Trait psychoticism is associated with greater intersubject neural dissimilarity during naturalistic social movie viewing

Abstract

Psychoticism refers to maladaptive traits associated with eccentricity, perceptual dysregulation, and unusual beliefs. This study explored how individual differences in psychoticism map into neural responses during naturalistic movie-watching. By utilizing social movies with differing demands of inference, we examined how relevant correspondence of individual differences emerges as a function of movie features. A total of 157 individuals were assessed with traits of psychoticism, and intersubject correlation of neural response was assessed with functional near-infrared spectroscopy (fNIRS). Three movie-watching conditions (low-social, high-social A/B) were compared in the examination of brain-trait correspondence. The result showed that the intersubject distance of psychoticism scores corresponds to lower neural synchrony (nearest neighbor pattern), while higher scores of psychoticism correspond to decreased synchrony between individuals (Anna Karenina pattern). Moreover, such patterns were pronounced in the movie condition that required more complex mentalizing and social inference. Also, individuals with higher psychoticism traits tended to correlate with more frequent use of excessive Theory of Mind (ToM) inference during movie-watching which further supported the basis of neural idiosyncrasy. Overall findings imply that specific task demands can be more effective in provoking individual differences of psychosis-proneness. Also, individuals with psychoticism traits may undergo maladaptive interpersonal functioning possibly due to idiographic neural response characteristics and hypermentalizing tendencies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-33646-8.

Introduction

Psychoticism represents a maladaptive personality domain characterized by eccentricity, perceptual dysregulation, and unusual beliefs. Within dimensional models of psychopathology, it aligns with schizophrenia-spectrum features spanning schizotypal and paranoid personality traits^1,2. Even at subclinical levels, these tendencies manifest as cognitive-perceptual disturbances, aberrant salience, and social detachment observable in the general population^3,4. Personality research increasingly views such traits not merely as taxonomic categories but as individual differences in interpretive and response schemas mediated by specific neural features^5–7. In this framework, psychoticism can be conceived as a personality dimension that modulates how social and perceptual information is encoded and integrated at the neural level.

While numerous neuroimaging studies have examined schizophrenia-spectrum traits, most have relied on resting-state or structural features, yielding inconsistent findings on connectivity and functional organization^8–11. Such paradigms may be suboptimal for identifying trait-dependent dynamics because they lack ecologically valid stimuli that elicit trait-relevant processes¹². To understand personality-related neural differences, one must examine the interaction between traits and task contexts—how particular cognitive or emotional demands accentuate trait expression. Recent approaches in “individual differences connectomics” propose that brain states can be manipulated to magnify between-person variability^13,14. Task-based functional connectivity, especially under naturalistic conditions such as movie watching, has shown stronger predictive power for behavioral and cognitive traits than resting-state data^15,16.

In parallel, inter-subject correlation (ISC) analysis has been proposed as a more effective method for interpreting the relationship between brain activity and social traits¹⁷. ISC does not simply map mean activation but examines shared temporal fluctuations, thereby preserving the ecological properties of social stimuli. Another advantage of the ISC approach is its ability to explore the nonlinear nature of individual differences through representational similarity analysis. In other words, individual differences in maladaptive traits may manifest not only as direct neural correlates corresponding to trait levels but also as deviations from the normative distribution in multiple directions^18,19. That is, individuals with stronger maladaptive traits may exhibit neural characteristics not as a single feature but as idiographic deviations, marked by increased inter-subject distance from the normative population. This pattern can be tested using two models: the Anna Karenina (AK) model, where “all happy families are alike; each unhappy family is unhappy in its own way,” and the Nearest Neighbor (NN) model, where increased distance between individuals corresponds to more dissimilar neural characteristics. It is notable that the ISC approach can differ from previous methods. Specifically, traditional predictive methods are limited to exploring linear correspondence patterns (i.e., higher or lower trait scores correspond to higher or lower neural features), which restricts them to identifying only the NN pattern of correspondence.

A study that effectively explored psychosis-proneness in a normative population analyzed whether high or low levels of trait paranoia—a characteristic related to impaired reality testing—were associated with increases or decreases in inter-subject correlation in response to naturalistic stimuli²⁰. However, the findings were somewhat inconsistent with those of clinical populations of schizophrenia, which would exhibit significant impairments. According to the normative model proposed in earlier studies, stronger psychotic features should lead to more idiographic neural responses^21,22, yet individuals with higher trait paranoia in the study of Finn et al. (2018) exhibited more similar neural responses in some brain regions. Although such results remain somewhat contradictory, the subjective phenomenology of psychotic symptoms, particularly the idiographic quality of bizarre mentation, suggests the importance of clarifying, through the present study, whether schizophrenia-spectrum traits are associated with more deviant neural responses. Given the scarcity of research investigating neural ISC correspondences in schizophrenia spectrum and psychoticism traits, definitive conclusions remain premature.

Another gap concerns the task content itself. Research employing movie-watching tasks has shown that the predictive power of neural information for behavior varies depending on the naturalistic nature of the film clips^23,24. Moreover, the stimuli used in ISC studies on schizophrenia often included silent films or public interest advertisements, which may not be truly representative of real-world contexts relevant to personality functioning, raising questions about their “naturalistic” quality^21,22. These stimuli were also typically low in ambiguity, limiting the extent to which mentalizing processes or abstract reasoning could be exerted. Given that schizophrenia spectrum disorders are associated with hypermentalizing tendencies in response to ambiguous social cues, stimuli inducing these cognitive processes could more effectively reveal the social-cognitive traits underlying schizophrenia-proneness^25,26.

