Abstract
Mentalizing is a core faculty of human social behaviors that involves inferring the cognitive states of others. This process necessitates adopting an allocentric perspective and suppressing one's egocentric perspective, referred to as self–other distinction (SOD). Meanwhile, individuals may project their own cognitive states onto others in prosocial behaviors, a process known as self–other mergence (SOM). It remains unclear how the two opposing processes coexist during mentalizing. We here combined functional magnetic resonance imaging (fMRI) and repetitive transcranial magnetic stimulation (rTMS) techniques with intranasal oxytocin (OTint) as a probe to examine the SOM effect in healthy male human participants, during which they attributed the cognitive states of decision confidence to an anonymous partner. Our results showed that OTint facilitated SOM via the left temporoparietal junction (lTPJ), but did not affect neural representations of internal information about others' confidence in the dorsomedial prefrontal cortex, which might be dedicated to SOD, although the two brain regions, importantly, have been suggested to be involved in mentalizing. Further, the SOM effect induced by OTint was fully mediated by the lTPJ activities and became weakened when the lTPJ activities were suppressed by rTMS. These findings suggest that the lTPJ might play a vital role in mediating SOM during mentalizing.
SIGNIFICANCE STATEMENT Every human mind is unique. It is critical to distinguish the minds of others from the self. On the contrary, we often project the current mental states of the self onto others; that is to say, self–other mergence (SOM). The neural mechanism underlying SOM remains unclear. We here used intranasal oxytocin (OTint) as a probe to leverage SOM, which is typically suppressed during mentalizing. We revealed that OTint specifically modulated the left temporoparietal junction (lTPJ) neural activities to fully mediate the SOM effect, while suppressing the lTPJ neural activities by transcranial magnetic stimulations causally attenuated the SOM effect. Our results demonstrate that the lTPJ might mediate SOM during social interactions.
Introduction
During interpersonal interactions, we infer the cognitive states of others, such as their desires, thoughts, and beliefs, to predict and influence their behaviors (Frith and Frith, 2012; Baker et al., 2017). This process is so-called mentalizing (Carruthers, 2009). Unlike our own cognitive states, which are introspectable, others' cognitive states cannot be directly accessed (Carruthers, 2009). We instead rely on apparent associations between external social cues and internal cognitive states to make such social inferences (Lurz, 2011; Kliemann and Adolphs, 2018). Critically, we need to adopt others' perspectives to specify their cognitive states, which are distinct from ours; that is, self–other distinction (SOD), a hallmark of human social cognition (Frith and Frith, 2012; Silani et al., 2013).
On the contrary, we often attribute mental states to others similar to ours. This process is referred to as self–other mergence (SOM; Robbins and Krueger, 2005; Toma et al., 2010; Wittmann et al., 2016). The SOM effect during mentalizing has been found to crucially depend on social contexts (Robbins and Krueger, 2005; Toma et al., 2010; Wittmann et al., 2016). For instance, the SOM effect emerges when the partner is an in-group member or similar to us, but disappears when the partner is an out-group member or dissimilar to us (Robbins and Krueger, 2005; Toma et al., 2010). Thereby, the two processes of SOD and SOM often coexist in mentalizing. However, the neural basis underlying their coexistence and relationship remains unclear.
SOD necessitates that the cognitive states of the self and others are separately represented in the brain. The neural representations of others' cognitive states in mentalizing have been found prevalently in the following two core brain regions: the dorsomedial prefrontal cortex (dmPFC) and the temporoparietal junction (TPJ; Saxe and Kanwisher, 2003; Amodio and Frith, 2006; Schurz et al., 2014), which are separated from those representing one's own cognitive states (Jiang et al., 2022). However, the functional roles of the two regions involved in mentalizing might differ. The dmPFC might engage in the allocentric perspective by encoding others' unique cognitive states (Amodio and Frith, 2006; Schurz et al., 2014; Jiang et al., 2022), while the TPJ might suppress the egocentric perspective taking to facilitate SOD (Schurz et al., 2013; Silani et al., 2013; Sowden et al., 2015; Cui et al., 2023; Li et al., 2023). Excitation of the TPJ activities by direct current stimulation strengthens the suppression of egocentric perspective taking (Martin et al., 2019), while anatomic or virtual lesions in the TPJ weaken the suppression of egocentric perspective taking; that is, enhancement of SOM (Samson et al., 2004; Heinisch et al., 2011; Soutschek et al., 2016). Hence, SOM in mentalizing might be mediated by the TPJ. However, this has not yet been formally tested, partially because SOM is typically suppressed in mentalizing. To investigate SOM, we then need an effective probe to modulate the TPJ activities to leverage SOM.
Administrations of intranasal oxytocin (OTint), a hormone that promotes social bonds between the self and others, have been reported to produce prosocial effects in a variety of social functions during interpersonal interactions (Kosfeld et al., 2005; Domes et al., 2007; De Dreu et al., 2010; Insel, 2010; Meyer-Lindenberg et al., 2011; Davis et al., 2013; Neto et al., 2020). Such prosocial effects might be achieved through SOM (Tomova et al., 2019; Yue et al., 2020). For instance, OTint enhances in-group cooperation and out-group competition (De Dreu et al., 2010). Notably, prior studies have shown that the TPJ activities can be selectively modulated by OTint (Mascaro et al., 2014; Lancaster et al., 2015; Hu et al., 2016; Lee et al., 2018; Wu et al., 2020). Thus, OTint could be used as a probe to modulate the TPJ activities in mediating SOM during mentalizing. In the current study, we combined OTint, functional magnetic resonance imaging (fMRI), and offline repetitive transcranial magnetic stimulation (rTMS) techniques to test the hypothesis that the TPJ plays a critical role in mediating SOM during mentalizing.
Materials and Methods
Participants
We recruited a total of 100 healthy right-handed male human participants in three experiments [mean ± SD; experiment 1 (Exp1; behavior): n = 40; mean age, 24.9 ± 2.3 years; Exp2 (fMRI): n = 30; mean age, 25.3 ± 2.3 years; Exp3 (rTMS): n = 30; mean age, 23.8 ± 2.9 years; Extended Data Fig. 1-1A, a priori power analysis]. Only male participants were voluntarily recruited in the current study to avoid gender-related confounding effects and possible interactions with gonadal steroids. All the participants were free of neurologic and psychiatric disorders, medications, smoking cigarettes, drug and alcohol abuse, nasal diseases such as rhinitis and nasal allergies, and respiratory diseases. Before the experiments, each participant was informed about the possible side effects of oxytocin and gave written informed consent. The experiments were conducted in accordance with the protocol approved by the Ethics Committee of the Institutional Review Board of Institute of Biophysics, Chinese Academy of Science.
Experimental paradigm
We designed three baseline-controlled, within-participant, double-blinded or single-blinded experiments (double-blinded in Exp1 and Exp2; single-blinded in Exp3; Fig. 1A). Each participant performed the same set of tasks twice. These tasks were pseudorandomly intermingled and counterbalanced across participants in each session. Forty minutes before the tasks, each participant received OTint of 24 international units or intranasal placebo (PLint) containing the same components other than the neuropeptide. The two sessions were separated by a week for the sake of the metabolism of oxytocin.
Experimental stimuli.
The present study involved a perceptual decision-making task in which participants were required to judge the net direction of random dot motion (RDM). In each trial, the RDM stimuli were presented within an aperture with a radius of 3° (visual angle), consisting of three hundred white dots (radius, 0.08°; density, 2.0%) moving in different directions at a speed of 8.0°/s on a black background. Each dot had a life span of three frames. A fraction of the dots moved coherently in one of the four directions (left, down, right, or up), while the remaining dots moved in random directions. The coherence of RDM stimuli was defined as the percentage of dots moving coherently.
Metacognition task.
In the metacognition task (Fig. 1B, left), participants were required to report their own confidence in a decision based on the RDM stimuli. In each trial, the participant perceived the RDM stimuli surrounded by four direction arrows (left, down, right, and up) and made a judgment on the moving direction of the coherent dots in 3 s. The elapsed time since stimulus onset (called the progress bar) was simultaneously displayed at the bottom of the screen, with a color gradient indicating the elapsed time, as follows: green for 1 s; yellow for 2 s; and red for 3 s. After making the decision, participants were asked to immediately rate their confidence within 2 s with regard to their level of belief in that the decision was correct. The confidence ratings were on a scale from 1 to 8, where 1 represented the lowest confidence level and 8 indicated the highest confidence level. Before the experiment, the RDM coherence was titrated individually and maintained at a constant level throughout the experiment, ensuring that the decision accuracy rate converged to 0.5 for each participant.
