Heroes and villains: opposing narrative roles engage neural synchronization in the inferior frontal gyrus

Abstract

Neuroscientific studies have highlighted the role of the default mode network (DMN) in processing narrative information. Here, we examined whether the neural synchronization of the DMN tracked the appearances of protagonists and antagonists when viewing highly engaging, socially rich audiovisual narratives. Using inter-subject correlation analysis on two independent, publicly available movie-watching fMRI datasets, we computed whole-brain neural synchronization during the appearance of the protagonists and antagonists. Results showed that the inferior frontal gyrus (IFG) had higher ISC values during the appearance of the protagonists than the antagonists. Importantly, these findings were generalized in both datasets. We discuss the results in the context of information integration and emotional empathy, which are relevant to functions of the IFG. Our study presents generalizable evidence that the IFG show distinctive synchronization patterns due to differences in narrative roles.

Introduction

Audiovisual narratives dominate modern lives. As in real-life interactions, readers and audiences actively infer the intentions and actions of characters. People attempt to understand the constructed reality of a narrative and emotionally identify with the characters through experience-taking, a process in which the perceivers replace part of their identity with the mindset of the characters (Kaufman and Libby 2012). This experience is similar to empathy—the sharing of the feelings of another person (Decety and Jackson 2004). Indeed, there is a growing body of research that links narrative engagement to social cognition (Tamir et al. 2016, Mar 2018). Relatedly, exposure to fiction improves the capacity for empathy (Kidd and Castano 2013), social inference (Black and Barnes 2015), and even affect functional brain connectivity in both short term and long term (Berns et al. 2013). As such, narrative engagement shapes a person’s experience and influences how they navigate their social worlds.

Although narrative engagement is a widely studied topic, only a few neuroimaging studies examined the neural basis of narrative-driven character perception. One recent functional magnetic resonance imaging (fMRI) study found that the activity of the default mode network (DMN) distinguishes fictional characters based on their narrative roles (Ron et al. 2022). Another recent fMRI work by Ohad and Yeshurun (2023) observed a dissociation in neural synchronization patterns during the appearance of positively and negatively engaging characters. Both studies highlight the role of the DMN in distinguishing different narrative roles.

The importance and contributions of these studies notwithstanding, there are some limitations that could be improved upon. For instance, Ron and colleagues (Ron et al. 2022) used static images of the characters from different movies to elicit brain activity, precluding potential effects of ongoing perception of actions, dynamic facial expressions, and any in-narrative contexts. Using naturalistic stimuli such as movies and audiobooks improves ecological validity compared to traditional static visual stimuli (van Atteveldt et al. 2018, Nastase et al. 2019, 2020). It is critical to examine the complex nature of socioemotional processing with naturalistic stimuli by capturing how our brains function in the wild (Saarimäki 2021, Tikka et al. 2023). Ohad and Yeshurun (2023) adopted such an approach to investigate the neural substrates of character engagement, but was limited to a single audio-only narrative. In the present study, we included two independent audiovisual narratives (i.e. movies), seeking to improve the generalizability of our laboratory-based findings while maximizing the advantages of using naturalistic stimuli in fMRI experiments.

Here, we investigated the involvement of brain regions, particularly the DMN, as participants watched audiovisual narratives using two independent, publicly available movie-watching fMRI datasets. If people indeed gain experience through the lens of characters as a story unfolds (Kaufman and Libby 2012), there would be differences in how they process each narrative role depending on the relatability or the amount of information about the character. We reasoned that the characters with the most prominent and distinctive narrative roles—protagonists and antagonists—would offer a useful window into the neural basis of such processes. We adopted the inter-subject correlation (ISC) method to test the neural synchronization among individuals while they experienced the same time-locked stimuli (Hasson et al. 2004). The basic premise of the ISC analysis is that, as the subjects watch the same movie, the blood-oxygen-level-dependent (BOLD) signal in brain regions that putatively process the stimuli will systematically synchronize across subjects. Therefore, ISC values reflect how consistent the region is involved in processing the stimuli (Nastase et al. 2019, Nguyen et al. 2019). In the present study, we posit that there will be different neural synchronization patterns, indexed by ISC, based on the contrasting narrative roles—protagonists and antagonists. We hypothesized that, across two fMRI datasets, the DMN will show separable neural synchronization patterns during the appearance of protagonists and antagonists. Specifically, we expected the DMN to show greater neural synchronization patterns toward protagonists compared to antagonists, in line with previous work (Ohad and Yeshurun 2023) that showed similar effects for positively engaging characters in a narrative.