For this purpose, the present study employed functional near-infrared spectroscopy (fNIRS) during movie watching. Although fNIRS has limited spatial depth, it enables ecologically natural recording in the prefrontal cortex—key to the default mode network (DMN) and social-cognitive processing^27,28. This technique allows participants to watch socially rich films in a comfortable, realistic posture, thereby preserving the authenticity of mentalizing engagement.

This study aimed to examine the relationship between individual differences in psychoticism traits and neural synchrony patterns by testing two models. The nearest-neighbor (NN) model predicted that greater inter-individual distance in psychoticism scores would correspond to greater neural dissimilarity during movie watching. In contrast, the Anna Karenina (AK) model predicted that both higher and lower extremes of psychoticism would be associated with neural dissimilarity. In this way, the study sought to identify neural markers of psychosis risk by considering both linear and non-linear embedded patterns.

We first hypothesized that, similar to other phenotypic individual differences, psychoticism would follow a nearest-neighbor pattern such that larger distances in trait scores would correspond to greater neural distances. Second, we hypothesized that higher psychoticism would be specifically associated with greater neural dissimilarity—reflecting a positive correspondence with AK pattern. Finally, we further expected that more ambiguous, socially demanding stimuli requiring interpretive processing would elicit greater neural divergence, and that this effect would be most pronounced among individuals with higher psychoticism traits.

Method

Participants

Participants were recruited via notices on campus bulletin boards and online platforms, with a focus on adults aged 19 and older. Following recruitment, participants completed the Personality Assessment Inventory (PAI) online before taking part in an in-person experiment. The PAI was administered online via the publisher-hosted, license-protected web platform operated by Insight (the authorized Korean distributor of the PAI). The platform provides encrypted, password-protected access to test materials and automated scoring reports, and raw item responses are stored on the vendor’s secure server rather than on local laboratory computers. Research staff only accessed de-identified scores for analysis. Participants accessed the test screen through individualized invitation links. Each link permitted a single submission, and item responses were required to be completed before submission.

Any participants whose data were incomplete, whether from the survey or the fNIRS experiment, were excluded from the analysis. Additionally, participants with Inconsistency scale (ICN) of 64 T or higher were prompted to retake the survey, receiving feedback on their extreme response inconsistency. Ultimately, three participants with invalid ICN scores were excluded from the study. After completing the online survey, the fNIRS experiment was administered to 177 individuals. Of these, data acquisition was incomplete for nine participants due to device failure, and 11 participants had missing data caused by Bluetooth wireless transmission errors. Ultimately, the remaining 157 participants were analyzed.

Ages ranged from 19 to 36 years (mean = 22.78, SD = 2.96), with 105 female participants (67%). For the occupational status, 14 participants were employed and four were unemployed; 11 were graduate students, and the remainder were undergraduates. All of the participants were South Korean without racial and ethnic heterogeneity. Among them, 44 (28%) had a history of psychiatric or psychological treatment, and 11 (7%) were receiving ongoing treatment or medication. The study protocols were approved by the Pusan National University Institutional Review Board (PNU IRB/2023_137_HR) and conducted in accordance with the Declaration of Helsinki. All participants gave their written agreement to use their information in the study and the research. All participants signed the informed consent.

The participant pool largely overlapped with that of our previous study within the same project on personality disorders²⁹. The present dataset was acquired in a separate run within the same recruitment pool; device failures and wireless transmission losses caused fNIRS data attrition, not differences in recruitment, yielding minor variation in the final analyzable sample. Although the current study shares themes related to maladaptive traits, the present work was conducted as a separate purpose that employs a different analytic strategy—intersubject synchrony—to relate neural characteristics to ways of interpersonal relating, and a distinct hypothesis that psychotic features are associated with reduced synchrony between individuals.

As conventional power analysis was not applicable to the current analytic approach (i.e., permutation-based matrix similarity test), we referred to studies with identical analytic approaches, whose sample sizes ranged from 22 to 93^19,20.

Social movie stimuli

The movie-watching task consisted of three clips, including one Low-Social (LS) and two High-Social (HS) conditions (A and B). To select the HS movies, we first screened publicly available videos that contained subtle yet intense social intentions and dynamics. After identifying suitable HS stimuli, we then searched for a contrasting LS movie that matched the HS clips in stimulus characteristics (i.e., number of characters, camera framing, speech density, and speaker-turn intervals). Detailed selection procedures and considerations are described in the Supplementary Method. Ultimately, three clips were chosen that differed in the level of social inferential demand while remaining comparable in duration and stimulus quality (Supplementary Method). Prior to the study, we obtained written permission from the respective content owners (production company and public institution) to use for non-commercial academic research. All three movie clips were based on pre-existing audiovisual works that had already been publicly released for general viewing, and informed consent was not required by the institutional review board (Supplementary Method).