Mentalizing task.
In the mentalizing task (Fig. 1B, middle), a pair of participants simultaneously engaged, with one individual acting as the partner and performing the metacognition task, while the other participant assumed the role of the observer and performed the mentalizing task. Before the experiment, the pair of participants convened in the experimental space and were subsequently physically segregated into separate rooms to ensure visual isolation while remaining aware of each other's actual presence. Note that during the fMRI experiment (Exp2), the observer was positioned inside the scanner, while the partner was outside the scanner. The two computers used for task presentations were interconnected through an Ethernet connection to synchronize the tasks performed by the participant pair. Specifically, the observer was required to infer the anonymous partner's decision confidence, which was reported by the partner while performing the metacognition task but was concealed from direct observation by the observer. Notably, as each observer had previously fulfilled the role of the partner in the mentalizing task, they possessed a comprehensive understanding of the partner's task context when assuming the role of observer.
Within each trial of the mentalizing task, the observer's task sequence was synchronized with the metacognition task performed by the partner. During the decision-making stage, the partner perceived an RDM stimulus and made a judgment on the moving direction of the RDM stimuli in 3 s; meanwhile, the participant observed the synchronized visual stimuli and a synchronized time bar positioned below it. To prevent the observer from evoking his own confidence regarding the same RDM stimulus that the partner was perceiving, we presented the observer with a noiseless RDM stimulus, in which only the coherently moving dots remained in motion, while the others were stationary. When the partner made a decision, the progress bar halted, allowing the observer to precisely track the partner's response time (RT). Notably, the partner's choice was not revealed to the observer. In the confidence-rating stage, the partner was required to rate his own decision confidence within 2 s as in the metacognition task, while the observer was asked to infer the partner's decision confidence (called the inferred partner's confidence) within the same duration. The partner's reported confidence was not revealed to the observer. Hence, the only external information that the observer could use to infer the partner's decision confidence was the RT. This task was instructed as follows:
“In this task, you are required to observe another participant performing a metacognition task and then infer his decision confidence. You will observe the movement of dots in the same proportion as the partner, while the remaining random dots will remain stationary. At this stage, your role is purely observational. When the partner makes a decision, the time bar will stop, allowing you to know his response time. As you engage in your own decision-making, you may experience that the longer response time corresponds to the lower confidence level. However, it is crucial to acknowledge that everyone has unique thought processes. Therefore, when assessing the partner's decision confidence, you should consider, 'If I were in their shoes, what level of confidence would I attribute?' After observing the partner's decision-making process, you will need to report the inferred confidence level of the partner within 2 seconds.”
To avoid individual differences across partners in task performance, we used the predetermined and identical task dataset in the mentalizing task for all observers, including the RT presented to the observer. Such a task dataset was accomplished by a participant who exhibited a high metacognitive ability, ensuring that the trial-by-trial reported confidence in the dataset exhibited high consistency with the actual decision correctness.
Association task.
An association task (Fig. 1B, right) was implemented as a control task alongside the mentalizing task. The task sequence for the association task was identical to the mentalizing task, with the only distinction being the instructional content. In this task, participants were instructed to estimate the probability of decision accuracy for a computer algorithm that used stimulus information to make judgments on the RDM stimuli, with its decision-making process incorporating some random noise. Thereby, any observed behavioral and neural differences between the mentalizing and association tasks were expected to originate from the underlying mentalizing process. The participants were instructed as follows:
“For this task, we have developed an algorithm that can generate responses to movement judgments, but its accuracy is not 100% certain. Whether the generated response is correct or not depends on the time it takes to generate the response. You need to observe the time taken by the computer to generate its response to the movement judgment and estimate the correct probability of the computer. After observing the response time, you rate your estimation of the computer's accuracy level within 2 seconds.”
Pretraining.
To stabilize their abilities in the metacognition task, the participants were trained before the experiments. After sufficient practice using a staircase procedure, by which RDM coherence was upgraded one level after two continuous correct trials, downgraded one level after two continuous erroneous trials, and otherwise remained unchanged, RDM coherence gradually decreased and later remained at a stable level (mean coherence, 8.2; Extended Data Fig. 1-2A), and the decision accuracy rate was close to 0.5 for each participant (chance level, 0.25; Extended Data Fig. 1-2B). Furthermore, the metacognitive ability [area under the curve (AUC); see subsection Behavioral data analyses] also became stable at the late practice phase (mean, 0.64; Extended Data Fig. 1-2C).
Besides, before the experiments, each participant practiced the three tasks, each consisting of 60 trials with feedback, so that the participant could quickly learn the RT–confidence (accuracy) associations that were used to infer the partner's confidence in the mentalizing task and estimate the accuracy of the computer in the association task. Except for feedback, other settings in pretraining are the same as in the formal experiments. The RT–confidence (accuracy) associations in the pretraining remained stable in the metacognition task (mean correlation, −0.18; Extended Data Fig. 1-2D), mentalizing task (mean correlation, −0.31; Extended Data Fig. 1-2E), and association task (mean correlation, −0.35; Extended Data Fig. 1-2F).
Procedure.
In the behavioral (Exp1) and fMRI (Exp2) experiments, each participant received the pharmacological administration twice with an interval of 1 week. One was the oxytocin treatment, and the other was the placebo treatment. The types of treatments remained blind to both the participants and the experimenters until the end of the study. In the TMS experiment (Exp3), each participant also received the administration twice with an interval of 1 week, but the same exogenous oxytocin was administered in each session. The treatments only remained blind to each participant. Notably, the participant who played a role as a partner did not take any intranasal administration in each experiment.
To counterbalance the order of the tasks and the elapsed time of each task since the onset of OTint or PLint administration, the participants were randomly and evenly assigned to four different task sequences (Metacognition → Mentalizing → Association; Metacognition → Association → Mentalizing; Mentalizing → Association → Metacognition; Association → Mentalizing → Metacognition). Each participant encountered the same task sequence in both sessions. There were two consecutive runs, each consisting of 60 trials in each task. Each task lasted 15 min. All the procedures were identical in the three experiments.
In each session, the participants abstained from food and drink (other than water) for 2 h before the tasks. As an apparent increase in oxytocin levels in CSF has been reported at ∼40 min after oxytocin administration in humans and monkeys (Porges and Carter, 2011, chapter 4; Striepens et al., 2013; Lee et al., 2018), the tasks were started 40 min after each intranasal administration and lasted for 45 min.
Oxytocin/placebo administration.
Exogenous oxytocin was administered by nasal spray. In each treatment, three puffs (each consisting of 4 IU in the case of oxytocin) per nostril with 30 s between each puff were sprayed using a standard protocol (Guastella et al., 2013). The placebo treatment was provided in the same type of bottle and contained the same ingredients other than the neuropeptide oxytocin (Sichuan Meike Pharmacy). In postexperiment interviews, the participants could not distinguish whether they had received the oxytocin or placebo treatment. No participants reported side effects after the experiments.
TMS protocol
Following the continuous theta-burst (cTBS) protocol (Huang et al., 2005), a continuous train of “theta burst” with three TMS pulses (biphasic waveform, width, 0.1 ms) at 50 Hz was repeated at the internals of 0.2 s for 200 times using a Magstim Rapid2 stimulator with a 70 mm figure-eight coil (Magstim), consisting of a total of 600 pulses at an intensity of 80% resting motor threshold (RMT). RMT was determined for each participant as the lowest TMS intensity, which can evoke at least 5 of 10 electromyographic (EMG) pulses with an amplitude of >50 μV peak-to-peak in the relaxed first dorsal interosseous muscle of the right hand. Each participant wore a pair of earplugs to reduce noise from the TMS delivery system. The target position was precisely localized on each individual anatomic brain structure by the navigation system (Brainsight, Rogue Resolutions) in reference to the left temporoparietal junction (lTPJ) region [Montreal Neurologic Institute atlas based on 152 structural images (MNI152) coordinates: x = −56, y = −70, z = 18] defined by the conjunction analysis on the fMRI data (see Fig. 4D).