Materials and methods

Datasets

We utilized two open fMRI datasets that leveraged naturalistic stimuli with a narrative focus. The first dataset comprises fMRI scans of 16 subjects (demographic information was only available for the initial 22 participants of the original study; mean age = 20.8, ages 18–26, 12 males, 10 females) watching the initial 50 minutes of the 2010 BBC TV series Sherlock (Chen et al. 2018). The second dataset contains fMRI scans of 25 subjects (mean age = 27.5, ages 22–31, 12 males, 13 females) watching the second half of the 2014 film The Grand Budapest Hotel (Di Oleggio Castello et al. 2020). Both datasets are available to download at OpenNeuro (Sherlock: https://openneuro.org/datasets/ds001132; The Grand Budapest Hotel: https://openneuro.org/datasets/ds003017). We chose these two datasets as both Sherlock and The Grand Budapest Hotel contain socially rich contents with an engaging narrative structure that centers around its main characters. Our focus was on the main protagonist and antagonist in each story, which corresponds to Sherlock and Mycroft (Sherlock), and Gustave and Dmitri (The Grand Budapest Hotel).

Preprocessing

We aimed to keep the preprocessing steps as consistent as possible across the two datasets. For Sherlock, we used the preprocessed fMRI data that were published along with the rest of the Sherlock dataset (Chen et al. 2018). In brief, Chen et al. (2018) performed realignment to MNI152 space and spatial normalization using fMRIPrep (Esteban et al. 2019). Spatial smoothing was conducted using a 6-mm full-width at half-maximum kernel, and denoising was performed using a general linear model (GLM). Specifically, denoising involved removing global and multivariate spikes, average CSF activity, linear and quadratic trends, as well as 24 motion covariates. Among the initial 22 participants, six of them were discarded due to excessive head motion (n = 2 participants), short recall data (n = 2), falling asleep (n = 1), and missing data (n = 1).

Since The Grand Budapest Hotel dataset (Di Oleggio Castello et al. 2020) only provided raw data, we performed preprocessing using an equivalent pipeline as the Sherlock dataset via fMRIPrep. The preprocessing steps included realignment, spatial normalization, smoothing, and denoising as described above. Framewise displacement (FD) was calculated for each functional run, and two subjects were excluded due to excessive mean FD (>0.2 mm) (Power et al. 2014). The final sample size consisted of 39 (16 from Sherlock and 23 from The Grand Budapest Hotel datasets) participants. For both datasets, we used the Schaefer-Yeo atlas (Schaefer et al. 2018) to parcellate the fMRI data into 100 regions, from which mean BOLD timeseries were extracted.

Annotation and scene selection

Rather than using the entire BOLD timeseries, we focused on the inter-subject synchronization of brain activity that was time-locked to each character’s appearance (Fig. 1). In Sherlock and The Grand Budapest Hotel, the protagonists are Sherlock and Gustave, whereas the antagonists are Mycroft and Dmitri, respectively. It is worth noting that although Mycroft is not the primary antagonist throughout the entire episode, he is considered as such due to his actions of threatening another main character, John Watson, while self-identifying as an “enemy” of Sherlock. This intentionally misleading portrayal confuses the audience, as the true criminal is not revealed within the first 50 minutes of the movie. Thus, the narrative role of Mycroft, at least in the first 50 minutes of the first episode, is consistent with that of an antagonist.

Figure 1.

Timepoints during which the protagonists (Sherlock, Gustave) and antagonists (Mycroft, Dmitri) appeared on screen in their respective movies (first episode of Sherlock, second half of the film The Grand Budapest Hotel). Dotted lines indicate segments selected to match the density and quantity of the annotated TRs between the narrative roles per dataset.

We annotated the repetition times (TRs) corresponding to the appearance of each character’s face alone on the screen or when the character was alongside other characters, with the camera focused on the specific target character (e.g. camera following the protagonist/antagonist as they moved) and concatenated these TRs per character (Ohad and Yeshurun 2023). To ensure accuracy, we excluded scenes where the protagonists and antagonists interacted in the same scene. Considering the hemodynamic delay of 3–5 s (Hasson et al. 2004, Chang et al. 2021), we selected TRs that were 4.5 s (3 TRs) after the characters appeared on screen for Sherlock and 5 s (5 TRs) for The Grand Budapest Hotel. The annotated time length for each character is 690 s for Sherlock, 123 s for Mycroft, 89 s for Gustave, and 92 s for Dmitri. We note that since Gustave mostly appeared with other characters, especially a sidekick character named Zero Moustafa, Gustave had fewer annotated timepoints compared to Sherlock. On the other hand, the antagonists, Mycroft and Dmitri, had similar length of appearance in their respective movies.