The LS Movie was extracted from the National Election Commission as a public service content and featured four people discussing the political participation of young adults. The clip consisted of cuts where two or more characters interacted through dialogue; however, in contrast to the following HS Movies, the content consisted of minimal emotional interaction and socioemotional ambiguity. Following this, the HS Movies (A and B) were presented. These were selected from short dramas dealing with friendship and conflict among young adults. Contrary to the LS Movie, the dialogues contained strong emotional interactions while also leaving room for interpreting the character’s intentions (Supplementary Method). HS Movie A is a story with a female protagonist with themes of disdain and subtle power struggles (Synopsis: “Y and S are university friends, but S feels uncomfortable with Y’s attempts to bond by frequently asking to meet or hang out. S grows closer to another friend, C, which makes Y jealous and overly dependent on S for attention and help. Despite Y’s apology for her behavior, the story ends without her recognizing how her actions affected S.”). HS Movie B features a male protagonist and depicts themes of excessive dependency in close friendships (Synopsis: “J is a composer who works at K’s company but continues pursuing his passion for music, which K mocks as impractical and belittles in front of others. K frequently criticizes J’s music, personality, and appearance, pushing J to his breaking point after enduring repeated insults. J apologizes to K the next day, but K ignores the message, leaving their conflict unresolved.”).

To minimize differences in clip lengths, key scenes from the HS Movies, particularly those that were critical to narrative flow, emotion-inducing, or prone to interpretive differences, were selected based on qualitative evaluations from the researcher and five raters. Scenes essential for conveying the main narrative were retained, while interim summarizing narrations and indirect advertisements (i.e., product placement, sponsor inserts) were excluded. The excluded advertisement segments did not affect the overall narrative; they consisted of approximately 20 s of explanation on a specific product (i.e., bed), accompanied by a character’s verbal comment on the product. In terms of narration, HS movie A included voice-overs from an external narrator who explicitly summarized the development of events and highlighted the moral implications of the story. HS movie B contained segments where the inner speech of the characters was overtly verbalized through narration. These narrations were judged to reduce the scope for inference regarding the flow of the narrative and the characters’ mental states, thereby conflicting with the objectives of the present study, and were removed. Additional edits involved the exclusion of segments consisting of simple repetitions, casual short conversations unrelated to the main theme, or other non-essential cuts. These adjustments were made to equalize the total duration across movie conditions. Importantly, all narrative contexts remain intact, and both the unedited full-length versions of the movies and detailed information on the included sub-episodes can be accessed through online links (Supplementary Method). The final clip was confirmed to ensure that the central narrative remained plausible (Supplementary Method). The clip duration of the LS Movie was 3 min, while the two HS Movies were 5 min each. All clips were provided with subtitles. The stimuli were edited with approval from the original creators.

After watching each clip, questionnaires were administered regarding the content of the clips. For the LS Movie, a single item (“Choose the most appropriate topic for the conversation the characters are having”) confirmed whether participants paid appropriate attention to the clip (Supplementary Method). The result showed that all participants responded correctly. For HS Movies, participants were asked about the thoughts, emotions, and intentions of the characters with eight items after each HS Movie clip (four items for A and B, respectively). Based on reference to the Movie Assessment for Social Cognition (MASC), the items were constructed to reflect the levels of inference and theory of mind (ToM) toward the main characters from none to excessive³⁰. We aimed to assess hypermentalizing tendency by constructing response options that reflected excessive use of inference³¹. Participants were given five response options, which were categorized as correct ToM, no ToM, low ToM, excessive emotional ToM, and excessive cognitive ToM (Supplementary Method). The two response options of excessive ToM featured strong inferences of characters’ intentions that cannot be explicitly drawn from the clips (e.g., Excessive-emotional: “Sohyun wants to end her relationship with Yeonhe”; Excessive-cognitive: “The library is the most optimal place for preparing for the presentation and printing”). The low ToM option is characterized by an overall understanding of context and flow, but with partial and superficial inferences (e.g., “Sohyun needs to print something at the library”). The no ToM option was made up of either a literal or a lack of inference on the narrative, which was not based on consideration of any mental state of the character, typically accompanied by judgment with misinformation (“Sohyun has an appointment with another friend at the library”). Three blind raters, independent from the current research, who were allowed to scrutinize the movie, all judged the correct option to be the most optimal and accurate answer. A separate participant pilot was not conducted; instead, stimuli underwent expert qualitative screening (researcher and five independent raters) to equalize narrative length/ambiguity. Overall, the average correct rate was 64% showing that participants showed a certain level of ToM in the HS movie. The rate of no ToM (3%) and low ToM (5.4%) was relatively low, while the response rate of excessive ToM showed 10.1% in emotional ToM and 16.7% in cognitive ToM (Supplementary Table S1). The response proportions varied by item according to difficulty and category, and the accuracy rates and response proportions for each were reported (Supplementary Method).

Trait psychoticism

Trait psychoticism was measured and derived based on the Personality Assessment Inventory (PAI)³². PAI is composed of 344 items and encompasses a variety of personality scales, including those that show convergent evidence for psychotic features and the presence of psychotic disorders^33,34. Recent Korean re-standardization of the test yielded good internal consistency, with schizophrenia-related scales ranging from 0.77 to 0.79³⁵.

