning. (Princeton University Press, 2013). doi:10.1515/9781400846504
Lindström, B., Haaker, J. & Olsson, A. A common neural network differentially mediates direct and social fear learning. Neuroimage 167, 121–129 (2018).
Lindström, B., Golkar, A., Jangard, S., Tobler, P. N. & Olsson, A. Social threat learning transfers to decision making in humans. Proc. Natl. Acad. Sci. U. S. A. 116, 4732–4737 (2019).
Olsson, A. & Phelps, E. A. Social learning of fear. Nat. Neurosci. 10, 1095–1102 (2007).
Debiec, J. & Olsson, A. Social fear learning: from animal models to human function. Trends Cogn. Sci. 21, 546–555 (2017).
Fiorito, G. & Scotto, P. Observational learning in octopus vulgaris. Science (80-.). 256, 545–547 (1992).
Sherry, D. F. & Galef, B. G. Cultural transmission without imitation: Milk bottle opening by birds. Anim. Behav. 32, 937–938 (1984).
Thornton, A. Social learning about novel foods in young meerkats. Anim. Behav. 76, 1411–1421 (2008).
van de Waal, E., Borgeaud, C. & Whiten, A. Potent social learning and conformity shape a wild primate’s foraging decisions. Science (80-.). 340, 483–485 (2013).
Askew, C. & Field, A. P. The vicarious learning pathway to fear 40 years on. Clin. Psychol. Rev. 28, 1249–1265 (2008).
Gerull, F. C. & Rapee, R. M. Mother knows best: effects of maternal modelling on the acquisition of fear and avoidance behaviour in toddlers. Behav. Res. Ther. 40, 279–287 (2002).
Skversky-Blocq, Y., Pine, D. S. & Shechner, T. Using a novel paradigm to examine observational fear‐learning across development. Depress. Anxiety 38, 731–738 (2021).
Hewlett, B. Social learning and innovation in adolescence. Hum. Nat. 32, 239–278 (2021).
Olsson, A., Nearing, K. I. & Phelps, E. A. Learning fears by observing others: The neural systems of social fear transmission. Soc. Cogn. Affect. Neurosci. (2007). doi:10.1093/scan/nsm005
Burke, C. J., Tobler, P. N., Baddeley, M. & Schultz, W. Neural mechanisms of observational learning. Proc. Natl. Acad. Sci. U. S. A. 107, 14431–14436 (2010).
Wisdom, T. N., Song, X. & Goldstone, R. L. Social learning strategies in networked groups. Cogn. Sci. 37, 1383–1425 (2013).
Zhang, L., Gläscher, J., Zhang & Gläscher, J. A brain network supporting social influences in human decision-making. Sci. Adv. 6, eabb4159 (2020).
Olsson, A., Knapska, E. & Lindström, B. The neural and computational systems of social learning. Nat. Rev. Neurosci. 21, 197–212 (2020).
Keysers, C., Knapska, E., Moita, M. A. & Gazzola, V. Emotional contagion and prosocial behavior in rodents. Trends Cogn. Sci. (2022). doi:10.1016/j.tics.2022.05.005
Han, Y. et al. Bidirectional cingulate-dependent danger information transfer across rats. PLOS Biol. 17, e3000524 (2019).
Pan, Y., Novembre, G. & Olsson, A. The interpersonal neuroscience of social learning. Perspect. Psychol. Sci. 17, 680–695 (2022).
Hasson, U., Ghazanfar, A. A., Galantucci, B., Garrod, S. & Keysers, C. Brain-to-brain coupling: A mechanism for creating and sharing a social world. Trends in Cognitive Sciences (2012). doi:10.1016/j.tics.2011.12.007
Pan, Y., Novembre, G., Song, B., Li, X. & Hu, Y. Interpersonal synchronization of inferior frontal cortices tracks social interactive learning of a song. Neuroimage 183, 280–290 (2018).
Kostorz, K., Flanagin, V. L. & Glasauer, S. Synchronization between instructor and observer when learning a complex bimanual skill. Neuroimage 116659 (2020). doi:10.1016/j.neuroimage.2020.116659
King-Casas, B. et al. Getting to Know You: Reputation and Trust in a Two-Person Economic Exchange. Science (80-.). 308, 78–83 (2005).
Zhang, W. & Yartsev, M. M. Correlated Neural Activity across the Brains of Socially Interacting Bats. Cell 178, 413–428.e22 (2019).