Each participant was administered in two sessions with a gap of 1 week. One was the cTBS session (the true rTMS condition in which the coil surface was placed perpendicular to the lTPJ region), and another was the Sham session (the false rTMS condition in which the coil surface was placed parallel to the lTPJ region). The two conditions were randomly scheduled and counterbalanced across the participants. To preclude the participants' awareness of the differences between the cTBS and Sham administrations (i.e., TMS placebo effects), we instructed each participant that the experiment would be aimed to test the TMS effects on the different (oxytocin and placebo) pharmacological interventions, and the intranasal administrations of oxytocin and placebo were blinded to both the participant and the experimenters (as assumed to be the same as the protocol used in the previous two experiments), while the TMS protocols in the two sessions were identical. For this purpose, the Sham control condition should be better than the active control condition by stimulating a different target region. In postexperiment interviews, no participants reported any side effect after the experiment. All participants reported no awareness of the differences between the two TMS administrations.
Statistical analyses
Statistical analyses of the data were performed with the MATLAB statistical toolbox (Matlab2015b, MathWorks). One-way or two-way repeated ANOVAs were used to test for significance. Pearson's correlation was used to measure the associations among internal mentalizing ability, internal metacognitive ability, and neural activities, and two-tailed t tests were used to test for significance unless otherwise mentioned. Two-sample Kolmogorov–Smirnov tests were used to compare the populations of reported confidence or accuracy with α = 0.05.
Behavioral data analyses
Metacognitive and mentalizing abilities were assessed by the consistency between the reported self's or inferred partner's confidence and the self's or partner's actual decision correctness, respectively. In doing so, we constructed the receiver operating characteristic (ROC) curve (Fleming et al., 2010) by characterizing the correct probabilities with different confidence levels as thresholds. The AUC of the ROC curve was then measured as a proxy of the participant's metacognitive and mentalizing abilities. A larger AUC indicates a higher metacognitive or mentalizing ability.
Across the three tasks, the reported self's confidence (metacognition task), the inferred partner's confidence (mentalizing task), and the estimated accuracy of the computer (association task) were associated with the RTs used for decision-making on the RDM stimuli (Fig. 2A), which was the only external cue information that could be used to predict the partner's confidence and the accuracy of the computer. Though RTs could be used to predict decision correctness commonly in all three tasks (Fig. 2A, bottom, blue area under the ROC curve), metacognition and mentalizing might involve additional components (Fig. 2A, bottom, gray and red regions under the ROC curves). For instance, the RTs were not the only source of the self's confidence during metacognition, as the decision-making process could be directly inspected. We defined the remaining component after regressing out the RT-associated component from the reported confidence/accuracy as the residual confidence/accuracy. We also calculated the residual AUCs by using the residual confidence/accuracy. The residual AUCs thus represent the internal metacognitive and mentalizing abilities. Hence, we could separately examine the OTint modulation effects on the RT-associated components and the residual confidence/accuracy. Specifically, the OTint effects on leveraging the influence of the internal components of the metacognitive ability on the internal components of the mentalizing ability (i.e., SOM) were the main focus of the current study.
To exclude the RT-associated components, we used a variety of models as a function of the RTs to fit the reported confidence/accuracy in the three tasks and selected the winning model (Extended Data Fig. 3-1). These models consisted of different (i.e., linear, quadratic, triple, biquadratic, and quantic) series functions with the RTs. To account for potential biases in individual perception regarding moving directions, some models also contained the components of moving direction biases and their interactions with the RTs. We then used the Bayesian information criterion (BIC) to determine the winning model. The linear model was unanimously the best for each task.
Modulation effects on SOM by OTint.
To evaluate the SOM effect induced by OTint through projecting the metacognitive experiences onto the intrinsic confidence state attribution to an anonymous human partner or a computer, we modeled the residual AUCs in the mentalizing and association tasks under the OTint and PLint treatments as a linear function of the residual AUCs, as follows:
Power analyses.
We conducted a priori power analysis to determine the sample size necessary for the current study and a post hoc power analysis to evaluate the power under the effect size acquired in the current study using G*Power (version 3.1.9.228; Erdfelder et al., 1996). The main behavioral measure for the current study was the SOM effect; that is, the associations between the internal mentalizing and metacognition abilities in the OTint and PLint treatments. We estimated the sample size with the a priori effect size of Pearson's correlation coefficient (r) as 0.4 under the power of 0.95. The required sample size was 70 (Extended Data Fig. 1-1A). Further, we also estimated the post hoc power under the empirical effect size (r = 0.44) with a sample size of 40 in Exp1 only, and 70 including Exp1 and Exp2 (Extended Data Fig. 1-1B).
fMRI data acquisition and analysis
All fMRI experiments were conducted using a 3 T Siemens Prisma MRI system with a 20-channel head coil (Siemens). Functional images were acquired with a single-shot gradient echo T2* echoplanar imaging (EPI) sequence (repetition time, 1 s; echo time, 30 ms; slice thickness, 3.0 mm; in-plane resolution, 3.0 × 3.0 mm2; field of view, 192 × 192 mm2; flip angle, 90°). Forty-eight axial slices were taken, with the interleaved acquisition parallel to the anterior commissure–posterior commissure line.
The fMRI analyses were conducted using the FMRIB Software Library (FSL; Smith et al., 2004). A standard preprocessing procedure was performed, as described in the previous study (Jiang et al., 2022). Briefly, all EPI images were realigned to the first volume of the first scan, transformed to the MNI space, resampled with a resolution of 2 × 2 × 2 mm, spatially smoothed with a 4 mm Gaussian kernel (full-width at half-maximum), then filtered with a cutoff of 0.005 Hz.
Whole-brain analyses
We first identified the voxel-wise neural correlates of different components of the reported confidence/accuracy (RTs and residual confidence/accuracy) separately in the three tasks across the whole brain. We used general linear modeling (GLM) to analyze the fMRI data. For the first-level analysis, two events were modeled in each trial: one event was time locked to the stimulus presentation, named “the stimulus event,” and another one was time locked to the onset of confidence rating with the duration of the rating time, named “the rating event.” The parametric modulations by the RTs and the residual confidences were concurrently added to the rating event (Qiu et al., 2018). All the regressors were convolved with the canonical hemodynamic response function (HRF) with a double-γ kennel. For the group-level analyses, we used the FMRIB local analysis of mixed effects (FLAME), which models both “fixed effects” of within-participant variance and “random effects” of between-participant variance using Gaussian random-field theory. Statistical parametric maps were generated by the threshold with z > 2.6 and p < 0.05 after cluster-level familywise error (FWE) correction for multiple comparisons for each contrast, unless mentioned otherwise.
The neural correlates of RTs were displayed in Extended Data Figure 3-3E for the metacognition task and in Extended Data Figure 4-2A for the mentalizing task and association task (z > 2.6, p < 0.05, cluster-level FWE correction). The neural correlates of residual confidence were displayed in Extended Data Figure 3-3F for the metacognition task and in Figure 4A for the mentalizing task and association task (z > 2.6, p < 0.05, cluster-level FWE correction).
Neural residual AUC
Similar to calculating the behavioral residual AUCs, we also conducted voxel-wise calculations of neural residual AUCs using the trial-by-trial estimated neural signals, following the steps outlined below (Fig. 2B).
Step 1.
Calculating the voxel-wise trial-by-trial neural activities. In the GLMs, each phase of each trial had an independent β parameter to be convoluted with the canonical HRF and linearly superposed together to fit with the preprocessed fMRI time series (Mumford et al., 2012). To obtain the trial-by-trial voxel-wise neural activities (
Step 2.
Calculating the voxel-wise residual AUCs. Like the behavioral analyses, we regressed out the linear RT-associated component from the voxel-wise trial-by-trial neural activities. Then we used the residual trial-by-trial neural activities as the judgment criteria to construct the ROC curve and calculated the neural residual AUC for each voxel.
Step 3.