Within the DSM-5 alternative model for personality disorders, psychoticism is one of the five pathological personality trait domains. It reflects extreme personality characteristics, such as unusual beliefs and experiences, eccentricity, and cognitive-perceptual dysregulation, that are typically observed in schizotypal personality disorder or severe personality dysfunction with psychotic features⁴. In most research, this domain is operationalized with the Personality Inventory for DSM-5 (PID-5)³⁶. Several studies have shown that combinations of existing PAI scales can be used to approximate the five maladaptive trait domains of the PID-5, including psychoticism, rather than developing entirely new scales^37,38. This trait-mapping approach has also been extended and validated in the South Korean version of the PAI³⁹. In particular, the Schizophrenia (SCZ) and Paranoia (PAR) clinical scales and their subscales have demonstrated convergent validity with other assessment indices of psychotic disorders and interview-based diagnoses of paranoid personality^40–42. Accordingly, the present study relied on these existing, validated PAI scales as indicators of trait psychoticism, instead of constructing a novel stand-alone measure.

In the present study, we derived a psychoticism index from the SCZ and PAR subscales following the trait-mapping results reported by Ruiz et al. (2018) and Kim et al. (2019)^39,43. In the Kim et al. report, SCZ-S (Social Detachment) and PAR-R (resentment) showed only small correlations with the psychoticism trait (r = 0.187–0.287), and SCZ-S only showed a strong correlation with the detachment trait (r = 0.807). The PAR-H (hypervigilance) subscale converged more strongly with negative affectivity trait, with weaker correlation with the psychoticism trait (Negative affectivity correlation: r = 0.445; Psychoticism correlation: r = 0.401). By contrast, the remaining three subscales, including SCZ-T (Thought Disorder), SCZ-P (Psychotic Experiences), and PAR-P (Persecution), showed a specific and strong correlation with the psychoticism trait (r = 0.508–0.725). On this basis, SCZ-T, SCZ-P, and PAR-P was selected as the three indicators of psychoticism. The decision is consistent with the cautions regarding the interpretation of SCZ-S noted in the PAI interpretive guide, which suggests the SCZ-S can easily elevate due to non-psychotic features. The rationale for the composite formulation was not based on a separate factor structure; rather, it was motivated to formulate a more specific and narrowly defined construct of schizophrenia spectrum syndrome. Each subscale is calculated with six items, and a prior South Korean validation study has reported internal consistency coefficients at levels comparable to the original PAI (SCZ-P: 0.64, SCZ-T: 0.76, PAR-P: 0.82)⁴⁴. A flow chart summarizing this derivation procedure is presented in Supplementary Figure S1.

To ensure that this composite behaved as a coherent index in the current sample, we evaluated its internal consistency and dimensionality. The three subscales showed acceptable internal consistency (ω = 0.70) when combined. Parallel analysis also confirmed that a single factor was the optimal solution and can be used as unidimensional construct. Thus, we interpret the composite as a research index capturing the shared variance of psychosis-proneness across these subscales, while relying on the extensive prior validation of the underlying PAI scales and their trait-domain mapping.

The final trait psychoticism score was formed by averaging the norm-standardized T-scores of these three subscales. In calculating inter-individual Euclidean distance, a three-dimensional vector [SCZ-T, SCZ-P, PAR-P] was directly calculated to obtain the distance metric. The subjects were also comprised of some proportion of cases with increased scales (T-score above 60) in SCZ-T (26), SCZ-P (n = 17), and PAR-P (n = 9). Additionally, the composite psychoticism trait did not show significant differences between genders (t = 1.26, p = 0.21).

Administration of online assessment and on-site experiment

Online PAI delivery and scoring were handled by the vendor platform; research staff verified completion and ICN flags. On-site sessions (consent, seating, NIRSIT placement, instruction script, debrief) were administered by trained graduate research assistants under supervision of the principal investigator. ToM questionnaires were computer-administered immediately after each clip; item keys were pre-specified, with three independent raters involved only in key confirmation (not in participant scoring).

fNIRS device

To measure neural responses, a wireless NIRS system (NIRSIT; OBELAB Inc., Seoul, Republic of Korea) was employed. This device features 24 laser sources (780/850 nm) and 32 detectors, supporting inter-optode distances up to 4.5 cm. Based on spatial resolution and differential path length, 48 channels were utilized with a source-detector spacing of 3.0 cm. The sampling rate of the device was 8.138 Hz. The equipment was secured on the head using a tightened strap. To ensure acquisition consistency across homologous brain regions, the device’s front calibration point was positioned at the midpoint between the eyes. The device covered prefrontal regions, including the ventrolateral, dorsolateral, frontopolar, and orbital cortex areas⁴⁵. All of the fNIRS acquisition procedures were identical to the previous work²⁹.

fNIRS preprocessing and ISC analysis

Negative raw intensity samples attributable to detector saturation were corrected via temporal interpolation using the nearest non-negative samples. No explicit onset-trimming window was applied at the beginning of each clip. Signals measured from each participant were then transformed into oxygenated hemoglobin (HbO) concentration using the Modified Beer-Lambert Law (MBLL)⁴⁶. Then signals from channels that met standard criteria for irrelevant signals (e.g., low signal-to-noise ratio below 15 dB, extreme correlation between HbO and HbR signals of r < −0.9, coefficient of variation below 7.5%) were identified as invalid. The criterion of extreme HbO–HbR anticorrelation is likely to reflect scalp and systemic physiological noise rather than neural activity, and a low signal-to-noise ratio (SNR) with limited variability indicates poor signal quality. Values for channels identified as invalid were then padded by replacing them with the mean of channels belonging to the same ipsilateral prefrontal Brodmann area (Supplementary Figure S2). For the calculation of SNR (dB), it was computed with the mean across the two wavelengths (780 nm and 850 nm) over a 10–15 s baseline as 20·log10(µ/σ). The quality metric of the SNR was quantified with previous recommendations officially implemented in the NIRSIT Analysis Tool’s quality control pipeline⁴⁷. To reduce physiological and environmental noise, a band-pass filter with a range of 0.005 Hz to 0.1 Hz was applied⁴⁸.