Kingsbury, L. et al. Correlated Neural Activity and Encoding of Behavior across Brains of Socially Interacting Animals. Cell 178, 429–446.e16 (2019).
Liu, Y. et al. Measuring speaker-listener neural coupling with functional near infrared spectroscopy. Sci. Rep. 7, 43293 (2017).
Dikker, S. et al. Brain-to-brain synchrony tracks real-world dynamic group interactions in the classroom. Curr. Biol. 27, 1375–1380 (2017).
Kang, O. & Wheatley, T. Pupil dilation patterns spontaneously synchronize across individuals during shared attention. J. Exp. Psychol. Gen. 146, 569–576 (2017).
Goldstein, P., Weissman-Fogel, I., Dumas, G. & Shamay-Tsoory, S. G. Brain-to-brain coupling during handholding is associated with pain reduction. Proc. Natl. Acad. Sci. U. S. A. 115, E2528–E2537 (2018).
Redcay, E. & Schilbach, L. Using second-person neuroscience to elucidate the mechanisms of social interaction. Nat. Rev. Neurosci. 20, 495–505 (2019).
Levy, J., Lankinen, K., Hakonen, M. & Feldman, R. The integration of social and neural synchrony: a case for ecologically valid research using MEG neuroimaging. Soc. Cogn. Affect. Neurosci. 16, 143–152 (2021).
Qu, C., Ligneul, R., Van der Henst, J.-B. & Dreher, J.-C. An integrative interdisciplinary perspective on social dominance hierarchies. Trends Cogn. Sci. 21, 893–908 (2017).
Qu, C. & Dreher, J.-C. Sociobiology: Changing the dominance hierarchy. Curr. Biol. 28, R167–R169 (2018).
Li, S., Krueger, F., Camilleri, J. A., Eickhoff, S. B. & Qu, C. The neural signatures of social hierarchy-related learning and interaction: A coordinate- and connectivity-based meta-analysis. Neuroimage 245, 118731 (2021).
Feng, C. et al. Social hierarchy modulates neural responses of empathy for pain. Soc. Cogn. Affect. Neurosci. 11, 485–495 (2016).
Takahashi, H. et al. When your gain is my pain and your pain is my gain: Neural correlates of envy and schadenfreude. Science (80-.). 323, 937–939 (2009).
Smith, R. H. et al. Envy and schadenfreude. Personal. Soc. Psychol. Bull. 22, 158–168 (1996).
Pärnamets, P., Espinosa, L. & Olsson, A. Physiological synchrony predicts observational threat learning in humans. Proc. R. Soc. B Biol. Sci. 287, 20192779 (2020).
Olsson, A. et al. Vicarious fear learning depends on empathic appraisals and trait empathy. Psychol. Sci. 27, 25–33 (2016).
Selbing, I., Lindström, B. & Olsson, A. Demonstrator skill modulates observational aversive learning. Cognition 133, 128–139 (2014).
Chang, S. W. C., Winecoff, A. A. & Platt, M. L. Vicarious reinforcement in rhesus macaques (Macaca Mulatta). Front. Neurosci. 5, 27 (2011).
Kendal, R. et al. Chimpanzees copy dominant and knowledgeable individuals: implications for cultural diversity. Evol. Hum. Behav. 36, 65–72 (2015).
Blue, P. R., Hu, J. & Zhou, X. Higher status honesty is worth more: The effect of social status on honesty evaluation. Front. Psychol. 9, 350 (2018).
Dalmaso, M., Pavan, G., Castelli, L. & Galfano, G. Social status gates social attention in humans. Biol. Lett. 8, 450–452 (2012).
Apesteguia, J., Huck, S. & Oechssler, J. Imitation—theory and experimental evidence. J. Econ. Theory 136, 217–235 (2007).
Kendal, J. R., Rendell, L., Pike, T. W. & Laland, K. N. Nine-spined sticklebacks deploy a hill-climbing social learning strategy. Behav. Ecol. 20, 238–244 (2009).
Leong, V. et al. Speaker gaze increases information coupling between infant and adult brains. Proc. Natl. Acad. Sci. U. S. A. 114, 13290–13295 (2017).
Stephens, G. J., Silbert, L. J. & Hasson, U. Speaker-listener neural coupling underlies successful communication. Proc. Natl. Acad. Sci. 107, 14425–14430 (2010).
Nummenmaa, L. et al. Mental action simulation synchronizes action–observation circuits across individuals. J. Neurosci. 34, 748–757 (2014).
Karalis, N. et al. 4-Hz oscillations synchronize prefrontal–amygdala circuits during fear behavior. Nat. Neurosci. 19, 605–612 (2016).