Inter-subject correlation calculation and statistical evaluation

In the present analysis, we adopted inter-subject correlation (ISC) as an index of inter-subject synchronization of brain activity. There are two major methods used to calculate ISC: pairwise or leave-one-out approach (Nastase et al. 2019). The pairwise approach involved calculating n(n-1)/2 correlation values for each voxel or parcel. In contrast, the leave-one-out approach computes ISC maps for each subject by correlating their BOLD timeseries with the average timeseries of the remaining subjects. Here, we chose the pairwise approach to account for individual variability in BOLD timeseries data, as the leave-one-out approach averages the activity time courses, potentially overlooking such variability. The synchronization of the parcels is summarized by either the mean or the median of the pairwise correlation values. This process is repeated for all protagonists and antagonists.

The next step is to evaluate the significance of the calculated ISC values. We adopted subject-wise bootstrapping, a non-parametric method that is often used when assumption of independence is violated during pairwise correlation approach. Subject-wise bootstrapping involves randomly sampling N subjects with replacement multiple times to create a bootstrap distribution. In this paper, we performed statistical evaluations using bootstrapping with 5000 samples. We used the median to represent ISC values for each parcel since bootstrapping samples with replacement, meaning it could sample the same participant, may lead to absurdly high correlations. Therefore, the median is the preferred choice when using subject-wise bootstrapping (Chen et al. 2016, Nastase et al. 2019). As we are testing across 100 parcels, we applied Bonferroni correction (P < .0005) for thresholding.

Inter-subject correlation comparison between protagonists versus antagonists

Fisher’s z transformation was performed prior to statistically comparing ISCs for each parcel. As the characters had different lengths of TRs, we randomly sampled TRs from characters with more TRs (Sherlock and Dmitri) to match the number of TRs of characters with less TRs (Mycroft and Gustave). We then employed paired t-tests to compare ISCs across conditions (protagonists vs antagonists) per parcel (Fig. 2). This allowed us to determine which condition (i.e. protagonist or antagonist) produced greater synchronization in each parcel. To ensure that the results of the paired t-tests were not driven by TR selection, we repeated the process of randomly sampling TRs, ISC calculation, and paired t-tests over 1000 iterations. We note that this procedure allowed us to focus on the neural responses to the perception of each character, regardless of narrative development (e.g. sampled TRs may be a collection of frames from different parts of the story, rather than a single coherent scene)—which may be a topic of scientific inquiry on its own right, but beyond the scope of the present study. If ISCs between characters differed significantly (Bonferroni-corrected P < .0005) in more than 950 iterations (i.e. consistently different in >95% of observations), we considered that parcel to synchronize differently between protagonists versus antagonists. Follow-up ISC analysis was conducted using Neurosynth parcellation (de la Vega et al. 2016) to consider involvement of amygdala, which was not covered in Schaefer-Yeo cortical parcellation. The amygdala was examined because it is often related to emotional processing (Adolphs et al. 1995, Baxter and Croxson 2012, Mattavelli et al. 2014). Examining the amygdala was relevant as we assumed that emotional responses in some capacity are likely to accompany these characters, since the conflict between protagonists and antagonists is one of the major driving forces of any narrative. This procedure was performed separately for the Sherlock and The Grand Budapest Hotel datasets.

Figure 2.

To calculate whole-brain ISC values, we computed pairwise Pearson correlations of BOLD time course data for the 100 regions of interest (ROIs) based on the annotated TRs of each character. Median pairwise ISC values in each ROI became the final ISC value of the ROI. To compare the ISC values between narrative roles, TRs of the characters with longer screen time (i.e. Sherlock and Dmitri) were randomly sampled within the annotated TRs to match the number of TRs of the characters with shorter screen time (i.e. Mycroft and Gustave). After computing pairwise correlations of BOLD time course data, paired t-tests between the narrative roles were performed on an ROI-by-ROI basis. Random sampling of the TRs, ISC calculation, and paired t-tests were repeated for 1000 iterations to ensure that the results were consistent regardless of TR selection. We considered the ROI to be differentially synchronized if the result was observed in 95% or more iterations.