Since specific channels were prone to noisy input, channels identified as invalid (and padded) in more than 20% of the participants (> 32 cases) were completely discarded from the analyses, leaving 36 channels of information. The decision was based on scrutinizing the distribution of the channel invalidity rate (Supplementary Figure S3). It was also confirmed that the total number of channel rejection (i.e., padded channels) of an individual showed no association with trait psychoticism (r = −0.02, p = 0.792). All of the procedures were identical to the previous report²⁹ except for the lowered padding criteria (median intensity from 20 to 15) and subsequent exclusion of channels with excessively high padding rates. The analysis decision differed since the current study included channel-level analyses.

After preprocessing, the computed HbO signals were used to calculate neural synchrony between all pairwise subjects (Fig. 1A). Temporal correlation (Pearson’s r) was calculated to assess the strength of neural ISC between every pair of subjects in a homologous channel. To assess whether neural intersubject correlation (ISC) differed across conditions, we performed a nonparametric permutation test at the subject level. For each movie condition, we computed 157 subject-level ISC values (the average ISC for each subject). We then generated a null distribution of the mean difference between two condition pairs (LS vs. HS-A and LS vs. HS-B) by randomly shuffling the condition labels and recomputing the between-condition mean difference at each permutation of 10,000 times. The observed mean difference computed with the true labels was compared against this null distribution; the p-values were obtained from the rank of the observed statistic within the 10,000 permuted values plus the observed value, calculated as (rank/10,001). Across the 36 channels, we identified those showing significant condition differences after controlling the false discovery rate (FDR-corrected q < 0.05). Condition comparisons targeted two prespecified contrasts: LS vs. HS-A and LS vs. HS-B.

Fig. 1.

Analysis scheme. (A) Neural synchrony assessed with Intersubject correlation (ISC) of fNIRS. ISC during three movie-watching conditions represents the pairwise neural similarity of subjects. The images are screenshots of the movie stimuli, which are accessible with full movie links (Supplementary Method). (B) Intersubject psychoticism trait pattern. The two intersubject patterns were represented by a matrix of pairwise score distances (Nearest Neighbor, NN) or pairwise mean scores (Anna Karenina, AK). (C) Intersubject Representational Similarity Analysis (IS-RSA). Similarity analysis between two intersubject matrices assessed whether the neural dissimilarity (1-ISC) corresponds to larger trait difference (NN pattern), higher trait scores (AK + pattern), or lower trait scores (AK- pattern).

Statistical analysis

Intersubject pattern of trait psychoticism

The analytical framework employed the intersubject representational similarity analysis (IS-RSA) method to investigate the brain-behavior relationship (Finn et al., 2020). The pairwise intersubject patterns of trait psychoticism were framed as distance and average (Fig. 1B). For the distance pattern matrix, the Euclidean distances across three scale measurements (SCZ-T, SCZ-P, PAR-P) were first calculated between all pairs of subjects. In this trait distance analysis, we did not aggregate the three scores but instead defined a three-dimensional trait vector [SCZ-T_i​, SCZ-P_i​, PAR-P_i​] and computed its pairwise distance matrix. In the case of the average pattern, the intersubject matrix was defined by the mean of the two trait scores of a pair. The intersubject pattern of the distance matrix and average matrix, respectively, was employed in the subsequent analysis of intersubject representational similarity analysis (i.e., correspondence with NN and AK patterns). We tested whether pairwise distances in psychoticism traits predicted pairwise neural dissimilarity (1–ISC) across subjects (i.e., NN pattern), and whether higher scores of psychoticism traits predicted pairwise neural dissimilarity (i.e., AK pattern).

Additionally, the association between the accuracy of social inferences, as measured by the response rate of ToM items (i.e., no ToM, low ToM, excessive cognitive ToM, excessive emotional ToM), and psychoticism was examined using Spearman’s correlation, with the hypothesis that higher psychoticism levels would lead to inaccurate ToM performance. To estimate 95% confidence intervals for Spearman’s rank correlation coefficients, we employed a nonparametric bootstrap procedure. Specifically, we generated 5,000 bootstrap samples of equal size to the original dataset by sampling with replacement and computed the correlation for each resample.

Inter-Subject representational similarity analysis (IS-RSA)

To match the direction of the representational correspondence, Pearson’s correlation coefficient of neural ISC was inverted (1–ISC) to represent intersubject dissimilarity in movie-watching neural responses. IS-RSA examines matrix similarity between the intersubject trait pattern and neural ISC as two patterns of representational similarity: NN and AK (Fig. 1C). The NN model tests that higher trait distance is associated with higher neural dissimilarity (inverse of ISC) and vice versa. The NN model primarily assumes that individual difference between psychoticism scores also corresponds to neural dissimilarity. On the contrary, the AK model tests a pattern that the overall level of trait score is associated with neural dissimilarity. Specifically, the pattern of AK was examined in both directions, in that either high (AK+) or low (AK-) psychoticism trait corresponds to the larger neural dissimilarity, although the main directional hypothesis was set to positive association (AK+).