Taub, A. H., Perets, R., Kahana, E. & Paz, R. Oscillations synchronize amygdala-to-prefrontal primate circuits during aversive learning. Neuron 97, 291–298.e3 (2018).
Chen, S. et al. Theta oscillations synchronize human medial prefrontal cortex and amygdala during fear learning. Sci. Adv. 7, eabf4198 (2021).
Choi, J. & Jeong, Y. Elevated emotional contagion in a mouse model of Alzheimer’s disease is associated with increased synchronization in the insula and amygdala. Sci. Rep. 7, 46262 (2017).
Bastuji, H., Cadic-Melchior, A., Magnin, M. & Garcia-Larrea, L. Intracortical functional connectivity predicts arousal to noxious stimuli during sleep in humans. J. Neurosci. 41, 5115–5123 (2021).
Zhu, Y. et al. Instructor–learner neural synchronization during elaborated feedback predicts learning transfer. J. Educ. Psychol. (2021). doi:10.1037/edu0000707
Zheng, L. et al. Enhancement of teaching outcome through neural prediction of the students’ knowledge state. Hum. Brain Mapp. 39, 3046–3057 (2018).
Pan, Y., Novembre, G., Song, B., Zhu, Y. & Hu, Y. Dual brain stimulation enhances interpersonal learning through spontaneous movement synchrony. Soc. Cogn. Affect. Neurosci. 16, 210–221 (2021).
Pan, Y., Guyon, C., Borragán, G., Hu, Y. & Peigneux, P. Interpersonal brain synchronization with instructor compensates for learner’s sleep deprivation in interactive learning. Biochem. Pharmacol. 191, 114111 (2021).
Hu, J. et al. Social status modulates the neural response to unfairness. Soc. Cogn. Affect. Neurosci. 11, 1–10 (2016).
Cheng, X. et al. Integration of social status and trust through interpersonal brain synchronization. Neuroimage 246, 118777 (2022).
Li, J., Schiller, D., Schoenbaum, G., Phelps, E. A. & Daw, N. D. Differential roles of human striatum and amygdala in associative learning. Nat. Neurosci. 14, 1250–1252 (2011).
Espinosa, L. et al. Enhanced social learning of threat in adults with autism. Mol. Autism 11, 71 (2020).
Kleberg, J. L., Selbing, I., Lundqvist, D., Hofvander, B. & Olsson, A. Spontaneous eye movements and trait empathy predict vicarious learning of fear. Int. J. Psychophysiol. 98, 577–583 (2015).
Haaker, J., Yi, J., Petrovic, P. & Olsson, A. Endogenous opioids regulate social threat learning in humans. Nat. Commun. 8, 1–9 (2017).
Davis, M. H. Measuring individual differences in empathy: Evidence for a multidimensional approach. J. Pers. Soc. Psychol. 44, 113–126 (1983).
Schlund, M. W., Hudgins, C. D., Magee, S. & Dymond, S. Neuroimaging the temporal dynamics of human avoidance to sustained threat. Behav. Brain Res. 257, 148–155 (2013).
Ferreira de Sá, D. S., Michael, T., Wilhelm, F. H. & Peyk, P. Learning to see the threat: temporal dynamics of ERPs of motivated attention in fear conditioning. Soc. Cogn. Affect. Neurosci. 14, 189–203 (2019).
Gallistel, C. R. & Gibbon, J. Time, rate, and conditioning. Psychol. Rev. 107, 289–344 (2000).
Hou, Y., Song, B., Hu, Y., Pan, Y. & Hu, Y. The averaged inter-brain coherence between the audience and a violinist predicts the popularity of violin performance. Neuroimage 211, 116655 (2020).
Kosinski, M., Stillwell, D. & Graepel, T. Private traits and attributes are predictable from digital records of human behavior. Proc. Natl. Acad. Sci. 110, 5802–5805 (2013).
Wohltjen, S. & Wheatley, T. Eye contact marks the rise and fall of shared attention in conversation. Proc. Natl. Acad. Sci. 118, e2106645118 (2021).
Kang, O. E., Huffer, K. E. & Wheatley, T. P. Pupil dilation dynamics track attention to high-level information. PLoS One 9, e102463 (2014).
Zhang, L., Lengersdorff, L., Mikus, N., Gläscher, J. & Lamm, C. Using reinforcement learning models in social neuroscience: frameworks, pitfalls and suggestions of best practices. Soc. Cogn. Affect. Neurosci. 15, 695–707 (2020).