Inter-subject correlation comparison while retaining the continuous nature of the narrative

While the decision to randomly sample TRs across 1000 iterations was to prevent arbitrary selection of TRs and maximize the generalizability of the ISC results, this approach necessitates the disruption of the temporal continuity of the data, which may end up limiting the advantages of using naturalistic stimuli. As such, we proceeded to analyze the data while retaining the continuous nature of the narrative while matching the density and quantity of the TRs for the protagonists and antagonists. We z-scored the timeseries data within each segment, concatenated the segments, and selected the TRs so that the density and quantity would match between the characters (Fig. 1). Density of TRs was calculated by the number of TRs within a temporal window over the length of the window.

If there were several timepoints with the same density value, the timepoints were excluded when the temporal window starts or ends in the middle of a continuous segment. A total of 88 TRs were included for the characters in Sherlock and 36 TRs for the characters in The Grand Budapest Hotel.

Low-level audiovisual features

To address the possibility that low-level features of the stimuli, such as luminance or audio amplitude, could affect the ISC values, we ran additional analyses to compare audio amplitude and luminance during the appearance of protagonists and antagonists to take account of the impact of such external features. To compute audio amplitude during the appearance of the characters, we first separated the audio clip from the video, and extracted root mean square (RMS) values on a frame-by-frame basis. Then, the RMS timeseries data were downsampled to 1 s for The Grand Budapest Hotel and 1.5 s for the Sherlock dataset to match the TR length. In a similar manner, we also computed mean luminance of each TR by converting the frame to grayscale and calculating the mean value of the grayscale. Then, once again the framewise luminance data were downscaled to 1 s for The Grand Budapest Hotel and 1.5 s for the Sherlock dataset. Lastly, we conducted Welch’s t-tests to compare the audio amplitude and luminance between the protagonists and antagonists from each dataset.

Results

Whole-brain inter-subject correlation

We performed whole-brain ISC analysis, as described by Hasson and colleagues (Hasson et al. 2004), to identify regions that were engaged during the appearances of antagonists and protagonists in both the Sherlock and The Grand Budapest Hotel datasets. As mentioned above, in the first episode of Sherlock, Sherlock and Mycroft effectively served as the protagonist and the antagonist, respectively. Whole-brain ISC analysis revealed a total of 93 regions that demonstrated statistically significant synchronization during the appearance of Sherlock (Supplementary Table S1). When categorized into canonical functional networks (Schaefer et al. 2018), 17 regions were assigned to the visual network, 9 to the somatomotor network, 15 to the dorsal attention network, 12 to the ventral attention network, 5 to the limbic network, 12 to the control network, and 23 to the default network (Fig. 3). On the other hand, 57 regions were shown to synchronize during the appearance of Mycroft (Supplementary Table S2), comprising 16 regions from the visual network, 6 from the somatomotor network, 10 from the dorsal attention network, 2 from the ventral attention network, 1 from the limbic network, 7 from the control network, and 15 from the default network (Fig. 3). In the additional follow-up analysis, we found that the ISC value of the amygdala during the appearance of Sherlock was not significant (ISC = 0.073, ns). Amygdala ISC was significant for Mycroft only at an uncorrected threshold (ISC = 0.078, P < .05).

Figure 3.

Results of the whole-brain ISC analysis during each of the character’s appearance (Bonferroni-corrected P < .0005). Brain regions with significant ISC values were categorized into Yeo 7 networks. SomMot = Somatomotor, DorsA = Dorsal attention, VentA = Ventral attention, Lim = Limbic, Cont = Control.

In The Grand Budapest Hotel dataset, whole-brain ISC analysis showed that 81 regions synchronized during the appearance of Gustave (Supplementary Table S3). These were assigned to the canonical functional networks as follows: 17 regions to the visual network, 11 to the somatomotor network, 15 to the dorsal attention network, 12 to the ventral attention network, 5 to the limbic network, 12 to the control network, and 23 to the default network (Fig. 3). Conversely, during the appearance of Dmitri, 82 regions showed significant synchronization (Supplementary Table S4). Among them, 17 regions were assigned to the visual network, 7 to the somatomotor network, 14 to the dorsal attention network, 10 to the ventral attention network, 2 to the limbic network, 11 to the control network, and 21 to the default network (Fig. 3). Follow-up analysis showed that both amygdala ISC values during the appearance of Gustave and Dmitri were not significant (Gustave ISC = 0.068, ns; Dmitri ISC = 0.044, ns).