To examine the correspondence of brain-trait association, the similarity between the two intersubject matrices (i.e., psychoticism distance and 1-ISC) was tested with the Mantel test of correlation using the MATLAB package bramila_mantel⁴⁹. The Mantel test evaluates the statistical significance of whether two matrices meaningfully correlate with each other. The estimate of matrix similarity is first calculated based on the Spearman correlation coefficient. Then, to generate a null distribution of the correlation coefficient, we randomly permuted the subject labels of one matrix by applying the same random permutation to its rows and corresponding columns simultaneously (thus preserving symmetry and the dependence structure), recomputed Spearman ρ between the vectorized upper triangles at each iteration, and built a null distribution of ρ. Permutation testing was conducted within the conditions only. The p-values were obtained through a permutation test with 5,000 iterations. In addition to the primary uncorrected test (p < 0.01), the FDR-corrected p-values (q < 0.05) were also noted. The FDR correction was applied to the multiple-channel testing within each specified set of conditions and RSA patterns.

Results

In the ratings of HS movie watching, individual differences in the overall correct rate of ToM and trait psychoticism composite score showed a tendency of association (ρ(155) = −0.14, 95% CI [−0.32 − 0.01], p = 0.080). The psychoticism composite was negatively correlated with a response rate of no ToM (ρ(155) = −0.17, 95% CI [−0.31 − 0.07], p = 0.031) while positively correlated with a response rate of excessive emotional ToM (ρ(155) = 0.18, 95% CI [0.03 0.30], p = 0.028) and excessive cognitive ToM (ρ(155) = 0.16, 95% CI [0.04 0.34], p = 0.049). On the contrary, no association was observed with the rate of low ToM (ρ(155) = −0.03, 95% CI [−0.14 0.12], p = 0.715). The individual variabilities are visualized in Supplementary Figure S4.

The fNIRS time-series correlations assessed neural ISC between subjects. For the main condition effects of neural ISC, the overall differences were compared between the three clip conditions. Upon inspecting regional patterns, the results showed that LS Movie exhibited overall higher ISC across channels, indicating more synchronized responses during movie-watching (Fig. 2). When we shuffled the per-channel, per-condition mean ISC values for the 157 participants to generate a null distribution and performed a permutation test (FDR-corrected q < 0.05), we confirmed that higher ISC for the LS movie was observed, particularly in the medial and right ventrolateral regions (Supplementary Figures S5).

Fig. 2.

FNIRS regional definition and overall neural synchrony effect during movie watching. (A) Anatomical label of fNIRS channels. IFG: Inferior Frontal Gyrus, MFG: Middle Frontal Gyrus, SFG: Superior Frontal Gyrus, OFC: Orbitofrontal Cortex. Gray area indicates rejected channels after preprocessing. (B) Neural ISC. Average of all pairwise neural ISC of fNIRS channels during movie watching by conditions of Low Social (LS) and High Social (HS-A/B). The color bar indicates the strength of ISC synchrony calculated with Pearson’s r.

In the ISC-RSA result, we showed two distinct correspondences between neural ISC and psychoticism traits as NN and AK patterns. First, the result of the NN pattern showed that higher psychoticism distance was positively correlated with higher neural dissimilarity (i.e., 1-ISC, lower synchrony) specifically in the condition of HS Movie B (Fig. 3A). Overall, the coefficients were in a positive direction, showing that the extent of difference in psychoticism scores corresponds with the dissimilarity of neural response during movie-watching. The areas with NN pattern distributed across the medial, orbital, and superior parts of PFC showed significant representation similarity (permutation p < 0.01, Supplementary Table S2). Also, all of the regions survived in the FDR correction (qs < 0.05).

Fig. 3.

Intersubject-Representation Similarity Analysis (IS-RSA) result. (A) NN pattern (Nearest neighbor: intersubject distance of psychoticism score associated with neural dissimilarity). (B) AK + pattern (Anna Karenina positive: Higher psychoticism associated with neural dissimilarity). (C) AK- pattern (Anna Karenina negative: Lower psychoticism associated with neural dissimilarity). Asterisks indicate statistical testing (p < 0.01). The horizontal axis represents the right (R) or left (L) hemisphere, while the vertical axis represents the dorsal or ventral part of the PFC regions. The color bar indicates Spearman’s correlation ρ between the vectorized upper triangles of NN/AK pattern matrix and neural dissimilarity matrix (1 – ISC).

The result of the AK pattern was examined in AK+ (higher psychoticism trait is associated with higher neural dissimilarity) and AK- (higher psychoticism trait is associated with lower neural dissimilarity). The result showed that only the AK + pattern was pronounced in HS Movie B. Specifically, the effect was observed in a region of the orbital part of PFC, though the effect in one channel did not survive in FDR correction (permutation p < 0.01, FDR qs > 0.05) (Fig. 3B, Supplementary Table S3). In contrast, the AK- pattern was not found in any of the channels (Fig. 3C, Supplementary Table S4).