Patron, E., Mennella, R., Messerotti Benvenuti, S. & Thayer, J. F. The frontal cortex is a heart-brake: Reduction in delta oscillations is associated with heart rate deceleration. Neuroimage 188, 403–410 (2019).
Mather, M. & Thayer, J. F. How heart rate variability affects emotion regulation brain networks. Curr. Opin. Behav. Sci. 19, 98–104 (2018).
Goubert, L., Vlaeyen, J. W. S., Crombez, G. & Craig, K. D. Learning about pain from others: An observational learning account. J. Pain 12, 167–174 (2011).
Rak, N., Bellebaum, C. & Thoma, P. Empathy and feedback processing in active and observational learning. Cogn. Affect. Behav. Neurosci. 13, 869–884 (2013).
Lindenberger, U., Li, S.-C., Gruber, W. & Müller, V. Brains swinging in concert: cortical phase synchronization while playing guitar. BMC Neurosci. 10, 22 (2009).
Sänger, J., Müller, V. & Lindenberger, U. Intra- and interbrain synchronization and network properties when playing guitar in duets. Front. Hum. Neurosci. 6, (2012).
Hu, Y. et al. Inter-brain synchrony and cooperation context in interactive decision making. Biol. Psychol. 133, (2018).
Bartra, O., McGuire, J. T. & Kable, J. W. The valuation system: A coordinate-based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value. Neuroimage 76, 412–427 (2013).
Zhang, L. & Gläscher, J. A brain network supporting social influences in human decision-making. Sci. Adv. 6, eabb4159 (2020).
Battaglia, S., Garofalo, S., di Pellegrino, G. & Starita, F. Revaluing the role of vmPFC in the acquisition of Pavlovian threat conditioning in humans. J. Neurosci. 40, 8491–8500 (2020).
Suzuki, S. et al. Learning to simulate others’ decisions. Neuron 74, 1125–1137 (2012).
Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S. J. & Passingham, R. E. The prefrontal cortex: Response selection or maintenance within working memory? Science (80-.). 288, 1656–1660 (2000).
Frith, C. D. The role of dorsolateral prefrontal cortex in the selection of action as revealed by functional imaging. in Control of Cognitive Processes. Attention and Performance XVI11 (eds. Monsell, S. & Driver, J.) 549–565 (MIT Press Cambridge, 2000).
Alexandra Kredlow, M., Fenster, R. J., Laurent, E. S., Ressler, K. J. & Phelps, E. A. Prefrontal cortex, amygdala, and threat processing: implications for PTSD. Neuropsychopharmacology 47, 247–259 (2022).
Morawetz, C., Bode, S., Derntl, B. & Heekeren, H. R. The effect of strategies, goals and stimulus material on the neural mechanisms of emotion regulation: A meta-analysis of fMRI studies. Neurosci. Biobehav. Rev. 72, 111–128 (2017).
Öhman, A. Differentiation of conditioned and orienting response components in electrodermal conditioning. Psychophysiology 8, 7–22 (1971).
Öhman, A. Orienting reactions, expectancy learning, and conditioned responses in electrodermal conditioning with different interstimulus intervals. Biol. Psychol. 1, 189–200 (1974).
Fredrikson, M., Annas, P., Georgiades, A., Hursti, T. & Tersman, Z. Internal consistency and temporal stability of classically conditioned skin conductance responses. Biol. Psychol. 35, 153–163 (1993).
Eaves, D. L., Riach, M., Holmes, P. S. & Wright, D. J. Motor imagery during action observation: A brief review of evidence, theory and future research opportunities. Front. Neurosci. 10, 514 (2016).
Haaker, J., Molapour, T. & Olsson, A. Conditioned social dominance threat: observation of others’ social dominance biases threat learning. Soc. Cogn. Affect. Neurosci. 11, 1627–1637 (2016).
Zink, C. F. et al. Know your place: Neural processing of social hierarchy in humans. Neuron 58, 273–283 (2008).
Dale, A. M., Fischl, B. & Sereno, M. I. Cortical surface-based analysis. Neuroimage 9, 179–194 (1999).