Calculating the correlation between the neural residual AUCs and behavioral residual AUCs across the participants. At the group level, we then searched the brain regions where the voxel-wise neural AUCs correlated with the behavioral AUCs across participants. We first generated a contrast neural residual AUC map between mentalizing and association tasks (Human – Computer). We also calculated the differences in behavioral residual AUC between the mentalizing task and the association task (Human – Computer). Subsequently, at the whole-brain level, we computed the correlation between the aforementioned neural and behavioral residual AUCs across participants (Extended Data Fig. 5-1A), Additionally, we calculated whole-brain correlations between the differences in neural residual AUC of each voxel (Human – Computer) and the behavioral residual AUCs in the metacognition task (Extended Data Fig. 5-1B).
Conjunction analyses
We conducted conjunction analyses on the neural correlates of the RTs and the residual confidence across the conditions. We also made conjunction analyses on the AUC correlation maps between metacognition and mentalizing under the comparison between the OTint and PLint treatments. We identified the significant regions in which there was evidence of effects in all contrasts of conditions using the FSL script easythresh. The statistical parametric maps were generated by the threshold with z > 2.6 and p < 0.05 after cluster-level FWE correction.
Regions-of-interest analyses
We first defined the regions of interest (ROIs) separately for different tasks. In the metacognition task, we defined the RT-associated ROIs as the RT neural activation map exceeding the threshold (Extended Data Fig. 3-3E; z > 4.5, p < 0.05, cluster-level FWE correction) and the residual confidence-associated ROIs as the residual confidence neural activation map exceeding the threshold (Extended Data Fig. 3-3F; z > 4.5, p < 0.05, cluster-level FWE correction) from the conjunction analyses across the conditions (OTint/PLint), respectively (Extended Data Fig. 3-4A).
In the mentalizing and association tasks, we defined the RT-associated ROIs as the RT neural activation map exceeding the threshold (Extended Data Fig. 4-2A; z > 4.5, p < 0.05, cluster-level FWE correction) from the conjunction analyses across the conditions (OTint/PLint, Human/Computer). We also defined the residual confidence-associated ROI as the residual confidence neural activation map exceeding the threshold (Fig. 4A; z > 2.6, p < 0.05, cluster-level FWE correction) from the conjunction analyses across the conditions (OTint/PLint, Human). And we defined the correlation-associated ROI of the residual AUCs as the behavioral metacognitive AUCs and the behavioral metalizing AUCs correlation neural activation map exceeding the threshold (Fig. 4D; z > 2.6, p < 0.05 cluster-level FWE correction) from the contrast analyses between the OTint and PLint conditions, respectively (Extended Data Fig. 3-4B).
Additionally, we conducted analyses on the meta-analyzed ROIs from Neurosynth (https://www.neurosynth.org). We defined the dmPFC as the neural activation map found using the keyword “mentalizing” exceeding the threshold (z > 5, p < 0.01, cluster-level FWE correction) and the TPJ as the neural activation map found using the keyword “TPJ” exceeding the threshold (z > 5, p < 0.01, cluster-level FWE correction; Yarkoni et al., 2011; Extended Data Figs. 3-4C, 4-1A).
Then, we deconvoluted the mean time series averaged across the voxels in each ROI with the trial-by-trial models at the rating event and obtained the trial-by-trial neural activities (β). These β values were then used to calculate coefficients within the GLM with RTs and residual confidence. Then, the neural residual AUCs were calculated based on the residual β, which is the residual component after regressing out the RT-associated component from the β.
Mediation analyses
To analyze the mediation effects through the TPJ (neural AUCs, m) on the association between the behavioral residual AUCs in metacognition (10) and mentalizing (y; Fig. 5A), we constructed a linear structural equation model to conduct mediation analyses on the associations among these variables as follows (Imai et al., 2010):
To test whether the mediation effect was significant under each treatment and whether the mediation effect difference between the two treatments was significant, we separately made 100,000 simulations of the randomized orders of pairs between the lTPJ neural AUCs and the metacognitive residual AUCs across the participants as the null population and made statistical tests on the simulated data (Fig. 5C).
We also made the same mediation effect analyses on the other ROIs defined by the significant neural correlations of the neural AUCs in the mentalizing task with the behavioral AUCs in the mentalizing task (Extended Data Fig. 5-1A) and in the metacognition task (Extended Data Fig. 5-1B) in the OTint and PLint treatments (Fig. 5D).
Bootstrap test on the TMS effect
We conducted a bootstrap test to validate the TMS effect. First, we used a bootstrapping procedure to create a distribution of the correlation between the residual AUC in the metacognition task and the residual AUC in the mentalizing task across participants in the OTint condition. On each iteration, we collapsed the data from all three experiments (n = 100) and randomly selected a subset of 30 participants of the total 100 participants in the OTint treatment, and then calculated the correlation coefficient between the mentalizing residual AUC and the metacognition residual AUC within this selected subset of participants. We repeated this procedure 100,000 times. Subsequently, we tested whether our empirical correlation results in the Sham condition and the true TMS condition were drawn from the same distribution as the one in the OTint condition of Exp1 and Exp2.
Results
One hundred healthy male human participants totally took part in a set of three baseline-controlled, within-participant, double-blinded, or single-blinded experiments (double blinded in Exp1 and Exp2; single blinded in Exp3; Fig. 1A). The cognitive state to be inferred in the tasks of the current study was decision confidence, a belief in that the decision was correct. The process during which the participant estimated his own decision confidence is referred to as metacognition. Correspondingly, the process during which the participant inferred a partner's decision confidence in a partner's decision is referred to as mentalizing. Each participant took part in both metacognition and mentalizing tasks twice by the OTint or PLint treatment. We examined whether the participant might use the self's metacognitive experiences to facilitate the inference of the partner's decision confidence during mentalizing; that is, the SOM effect that we measured in the current study.
Experimental paradigm and task settings. A, Experimental paradigm. B, The set of tasks used in the three experiments. The power analyses were conducted to determine the number of participants used in the experiment (Extended Data Fig. 1-1). The performance of pretraining is shown in Extended Data Figure 1-2.
Figure 1-1
The power analysis. A, a prior power analysis under different effect size [the correlation of the residual AUCs between the mentalizing task (human – computer) and the metacognition task] to estimate the necessary sample size. B, The post hoc power analysis under the sample size as used in the behavioral experiment (n = 40) and both Exp1 and Exp2 (n = 70). Download Figure 1-1, TIF file.
Figure 1-2
The performance of pretraining. A, The RDM stimulus coherence changed with the training trials and was eventually stabilized. B, The participants' performance accuracy rate was gradually reduced with the training trials and was eventually stabilized close to the control level. C, The AUC changed with the training trials and was eventually stabilized. Shading indicates the SEM across participants (n = 100), and the gray lines indicate the eventually converging levels. D, The correlation between RT and confidence was stable in pretraining of metacognition task (mean correlation, –0.18). E, Same as D in mentalizing task (mean correlation, –0.31). F, Same as D in association task (mean correlation, –0.35). Download Figure 1-2, TIF file.
Our hypotheses in this study are specified as follows (Fig. 2) under the PLint condition: (1) the residual confidence during mentalizing would not predict the partner's decision correctness; (2) the internal metacognitive ability and the internal mentalizing ability would be uncorrelated; and (3) the neural representations of the residual confidence of the self and the partner would be separated, reflecting their distinction (SOD). Conversely, under the OTint condition, (1) the residual confidence during mentalizing might predict the partner's decision correctness; (2) the internal mentalizing ability might become correlated with the internal metacognitive ability, indicating the SOM effect; and (3) OTint might specifically modulate the TPJ activities to induce the SOM effect.
The framework of behavioral and neural analyses. A, The schematic descriptions of behavioral analyses. B, The schematic descriptions of neural analyses.
OTint selectively and reliably induced SOM in mentalizing
To examine the impact of OTint on the participants' mentalizing abilities in attributing decision confidence to an anonymous human partner in the mentalizing task and in attributing decision accuracy to a computer in the association task, we assessed the participants' mentalizing abilities (behavioral raw AUC; Fig. 2A). The behavioral raw AUCs in both Exp1 and Exp2 were significantly greater than the chance level (0.5), but there were no differences across the four conditions (Extended Data Fig. 3-2C; two-way repeated ANOVA; Exp1: OTint/PLint: F(1,39) = 0.037, p = 0.85; human/computer: F(1,39) = 0.002, p = 0.96; interaction: F(1,39) = 1.18, p = 0.28; Exp2: OTint/PLint: F(1,29) = 3.42, p = 0.07; human/computer: F(1,29) = 0.78, p = 0.38; interaction: F(1,29) = 0.043, p = 0.84).