Discussion

This study explored how individual differences in psychoticism traits correspond to variations in dynamic neural responses while watching diverse social movie stimuli. The key finding was that greater inter-individual distances in trait scores were associated with weaker neural synchrony between individuals. This pattern of idiosyncrasy became more pronounced as psychoticism trait scores increased. The research demonstrated that individual differences in psychoticism traits during social movie watching manifest in two distinct neural patterns, which can be mapped to the Nearest Neighbor (NN) and Anna Karenina (AK) principles.

The first major observation was the distinct pattern of neural synchrony across three movie conditions. Two results were noted here: (1) neural synchrony was significantly stronger during the LS Movie compared to the HS Movie. While the PFC generally exhibits lower intersubject correlation ISC compared to sensory cortices, the demands imposed by naturalistic stimuli still elicited a certain level of synchronized responses. Previous studies suggest that neural synchrony increases when watching stimuli with a clear narrative flow, whereas synchrony decreases when watching stimuli with unclear narratives⁵⁰. Additionally, when viewers share a common understanding of the movie content, their neural responses become more similar⁵¹. In this study, the LS Movie likely evoked more homogeneous mental processes, while the HS Movie allowed for greater idiographic interpretation during narrative construction, potentially leading to reduced neural synchrony.

For the examination of brain-trait correspondence, the primary finding was the identification of the NN pattern. Greater neural response dissimilarity during HS Movie B was linked to larger differences in psychoticism trait scores, indicating that trait differences were associated with the dynamic neural fluctuations elicited by this particular movie. Prior research has shown that movie-watching ISC reflects individual differences in narrative interpretation and emotional responses during specific segments of the film^51–53. In this study, the observed intersubject neural ISC corresponding to the NN pattern (trait distance) appears to reflect variations in the extent of narrative interpretation, personal relevance, and emotional experiences linked to psychoticism traits. It is also noteworthy that the NN pattern was not observed in either LS Movie or HS Movie A, suggesting that these stimuli did not elicit the neural dynamics relevant to highlighting individual differences of psychoticism. This underscores the notion that the relationship between self-reported traits and neural responses in movie-watching paradigms can vary substantially depending on the specific content and interpretive demands of the stimuli.

Additional examination of brain-trait correspondence was conducted as an AK pattern. In this study, the AK + pattern, where higher psychoticism trait scores were associated with greater neural dissimilarity, was identified in HS Movie B. This is consistent with the original AK principle (i.e., “each unhappy family is unhappy in its own way”) and aligns with findings from previous studies on clinical disorders such as schizophrenia^21,22,54. Unlike the NN pattern, the AK + pattern represents non-linear brain-trait relationships, where individuals with lower trait scores show relatively homogeneous neural ISC, while those with higher scores display more multidirectional and deviant neural patterns. This is similar to findings in neuroimaging studies of schizophrenia, where heterogeneity in neural features rather than a singular predictive feature is associated with psychopathology⁵⁵. Our study demonstrates that even for a milder range of differences on the schizophrenia spectrum, namely the psychoticism trait, the normative model approach that characterizes mental disorders as atypical deviations remains a valid explanation. Although the AK + effect was not as extensive as the NN effect, the result was consistent in HS Movie B, indicating that idiographic neural responses were more pronounced for individuals with high psychoticism scores when watching this particular clip. Conversely, the AK- pattern, where higher trait scores would predict greater neural synchrony, was not observed. If higher psychoticism traits had led to stronger or more homogeneous neural responses to specific narrative cues, the AK- pattern might have been evident. However, this study suggests that high psychoticism traits did not induce homogeneous neural responses.

Interestingly, our study observed a different directionality in the AK pattern compared to prior research. Specifically, a previous study²⁰ found that higher trait paranoia was associated with stronger neural synchrony while listening to an auditory social narrative, which corresponds to AK- pattern. However, in our study, the AK + pattern was more prominent. Several factors may explain this discrepancy, including the relatively narrow focus on the PFC, the broader psychoticism construct, the larger participant sample, and the nature of the naturalistic stimuli, which may have modulated individual differences differently. Additionally, the two directions of the AK pattern (AK + and AK-) relate to how one conceptualizes “unhappy family” traits, particularly in terms of whether higher scores on certain traits indicate suboptimal brain states. For example, in individual differences in working memory performance, lower scores may correspond to suboptimal brain states of higher brain dissimilarity, leading to AK + pattern. On the other hand, for traits like psychoticism, higher scores might indicate more suboptimal deviance from the normative population and exhibit heterogeneous neural synchrony¹⁸. Our study provides partial evidence that neural correlates of psychoticism can be more characterized by an AK + pattern.

The AK + pattern observed in this study may offer a hypothetical explanation for the interpersonal maladaptiveness seen in individuals with high psychoticism traits. We found that higher psychoticism was associated with poorer theory of mind (ToM) performance, particularly due to a tendency toward hypermentalizing—excessive filling-in based on nonexistent evidence—rather than inaccuracies in concrete or superficial inferences. Previous research has also linked psychoticism traits to increased activation of representations that are contextually inappropriate or self-referential^26,56. This idiographic, self-referential reasoning may ultimately lead to inaccurate ToM attributions. The AK + pattern observed in our study resonates with this idea, suggesting that individuals with high psychoticism may be more prone to self-immersed inferences, resulting in idiosyncratic neural responses rather than showing coinciding brain features across individuals. Therefore, the dysfunction in individuals with high psychoticism may stem not from a lack of mentalizing process but rather from the lack of appropriateness and accuracy of their mentalizing processes.