Bates, D., Sarkar, D., Bates, M. D. & Matrix, L. The lme4 package. R Packag. version 2, 74 (2007).
Kuznetsova, A., Brockhoff, P. B. & Christensen, R. H. B. lmerTest Package: Tests in Linear Mixed Effects Models. J. Stat. Softw. 82, 1–26 (2017).
Luke, S. G. Evaluating significance in linear mixed-effects models in R. Behav. Res. Methods 49, 1494–1502 (2017).
Lenth, R., Singmann, H., Love, J., Buerkner, P. & Herve, M. Emmeans: Estimated marginal means, aka least-squares means. R Packag. version 1, 3 (2018).
Boucsein, W. Principles of electrodermal phenomena. in Electrodermal Activity 1–86 (Springer US, 2012). doi:10.1007/978-1-4614-1126-0_1
Knapen, T. et al. Cognitive and ocular factors jointly determine pupil responses under equiluminance. PLoS One 11, e0155574 (2016).
Urai, A. E., Braun, A. & Donner, T. H. Pupil-linked arousal is driven by decision uncertainty and alters serial choice bias. Nat. Commun. 8, 14637 (2017).
Sutton, R. & Barto, A. Reinforcement Learning: An Introduction. (MIT Press Cambridge, 2018).
Ahn, W.-Y., Haines, N. & Zhang, L. Revealing Neurocomputational Mechanisms of Reinforcement Learning and Decision-Making With the hBayesDM Package. Comput. Psychiatry 1, 24 (2017).
Taulu, S. & Simola, J. Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements. Phys. Med. Biol. 51, 1759–1768 (2006).
Oostenveld, R., Fries, P., Maris, E. & Schoffelen, J.-M. FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Intell. Neurosci. 2011, 1–9 (2011).
Jung, T.-P. et al. Removal of eye activity artifacts from visual event-related potentials in normal and clinical subjects. Clin. Neurophysiol. 111, 1745–1758 (2000).
Garcés, P., López-Sanz, D., Maestú, F. & Pereda, E. Choice of magnetometers and gradiometers after signal space separation. Sensors 17, 2926 (2017).
Maris, E. & Oostenveld, R. Nonparametric statistical testing of EEG- and MEG-data. J. Neurosci. Methods 164, 177–190 (2007).
Jammalamadaka, R. & Sengupta, A. Topics in Circular Statistics. (World Scientific Publishing, 2001).
Burgess, A. P. On the interpretation of synchronization in EEG hyperscanning studies: a cautionary note. Front. Hum. Neurosci. 7, 881 (2013).
Pérez, A., Dumas, G., Karadag, M. & Duñabeitia, J. A. Differential brain-to-brain entrainment while speaking and listening in native and foreign languages. Cortex 111, 303–315 (2019).
Berens, P. CircStat: A MATLAB toolbox for circular statistics. J. Stat. Softw. 31, 1–21 (2009).
Reindl, V., Gerloff, C., Scharke, W. & Konrad, K. Brain-to-brain synchrony in parent-child dyads and the relationship with emotion regulation revealed by fNIRS-based hyperscanning. Neuroimage 178, 493–502 (2018).
Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B 57, 289–300 (1995).
Delaherche, E., Dumas, G., Nadel, J. & Chetouani, M. Automatic measure of imitation during social interaction: A behavioral and hyperscanning-EEG benchmark. Pattern Recognit. Lett. 66, 118–126 (2015).
Gaujoux, R. & Seoighe, C. A flexible R package for nonnegative matrix factorization. BMC Bioinformatics 11, 367 (2010).
Brunet, J.-P., Tamayo, P., Golub, T. R. & Mesirov, J. P. Metagenes and molecular pattern discovery using matrix factorization. Proc. Natl. Acad. Sci. 101, 4164–4169 (2004).
Dale, A. M. et al. Dynamic Statistical Parametric Mapping. Neuron 26, 55–67 (2000).
Gramfort, A. MEG and EEG data analysis with MNE-Python. Front. Neurosci. 7, 267 (2013).
Desikan, R. S. et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 31, 968–980 (2006).
Jiang, J. et al. Leader emergence through interpersonal neural synchronization. Proc. Natl. Acad. Sci. U. S. A. 112, 4274–4279 (2015).
Zhu, Y., Pan, Y. & Hu, Y. Learning desire is predicted by similar neural processing of naturalistic educational materials. eNeuro 6, ENEURO.0083-19.2019 (2019).