We then analyzed the impact of OTint on the components of the estimated partner's confidence. We first assessed the extent to which the estimated partner's confidence was associated with the RTs. In our experimental design, since the RTs were the only external cue available to predict whether the partner's or computer's decision was correct, it is important to assess the OTint effect in the RT association. To do so, we conducted GLM analyses, which showed that the RT-associated contributions were stable and similar across the four conditions (Extended Data Fig. 3-2D, two-way repeated-measures ANOVA; Exp1: OTint/PLint: F(1,39) = 0.36, p = 0.55; human/computer: F(1,39) = 0.002, p = 0.96; interaction: F(1,39) = 2.23, p = 0.14; Exp2: OTint/PLint: F(1,29) = 0.03, p = 0.86; human/computer: F(1,29) = 0.49, p = 0.49; interaction: F(1,29) = 0.005, p = 0.95).
Furthermore, to assess participants' internal mentalizing abilities that were independent of the external information, we regressed out the RT-associated component from the estimated decision confidence/accuracy and calculated the residual AUCs. Because of the lack of further social information, it was expected that the residual AUCs would be close to the chance level. In the behavioral experiment, the residual AUCs in both the mentalizing and association tasks were indeed at the chance level in the PLint treatment (one-way ANOVA; other: F(1,39) = 0.44, p = 0.51; computer: F(1,39) = 3.51, p = 0.068; Fig. 3A). However, the residual AUCs in inferring the human partner's confidence were larger than the chance level in the OTint treatment (one-way ANOVA; F(1,39) = 19.80, p = 0.000069; Fig. 3A), and also larger than that in the PLint treatment (one-way ANOVA; F(1,39) = 7.19, p = 0.011), while the residual AUCs in inferring the computer's accuracy were not different between the two treatments (one-way ANOVA; F(1,39) = 0.026, p = 0.87). Therefore, OTint appeared to selectively enhance the internal mentalizing abilities only when the target was a human partner rather than a computer.
OTint selectively and reliably induced SOM in mentalizing. A, The mentalizing residual AUCs in Exp1. The solid and open bars denote the original mentalizing residual AUCs and those after SOM was regressed out, respectively. ns, No significance, *p < 0.05, ***p < 0.001. Error bars indicate the SEM across participants. B, The mentalizing residual AUCs in Exp2. C, SOM was induced by OTint, but not PLint in Exp1. The correlations between the metacognitive residual AUCs and the mentalizing residual AUCs (human – computer) in the OTint and PLint treatments, respectively. D, SOM in Exp2. E, Positive social projection to the human partner induced by OTint, but not by PLint. F, Negative social projection to the computer induced by OTint, but not PLint. The SOM effect induced by OTint using different models characterizing the associations with the RTs was shown in Extended Data Figure 3-1. The OTint modulation effect on the RT-associated components in mentalizing was shown in Extended Data Figure 3-2. The OTint modulation effect in decision-making and metacognition was shown in Extended Data Figure 3-3. The definition of ROIs in Extended Data Figure 3-3 was shown in Extended Data Figure 3-4.
Figure 3-1
The reliability of the OTint modulation effect on social projection was independent of models as a function of RTs fitting with the inferred confidence and accuracy in the mentalizing and association tasks, respectively. A, The mentalizing residual AUCs were measured by the residual confidence after the RT-associated components as different functions of RT were regressed out for the data collapsed from both the behavioral and fMRI experiments (n = 70). Although the enhancement of internal mentalizing abilities was not always significant (left column), the associations between the differences of mentalizing abilities (mentalizing residual AUCs), and the internal metacognitive abilities (metacognitive residual AUCs) were stable for all the models in the OTint treatment, but not in the PLint treatment. ns, No significance, *p < 0.05, **p < 0.01. Error bars indicate the SEM across participants. B, The models are listed details in the table. Download Figure 3-1, TIF file.
Figure 3-2
OTint did not alter the participants' behavioral correlates of the association with the RT information in mentalizing. A, The distributions of the confidence ratings in the metacognition task. B, Same as A in the mentalizing and association tasks. C, The raw AUCs measured by the ratings of reported confidence/accuracy. D, The regression β values of the reported confidence/accuracy with RT. Download Figure 3-2, TIF file.
Figure 3-3
OTint did not alter the participants' behavioral and neural correlates of decision-making and metacognition. A, The RDM performance accuracy rates. B, The raw AUCs measured by the reported confidence. C, The regression β values of confidence with RTs. D, The residual AUCs measured by the residual confidence after the RT-associated component was regressed out. E, The whole-brain activation map in association with RTs commonly across the OTint/PLint treatments. F, Same as E with residual confidence. G, The regression β values with RTs in the significantly activated brain regions in E. H, The regression β values with residual confidence in the significantly activated brain regions in F. Download Figure 3-3, TIF file.
Figure 3-4
Definition of ROIs. A, Definition of ROIs in the metacognition task. B, Definition of ROIs in the mentalizing and association task. C, Definition of ROIs in meta-analyses. Download Figure 3-4, TIF file.
We then tested whether the facilitation of mentalizing by OTint was undergone through projection from the self's metacognitive experiences onto others' confidence attribution, even when the partner was a stranger (Krueger, 2007). To do so, we calculated the correlation between the residual AUCs in the mentalizing/association task and in the metacognition task under the OTint and PLint treatments, respectively. The analyses of the behavioral data in Exp1 revealed a trivial correlation between internal mentalizing and metacognitive abilities in the PLint treatment (r = −0.039; two-tailed t test: t(38) = −0.24, p = 0.81; Fig. 3C), confirming that SOM was suppressed in mentalizing, because the human partner was anonymous (Krueger, 2007). However, a significant correlation between the two internal abilities emerged in the OTint treatment (Pearson's r = 0.44; two-tailed t test: t(38) = 2.99, p = 0.0048). Further, the difference between correlations in the two treatments was also significant (two-tailed t test: t(38) = 3.20, p = 0.0014). Although the enhancement of the participants' mentalizing abilities by OTint was not repeated in Exp2 with another independent population of participants (Fig. 3B), the SOM effect (a high correlation between the residual AUCs in the mentalizing and metacognition tasks) induced by OTint was robust and was repeatedly observed in Exp2 (Fig. 3D). The post hoc power of the empirical association between internal mentalizing and metacognitive abilities (r = 0.44) under the current sample size (n = 40 and 70) was estimated to be as large as 0.98 and 0.99, respectively (Extended Data Fig. 1-1B). Notably, OTint did not affect the participants' performance in decision-making or metacognition [Extended Data Fig. 3-3; one-way repeated ANOVA; accuracy rate: Exp1: F(1,39) = 0.38, p = 0.54; Exp2: F(1,29) = 2.99, p = 0.09; raw AUC: Exp1: F(1,39) = 1.50, p = 0.23; Exp2: F(1,29) = 0.26, p = 0.61; regression with RT: Exp1: F(1,39) = 0.11, p = 0.74; Exp2: F(1,29) = 1.44, p = 0.24; residual AUC: Exp1: F(1,39) = 1.66, p = 0.21; Exp2: F(1,29) = 0.002, p = 0.97; neural correlates with RT: intraparietal lobule (IPL): F(1,29) = 1.76, p = 0.19; inferior frontal junction (IFJ): F(1,29) = 0.01, p = 0.92; dorsal anterior cingulate cortex (dACC): F(1,29) = 0.04, p = 0.84; neural correlates with residual confidence: IFJ: F(1,29) = 0.82, p = 0.37; dACC: F(1,29) = 0.25, p = 0.62]. Hence, OTint selectively enhanced SOM, causing the internal mentalizing abilities to increase proportionally with the internal metacognitive abilities.
To illustrate this SOM effect, we regressed out the linear component associated with the self's metacognitive abilities (see Materials and Methods). We found that the selective enhancement of internal mental state attributions in Exp1 caused by OTint disappeared (Fig. 3A, open bars). Further, OTint differentially induced the SOM effect, positively toward the human partner, but negatively toward a computer (Fig. 3E,F). Notably, the SOM effect induced by OTint was robust even using different models characterizing the associations with the RTs (Extended Data Fig. 3-1).