In the search for valid neural mechanisms of the schizophrenia-spectrum trait, our findings may also offer an alternative explanation for the inconsistent identification of specific brain regions associated with psychoticism-related traits¹¹. Rather than confining targets to specific brain regions, traits like schizotypy may involve functional characteristics of idiographic derailment in response to contextualized stimuli, such as social movies, which task-free brain characteristics may not fully capture.

Another notable finding was the distinct brain-trait correspondence as a function of task demand (LS vs. HS). According to our findings, merely watching social interactions (i.e., LS condition) did not significantly elicit individual differences in neural responses. Instead, it was the clips that required complex social reasoning and strong emotionality that revealed the neural correlates of psychoticism traits. By identifying specific movie stimuli that highlight relevant neural differences, our study proposes that the advantages of the naturalistic movie-watching paradigm can be augmented with relevant task demands²⁴. This also suggests that the “cardiac stress test” analogy for neuroimaging prediction¹³ extends from cognitive function (e.g., working memory) to individual differences in personality functioning.

Despite the limited anatomical coverage, our findings also highlight distinctive regional patterns of ISC during movie watching, along with channel-level regional significance in IS-RSA. Notably, the medial portion of the PFC exhibited stronger intersubject synchrony compared to the overall effect observed in the broader prefrontal cortex. This is a contrasting pattern from other conventional neurocognitive tasks that induce the most lateral part of the PFC responses. Previous studies have reported high inter-subject correlation (ISC) in the superior and medial frontal cortex when interpreting stimuli requiring social narrative processing^20,54. During the viewing of video stimuli, there is a tendency for increased engagement of the default mode network (DMN), particularly in the midline cortex, which is associated with its role in social functioning⁵⁷. This aligns with findings from other studies demonstrating that the primary ISC effects within the DMN regions extend beyond the direct sensory modality of the stimuli⁵⁸.

There were some unexpected results that warrant discussion. Individual differences were not augmented in both high-demand HS conditions but were observed only in HS Movie B. If psychoticism traits had modulated neural responses differently across LS and HS conditions, the lack of effects in HS Movie A remains unexplained. This introduces a key limitation in interpreting our findings in that the emotional characteristics of the stimuli remain unclear. Although both HS Movies featured conflict and relationship difficulties, HS Movie B contained more explicit and offensive behaviors which may potentially heighten the perceiver’s vigilance and arousal. Moreover, HS Movie B featured a more destructive progression of the conflict, leading to a breakdown of the relationship, which distinguished it from Movie A. Previous evidence also suggests that the emotional features of valence and arousal played a significant role^52,53. In addition, we cannot rule out potential effects arising from the presentation procedure and stimulus order. In this study, the order of clip presentation was not randomized, and HS Movie B may have included carryover effects from the preceding clips, thereby possibly imposing relatively greater emotional load. Future studies should incorporate physiological responses or affective experiences to clarify the role of these factors.

Several other limitations should also be addressed. First, the durations of the movie stimuli were not perfectly matched. Although movie-watching reaches a maximum level of informativeness within three minutes of acquisition⁵⁹, further confirmation is needed to ensure that differences are not driven by the amount of information presented. Second, this study was limited to neural ISC patterns in the PFC and therefore cannot be assumed that the observed ISC patterns represent the entire brain pattern. Previous research has shown that the AK pattern of ISC varies across brain regions²⁰, and whether similar results would be observed in regions involved in lower-level processing, such as the lateral temporal cortex, which is crucial for mentalizing and recognizing social interactions, remains an open question⁶⁰. Third, it is unclear whether our findings are specific to psychoticism. While the study focused on traits associated with syndromes of reality monitoring problems, the high psychoticism participants likely exhibited perceptual distortions and cognitive confusion that may also reflect traits seen in other conditions, such as borderline personality traits and bipolar disorder. Previous research has suggested that idiographic neural responses can also be observed in other disorders involving social deficits, indicating that dissimilar neural ISC may be a more transdiagnostic neural correlate⁶¹. Lastly, the participants of this study were predominantly recruited from a campus population, which may limit the generalizability of the findings.

This study reveals how psychoticism traits shape neural dynamics during naturalistic social movie viewing, highlighting distinct patterns aligned with the NN and AK principles. Higher psychoticism was associated with idiosyncratic neural responses, especially under conditions requiring complex social reasoning and emotional engagement. These findings emphasize the utility of naturalistic stimuli for studying personality traits and the importance of task demands and emotional context in shaping neural responses. While the study advances our understanding of psychoticism within the schizophrenia spectrum, future research should explore broader neural and psychological contexts to refine these insights further.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Min Kyoung Kim: Writing – original draft, Methodology, Formal analysis, Conceptualization. Da Won Jeong: Methodology, Data curation, Conceptualization. Do Yeon Yoo: Methodology, Data curation. Seyul Kwak: Writing – review & editing, Writing – original draft, Methodology, Funding acquisition, Formal analysis, Conceptualization.

Funding

This work was supported by the Humanities·Social-Science Research Promotion of Pusan National University (2025).

Data availability

The datasets and analysis codes are available at OSF storage ([https://osf.io/a49ct/](https:/osf.io/a49ct)).

Declarations

Competing interests

The authors declare no competing interests.