There is NO Competing Interest.

Supplement.docx
Supporting information
CS.mp4
Example video clip for CS-
CS.mp4
Example video clip for CS+

Download PDF

Version 1

posted

You are reading this latest preprint version

A multi-brain mechanism for observational threat learning

Status:

Version 1

Abstract

Figures

1. Introduction

2. Results

2.1. Successful observational learning in the sequential MEG imaging paradigm

2.2. Observational learning decreased widespread low-frequency neural response in observers

2.3. Shared neural responses between demonstrators and observers

2.4. Shared neural responses selectively predicted observational learning

2.5. Source-level shared neural responses associated with social status predicted learning

2.6. Shared attention and emotion as likely sources for BtBC predicting learning outcome

3. Discussion

4. Methods

4.1. Ethics statement

4.2. Participants

4.3. Study procedure

4.4. Electrical nerve stimulation

4.5. Data acquisitions

4.5.1. MEG imaging

4.5.2. Structural MRI

4.5.3. Eye tracking

4.5.4. Skin conductance

4.6. Data analyses

4.6.1. Learning assessment

4.6.2. Participant’s response to conditioned stimuli

4.6.3. Brain-to-brain coupling (BtBC)

4.6.4. Clustered BtBC

4.6.5. Source-level BtBC

4.6.6. Predicting learning from BtBC

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1