The dmPFC selectively supported SOD in mentalizing
To reveal the underlying neural correlates of SOD and SOM, we analyzed the fMRI data acquired in Exp2. Based on our hypotheses, we first searched for the internal mental state representations that were uncorrelated with the external information (RTs); that is, neural correlates with residual confidence. If the neural representations of the partner's residual confidence were separated from those of the self's residual confidence, then these separated neural representations should reflect SOD. To do so, we conducted whole-brain voxel-wise GLM analyses on the neural correlates of residual confidence in the mentalizing task.
The residual confidence during mentalizing was negatively correlated with the neural activities in the dmPFC (the conjunction analysis between the two treatments: z > 2.6, p < 0.05; cluster-level FWE correction; Fig. 4A), which was separated from the brain areas during metacognition (the conjunction analysis between the two treatments: z > 2.6, p < 0.05, cluster-level FWE correction; Extended Data Fig. 3-3F), but not in the TPJ (defined by the conjunction analysis as described below; Fig. 4D). Further, the regression values in either region of the dmPFC or the TPJ had no significant differences across the four conditions (two-way repeated-measures ANOVA; dmPFC: OTint/PLint: F(1,29) = 0.14, p = 0.71; human/computer: F(1,29) = 0.41, p = 0.52; interaction: F(1,29) = 0.24, p = 0.63; Fig. 4B; TPJ: OTint/PLint: F(1,29) = 0.19, p = 0.69; human/computer: F(1,29) = 0.23, p = 0.65; interaction: F(1,29) = 0.20, p = 0.67; Fig. 4E). These results were repeated in the ROIs of dmPFC and TPJ independently defined by the neurosynth system (https://www.neurosynth.org) with the key word “mentalizing” (Extended Data Fig. 4-1; two-way repeated-measures ANOVA: dmPFC: OTint/PLint: F(1,29) =0.23, p = 0.63; human/computer: F(1,29) = 0.08, p = 0.78; interaction: F(1,29) = 1.49, p = 0.23; TPJ: OTint/PLint: F(1,29) = 0.08, p = 0.78; human/computer: F(1,29) = 0.10, p = 0.75; interaction: F(1,29) = 0.87, p = 0.36; Yarkoni et al., 2011). Hence, OTint did not alter the internal mental state representations in the dmPFC, which might selectively be involved in SOD in mentalizing.
Dissociated functional roles of the TPJ and dmPFC in mentalizing. A, The activation map in association with the residual confidence in mentalizing across the OTint and PLint treatments. B, The regression β values with the residual confidence in the dmPFC. ns, No significance, ***p < 0.001. Error bars indicate the SEM across participants. C, The dmPFC neural AUCs (human – computer) in association with the metacognitive residual AUCs (left), and the mentalizing residual AUCs (human – computer; right) in the OTint (red) and PLint (blue) treatments, respectively. D, The activation map jointly in association with the mentalizing residual AUCs (human – computer) and the metacognitive residual AUCs across participants (the conjunction analysis). E, The regression β values with the residual confidence in the lTPJ. ns, No significance. Error bars indicate the SEM across participants. F, The lTPJ neural AUCs (human – computer) in association with the metacognitive residual AUCs (left), and the mentalizing residual AUCs (human – computer, right) in the OTint (red) and PLint (blue) treatments, respectively. Additionally, we conducted similar analyses on the meta-analyzed ROIs from Neurosynth (https://www.neurosynth.org; Extended Data Fig. 4-1). And OTint did not alter the RT-associated components in mentalizing at the neural level (Extended Data Fig. 4-2).
Figure 4-1
Comparisons of neural correlates of residual confidence/accuracy in the mentalizing and association tasks between the OTint and PLint treatments. A, The ROIs of the dmPFC and TPJ defined by meta-analysis on the Neurosynth database with the key word “mentalizing” and “TPJ.” B, The regression β values were not different across the four conditions in the dmPFC or the TPJ. ns, No significance. Error bars indicate the SEM across participants. Download Figure 4-1, TIF file.
Figure 4-2
OTint did not alter the participants' neural correlates of the association with the RT information in mentalizing. A, The whole-brain activation map in association with RT commonly across the four conditions. B, OTint did not alter the brain regions whose activities were correlated with RT. There were no differences across the four conditions in the dACC, the IFJ, and the IPL. Download Figure 4-2, TIF file.
Similar analyses were also conducted on the neural correlates of RTs, but no OTint impacts were also found (Extended Data Fig. 4-2; two-way repeated-measures ANOVA; dACC: OTint/PLint: F(1,29) = 3.81, p = 0.06; human/computer: F(1,29) = 2.59, p = 0.12; interaction: F(1,29) = 0.006, p = 0.94; IFJ: OTint/PLint: F(1,29) = 0.025, p = 0.87; human/computer: F(1,29) = 0.35, p = 0.56; interaction: F(1,29) = 0.40, p = 0.53; IPL: OTint/PLint: F(1,29) = 1.67, p = 0.21; human/computer: F(1,29) = 1.78, p = 0.19; interaction: F(1,29) = 0.001, p = 0.97).
The lTPJ selectively supported SOM in mentalizing induced by OTint
We then searched for the neural signatures of SOM induced by OTint, which should concurrently comply with the following criteria: (1) the neural activities were associated with the internal metacognitive abilities (metacognitive residual AUCs); (2) the neural activities were associated with the internal mentalizing abilities (mentalizing residual AUCs); and (3) the neural activities mediate the association between the two internal abilities.
To do so, we first calculated the voxel-wise neural residual AUCs across the whole brain using the trial-by-trial fMRI signals in each voxel as the judgment criteria to predict whether the decision was correct or not, similar to computing the behavioral residual AUCs (Fig. 2B; see Materials and Methods). We then searched for two types of brain activations under the contrast between the OTint and PLint treatments, as follows: (1) the voxel-wise residual neural AUCs (human – computer) were significantly correlated with the residual behavioral AUCs (human – computer) in mentalizing across participants (z > 2.6, p < 0.05, cluster-level FWE correction; Extended Data Fig. 5-1A); and (2) the voxel-wise residual neural AUCs (human – computer) were significantly correlated with the residual behavioral AUCs in metacognition across participants (z > 2.6, p < 0.05, cluster-level FWE correction; Extended Data Fig. 5-1B). The two analyses separately identified a number of brain regions, the majority of which overlapped with the neural regions known to be associated with mentalizing (Zaki and Ochsner, 2012). A further conjunction analysis identified that the lTPJ was the only brain region in which the residual neural AUCs were positively correlated with both residual mentalizing and metacognitive AUCs (z = 2.6, p < 0.05, cluster-level FWE correction; Fig. 4D,F). By contrast, the dmPFC residual neural AUCs were not correlated with either residual mentalizing or metacognitive AUCs (Fig. 4C).
The lTPJ fully mediated SOM in mentalizing induced by OTint
We then tested whether the lTPJ activities could mediate SOM induced by OTint (Fig. 5A). The mediation analyses showed that the lTPJ activities (residual neural AUCs, human – computer) fully mediated the association between the internal mentalizing abilities (residual behavioral AUCs, human – computer) and the internal metacognitive abilities (metacognitive residual AUCs) in the OTint treatment, but not in the PLint treatment (Fig. 5B). The permutation analysis by randomly shuffling the order of pairs between the lTPJ neural AUCs and the metacognitive residual AUCs across participants further indicated that the difference in the lTPJ mediation effects between the OTint and PLint treatments was significant (p = 0.023; Fig. 5C). Notably, among these brain regions that were associated with either the internal mentalizing abilities or the internal metacognitive abilities, the lTPJ was the only brain region showing a significantly positive mediation effect on SOM (Fig. 5D). Hence, the SOM effect of sharing metacognitive experiences in mentalizing induced by OTint was fully mediated by lTPJ activities.
The lTPJ activities fully mediated SOM induced by OTint. A, Schematic diagram of the lTPJ mediation effect on SOM. B, The lTPJ mediation effect on SOM in the OTint (red) and PLint (blue) treatments, respectively. C, The null distribution of the mediation effect differences (c–c′) between the OTint and PLint treatments by randomly shuffling the order of pairs between the lTPJ neural AUCs and the metacognitive residual AUCs across the participants. The arrow indicates the mean, and the red line indicates the empirical value. D, The mediation effect difference between the OTint and PLint treatments in the brain regions associated with the mentalizing residual AUCs (human – computer) or the metacognitive residual AUCs across participants (Extended Data Fig. 5-1). The gray bars indicate the 95% confidence intervals by permutations. *p < 0.05. The dACC and IFJ neural activities during mentalizing did not mediate the association between the two internal abilities (Extended Data Fig. 5-2). vmPFC, ventromedial prefrontal cortex; PCC, posterior cingulate cortex; STS, superior temporal sulcus.
Figure 5-1
The neural correlates with behavioral AUCs. A, The whole-brain statistical map of the regression between neural AUCs (human – computer) and behavioral AUCs (human – computer) in mentalizing across participants (n = 30) between the OTint and PLint treatments. B, The whole-brain statistical map of the regression between neural AUCs (human – computer) and behavioral AUCs (metacognition) in metacognition across participants (n = 30) between the OTint and PLint treatments. vmPFC, Ventromedial prefrontal cortex; PCC, posterior cingulate cortex; STS, superior temporal sulcus. Download Figure 5-1, TIF file.
Figure 5-2
The metacognitive neural system was not involved in social projection from metacognitive experiences to mentalizing. A, The dACC and IFJ ROIs of the metacognitive neural system in the metacognition task (Extended Data Fig. 3-3F). B, Neither dACC nor IFJ activities were correlated with residual confidence in mentalizing. C, Neither dACC nor IFJ neural AUCs (human − computer) in mentalizing were correlated with the metacognitive residual AUCs. D, The dACC and IFJ neural AUCs (human − computer) were correlated with mentalizing residual AUCs (human − computer) negatively in the OTint treatment, but positively in the PLint treatment, which were reverse to those in the lTPJ. E, F, Neither dACC (E) nor IFJ (F) mediated social projection from metacognitive experiences to mentalizing. Download Figure 5-2, TIF file.
Alternatively, the share of metacognitive experiences in mentalizing might be evoked by concurrent reactivations of the metacognitive neural system, in which the participants might vicariously experience trial-by-trial decision confidence as in the metacognition task, and likewise, experience sharing in empathy (Singer et al., 2004; Zaki and Ochsner, 2012). However, this should not be the case. First, the noiseless stimuli presented to the participants should not evoke their vicarious experience of trial-by-trial decision confidence in the mentalizing task. Second, as a matter of fact, the activities during mentalizing in the metacognitive neural system, including the dACC and the IFJ (Extended Data Fig. 5-2A), were not correlated with the residual confidence (Extended Data Fig. 5-2B). Third, the dACC and IFJ neural residual AUCs were also not correlated with the internal metacognitive abilities in each condition (Extended Data Fig. 5-2C), but were negatively correlated with the internal mentalizing abilities under the contrast between the OTint and PLint treatments (Extended Data Fig. 5-2D). These correlations were reversed in the lTPJ. Fourth, the dACC and IFJ neural activities did not mediate the association between the two internal abilities (Extended Data Fig. 5-2E,F), a hallmark of sharing metacognitive experiences with mentalizing. Hence, the facilitated metacognitive experience sharing with mentalizing should not originate from concurrent vicarious mental state representations in the metacognitive neural system.
Suppressing the lTPJ activities by cTBS altered the SOM effect induced by OTint
Finally, to verify that the SOM effect induced by OTint was causally mediated by the lTPJ activities, we administered the cTBS protocol of offline rTMS (Huang et al., 2005), to attenuate the lTPJ activities on another independent population of participants (n = 30), and examined the behavioral consequences in mentalizing using this manipulation in Exp3. If the lTPJ was critical to mediate the SOM effect induced by OTint, then suppressing its neural activities by cTBS should causally reduce the SOM effect; that is, the association between the internal metacognitive abilities and the internal mentalizing abilities should be weakened (Fig. 6A).
The lTPJ activities causally mediated SOM induced by OTint. A, Schematic diagram of perturbations on the lTPJ mediation effect by rTMS. B, The SOM effect in the cTBS-OTint treatment (n = 30) or the no_cTBS-PLint treatment (n = 70; left), and in the Sham (n = 30) or no-Sham OTint treatments (n = 70; right). Gray lines indicate the SOM effect for all the participants across the three experiments, and dotted lines indicate the SOM effect for the participants only in Exp3. C, The distribution of the SOM effects by a bootstrap procedure of randomly drawing the data of 30 participants from the total population of 100 participants in the OTint treatment. The arrow indicates the mean (0.40). The green line indicates the empirical value in the cTBS-OTint condition (p = 0.014), and the yellow line indicates the empirical value in the Sham-OTint condition (p = 0.26). The TMS effect on the participants' performance in the metacognition task was shown in Extended Data Figure 6-1, and the confounding learning effect was shown in Extended Data Figure 6-2.
Figure 6-1
The TMS behavioral results. A, The behavioral results in comparisons between the Sham and cTBS conditions in the metacognition task. B, The behavioral results in comparisons between the Sham and cTBS conditions in the mentalizing and association tasks. ns, No significance, Error bars indicate the SEM across participants. Download Figure 6-1, TIF file.
Figure 6-2
The learning effect results. A, The behavioral results in comparisons between session 1 and session 2 in the metacognition task. B, The behavioral results in comparisons between session 1 and session 2 in the mentalizing and association tasks. C, SOM was induced by OTint, but not PLint in before and after sessions. The correlations between the metacognitive residual AUCs and the mentalizing residual AUCs (human – computer) in the OTint and PLint treatments, respectively. Download Figure 6-2, TIF file.
Our empirical results confirm this prediction. The SOM effect (the association between the two internal abilities) induced by OTint under the Sham condition (r = 0.30; one-tailed t test: t(28) = 3.1, p = 0.050; pooling all data of the three experiments together: r = 0.39, t(98) = 4.18, p = 0.000032; Fig. 6B, right) was disrupted by cTBS (r = 0.053, t(28) = 0.53, p = 0.39; Fig. 6B, left) and became similar to the PLint treatment without cTBS (pooling all data of the three experiments together: r = −0.023, t(98) = −0.19, p = 0.41). Differing from the SOM effect in the Sham condition (p = 0.26; Fig. 6C), the SOM effect in the cTBS condition was unlikely drawn from the normal OTint population (30 of 100 by a bootstrap test; see Materials and Methods; p = 0.014; Fig. 6C). By contrast, attenuations of the lTPJ activities by cTBS did not affect the participants' performance in the metacognition task (Extended Data Fig. 6-1A; one-way repeated-measures ANOVA; accuracy rate: F(1,29) = 0.12, p = 0.73; regression with RT: F(1,29) = 2.78, p = 0.11; raw AUC: F(1,29) = 1.51, p = 0.23; residual AUC: F(1,29) = 1.09, p = 0.31), the RT association, or the residual AUCs in both the mentalizing task and the association task in Exp3 (Extended Data Fig. 6-1B; two-way repeated-measures ANOVA; raw AUC: cTBS/Sham: F(1,29) = 2.79, p = 0.11; human/computer: F(1,29) = 3.05, p = 0.09; interaction: F(1,29) = 0.017, p = 0.90; regression with RT: cTBS/Sham: F(1,29) = 1.18, p = 0.29; human/computer: F(1,29) = 1.01, p = 0.32; interaction: F(1,29) = 0.016, p = 0.90; residual AUC: cTBS/Sham: F(1,29) = 0.55, p = 0.47; human/computer: F(1,29) = 1.10, p = 0.31; interaction: F(1,29) = 0.28, p = 0.60).
Finally, to assess the potential confounding learning effect, we compared the behavioral results between the two consecutive sessions (Extended Data Fig. 6-2). Despite that the RT-associated components were larger in session 2 than in session 1, the residual AUCs, which were the critical variable induced by OTint, remained stable across the two sessions in both the metacognition task (Extended Data Fig. 6-2A; one-way repeated ANOVA; accuracy rate: F(1,99) = 0.88, p = 0.35; raw AUC: F(1,99) = 0.0029, p = 0.96; regression with RT: F(1,99) = 7.90, p = 0.0060; residual AUC: F(1,99) = 0.99, p = 0.32) and the mentalizing tasks (Extended Data Fig. 6-2B; one-way repeated-measures ANOVA; raw AUC: human: F(1,99) = 0.43, p = 0.51; computer, F(1,99) = 1.54, p = 0.22; regression with RT: human: F(1,99) = 6.96, p = 0.0097; computer; F(1,99) = 10.84, p = 0.0014; residual AUC: human: F(1,99) = 0.020, p = 0.89; computer: F(1,99) = 0.000047, p = 0.99). Notably, even in each separated session, the SOM effect (the association between the internal mentalizing and metacognition abilities) induced by OTint was robust (Extended Data Fig. 6-2C). Note that the two conditions (OTint and PLint in Exp1 and Exp2, cTBS and Sham in Exp3) were randomly scheduled and counterbalanced across the participants. Therefore, even if the learning effect existed, it should not impact the current findings.
Discussion
OTint selectively mediates SOM but not SOD in mentalizing
SOD is a hallmark of mentalizing. It is necessary for proper social cognition and behaviors to distinguish the self's and others' mental states that are concurrently represented in the brain (Insel, 2010; Meyer-Lindenberg et al., 2011). However, SOM in considering others' cognitive mental states similar to ours is also often present during interpersonal interactions, particularly in prosocial behaviors seeking social affiliation and cooperation (Robbins and Krueger, 2005; Toma et al., 2010). It thus far remains unclear how the brain concurrently maintains the two processes during mentalizing, and whether the two processes are mutually competitive with each other (e.g., enhancement of SOM is indicated by suppression of SOD) or independent of each other (Robbins and Krueger, 2005; Wittmann et al., 2016). Meanwhile, OTint often promotes prosocial effects in social cognition, probably through SOM (Tomova et al., 2019; Yue et al., 2020). In this study, we used OTint as a probe to induce SOM when attributing decision confidence to an anonymous human partner. We then took advantage of this induced effect to investigate the neural mechanisms of SOM in mentalizing. Our results showed that OTint selectively and reliably induced SOM, although the effect of enhancement of internal mentalizing abilities by OTint was not reliable. By contrast, OTint did not affect SOD. The RT was the only external cue that could be used to estimate this partner's confidence, thereby representing the social cue salience. We confirmed that the RT-associated components were unanimously similar across different conditions in behaviors (Extended Data Fig. 3-2D) and brain activities (Extended Data Fig. 4-2). On the other hand, we also found that OTint did not affect residual confidence/accuracy, reflecting the internal state representations (Fig. 4A,B). Thereby, oxytocin did not modulate the two key components of mentalizing—the associations with the external cue (i.e., the social cue salience) and the internal mental state representations (Shamay-Tsoory and Abu-Akel, 2016). Further, OTint did not affect egocentric cognitive processes, such as decision-making and metacognition. Both the metacognitive abilities and the associated brain activities in the metacognition task were also not altered by OTint (Extended Data Fig. 3-3). Hence, the selective mediation of SOM, rather than SOD, by OTint suggests that SOM and SOD may independently coexist in the brain during mentalizing. Further studies are needed to comprehensively investigate the relationship between SOD and SOM by manipulating SOD alternatively and analyzing subsequent changes in SOM.
Dissociable functional roles of the TPJ and the dmPFC in mentalizing
Importantly, our findings implicate that the functional roles of the TPJ and the dmPFC in mentalizing should be dissociated (Fig. 7), although both regions have been found to prevalently represent others' cognitive states in mentalizing (Saxe and Kanwisher, 2003; Amodio and Frith, 2006; Schurz et al., 2014). In this study, we found that the dmPFC was involved in representing others' internal mental states of decision confidence in SOD, rather than SOM. By contrast, the TPJ was specifically involved in SOM by projecting the egocentric metacognitive experiences to the allocentric mental state attribution, but not representing others' internal mental states in SOD. This double dissociation suggests that the dmPFC might specifically subserve SOD to construct others' unique mental states, but the TPJ might instead mediate SOM, which was modulated by OTint. Critically, we found that suppressing the TPJ activities by cTBS attenuated the SOM effect. By contrast, suppressing the dmPFC activities by cTBS affects allocentric perspective taking and enhances the SOM effect in attributing the performance abilities to a human partner under a socially cooperative context (Wittmann et al., 2021). The division of labor between the two core brain regions involved in mentalizing thus supports the harmonious coexistence of SOD and SOM in mentalizing.
Schematic diagram of the OTint selective modulation effect in SOM during mentalizing. OTint selectively modulated the TPJ activities to induce SOM through social projection from metacognitive experiences to mentalizing, which was normally suppressed in mentalizing, without affecting the internal mental state representations in the dmPFC that is specified for SOD.
Notably, the manner of experience sharing through the TPJ during mentalizing induced by OTint was entirely different from empathy in that the neural representations of others' affective states overlapped with those of the self's affective states (Singer et al., 2004; Zaki and Ochsner, 2012). First, the internal mental state representations in attributing decision confidence to others were separated from those in metacognition when the participants experienced such decision confidence by themselves (Jiang et al., 2022). Second, the TPJ that mediated SOM induced by OTint represented neither the components of decision confidence attributed to others in mentalizing, nor the components of the self's decision confidence in metacognition. Third, metacognitive experience sharing was not realized by reactivations of the metacognitive neural system as vicariously experiencing such decision confidence during mentalizing. By contrast, it might rely on the memory system vivifying past experiences to complement allocentric perspective taking in mentalizing (Heinrichs et al., 2004; Buckner and Carroll, 2007; Spreng et al., 2009).
The relationship between SOM and human prosociality by OTint
Human prosociality broadly refers to a wide range of positive social behaviors, including trust, cooperation, and altruism (Donaldson and Young, 2008; Marsh, 2019; Marsh et al., 2021). Most of these prosocial behaviors involve benefitting the welfare of others at the potential cost of one's own resources. In other words, the rewards or losses to others, even strangers, are mentally similar to the self's rewards or losses. These prosocial behaviors might be achieved by the interactions between endogenous OT and the dopamine reward system (Hung et al., 2017). However, because mentalizing is usually irrelevant to social rewards, as in the current study, SOM in mentalizing might not be supported by endogenous OT and is thus usually suppressed. By contrast, the current study implicates that exogenous OTint might efficiently modulate SOM in mentalizing via a neural pathway associated with the TPJ.
The current findings suggest that the enhancement of egocentric experience sharing (SOM) in the TPJ rather than allocentric perspective taking (SOD) in the dmPFC might be at the root of human prosociality for social affiliation and cooperation induced by OTint (Kosfeld et al., 2005; Domes et al., 2007; De Dreu et al., 2010; Insel, 2010; Wittmann et al., 2016, 2021). However, the augmentation of the SOM effect induced by OTint does not necessarily lead to the enhancement of prosociality, and may even be directed against others, such as out-groups, particularly under socially competitive contexts (De Dreu et al., 2010; Marsh, 2019). Hence, human prosociality through SOM should be crucially dependent on social contexts and personal attributes of the participants (Bartz et al., 2011). In the current study, although OTint reliably induced SOM in all three experiments with independent populations of participants, the consequential prosocial effects (the enhancement of internal mentalizing abilities) were unstable. This mismatch between SOM and human prosociality may help to resolve the controversies of OTint modulation effects on social cognition and behaviors observed in the literature, both in healthy and patient populations (Buckner and Carroll, 2007; Nave et al., 2015; Alvares et al., 2017; Keech et al., 2018; Declerck et al., 2020; Sikich et al., 2021). The current findings call for caution in using the apparent behavioral measures to assess the OTint effects and the OTint therapeutic potentials in ameliorating social functions in psychiatric disorders (Insel, 2010; Meyer-Lindenberg et al., 2011).
Footnotes
This research was funded by the Key Program for International S&T Cooperation Projects of China (Grant MOST, 2016YFE0129100; to X.W.), the Fundamental Research Funds for the Central Universities (Grant 2017EYT33; to X.W.), STI2030-Major Project (Grant No. 2021ZD0204200; to N.L.), the National Natural Science Foundation of China (Grant No. 32071094; to N.L.), and the Chinese Academy of Sciences (Grant QYZDB-SSW-SMC033; to N.L.). We thank Xuejing Sun, Shaohan Jiang, and Liang Xian for technical help.
The authors declare no competing financial interests.
- Correspondence should be addressed to Ning Liu at liuning{at}ibp.ac.cn or Xiaohong Wan at xhwan{at}bnu.edu.cn