Generalizable inter-subject correlation differences between protagonists versus antagonists

To determine regions with significantly different ISC values, we conducted 1000 iterations of paired t-tests for each parcel with the length of appearance controlled, given that each character from each dataset appeared on screen for different durations. If the results consistently showed significant differences between characters (Sherlock vs Mycroft and Gustave vs Dmitri) in over 950 iterations (i.e. >95% observations), that parcel was considered to exhibit distinctive synchronization based on narrative roles (protagonist vs antagonist). Across both datasets, we observed four parcels (parcels 10, 42, 43, and 59) that consistently displayed different synchronization patterns between characters with opposite narrative roles (Fig. 4). According to the Harvard-Oxford cortical structural atlas (Makris et al. 2006), parcel 10 corresponds to the left auditory cortex, parcels 42 and 43 correspond to the left inferior frontal gyrus (IFG) and orbitofrontal cortex (OFC), and parcel 59 corresponds to the right auditory cortex. Of these four, parcels 42 and 43 exhibited higher ISC during the appearance of the protagonists compared to the antagonists (parcel 42: Sherlock ISC = 0.123; Mycroft ISC = 0.010, ns; Gustave ISC = 0.163; Dmitri ISC = 0.061, ns; parcel 43: Sherlock ISC = 0.217; Mycroft ISC = 0.072, ns; Gustave ISC = 0.145; Dmitri ISC = 0.036, ns, all Ps = .0002 unless otherwise noted) (Fig. 5). Conversely, the other two parcels (10 and 59) showed higher ISC during the appearance of the antagonists compared to the protagonists (parcel 10: Sherlock ISC = 0.439; Mycroft ISC = 0.543; Gustave ISC = 0.527; Dmitri ISC = 0.620; parcel 59: Sherlock ISC = 0.383; Mycroft ISC = 0.494; Gustave ISC = 0.426; Dmitri ISC = 0.522, all Ps = .0002 unless otherwise noted) (Fig. 5). The results of follow-up analysis in both Sherlock and The Grand Budapest Hotel showed that the amygdala ISC values between the protagonists and antagonists were not significantly different.

Figure 4.

Summary of all ROIs with their number of iterations in which significant differences between protagonists and antagonists were observed. The horizontal gray dotted line marks 950 iterations. Gray bars represent the ROIs that the ISC values between the characters did not show consistent difference (>95% of observations) from 1000 iterations. Red bars represent ROIs whose ISC during the appearance of the protagonist was consistently greater than that of the antagonist, while blue bars show ROIs whose ISC during the appearance of the antagonist was consistently greater than that of the protagonist (>95% of observations). The overlapping ROIs between the two datasets are highlighted in red boxes (parcels 10, 42, 43, and 59). The horizontal color bars show which of the Yeo 7 networks the ROIs are categorized into. SomMot = Somatomotor, DorsA = Dorsal attention, VentA = Ventral attention, Lim = Limbic, Cont = Control.

Figure 5.

Parcels that showed distinctive synchronization pattern based on narrative roles when the TRs were randomly sampled across 1000 iterations (Bonferroni-corrected P < .0005). Darker colors represent greater ISC values. Across both datasets, parcels 42 and 43 showed greater synchronization in response to the protagonists, while parcels 10 and 59 showed greater synchronization in response to the antagonists. IFG = inferior frontal gyrus, OFC = orbitofrontal cortex.

Converging inter-subject correlation differences when retaining the continuous nature of the narrative

A total of six parcels (parcels 10, 11, 16, 43, 53, 59) showed significantly different synchronization levels across two datasets (Fig. 6). Notably, these include the same 3 out of 4 parcels (parcels 10, 43, 59) that were identified using the random sampling approach used in the main analyses. Specifically, parcels 16 (left lateral occipital cortex/inferior temporal gyrus), 43 (left inferior frontal gyrus), and 53 (right lateral occipital cortex/inferior temporal gyrus) showed higher ISC during the appearance of the protagonists parcel 16: Sherlock ISC = 0.551; Mycroft ISC = 0.149; Gustave ISC = 0.248; Dmitri ISC = 0.161, parcel 43: Sherlock ISC = 0.351; Mycroft ISC = 0.141; Gustave ISC = 0.145; Dmitri ISC = 0.046, ns, parcel 53: Sherlock ISC = 0.467; Mycroft ISC = 0.264; Gustave ISC = 0.364; Dmitri ISC = 0.232 all Ps = .0002 unless otherwise noted). On the other hand, parcels 10 (left auditory cortex), 11 (left central opercular cortex/insular cortex), and 59 (right auditory cortex) showed higher ISC during the appearance of the antagonists compared to the protagonists (parcel 10: Sherlock ISC = 0.358; Mycroft ISC = 0.555; Gustave ISC = 0.587; Dmitri ISC = 0.733, parcel 11: Sherlock ISC = 0.099; Mycroft ISC = 0.171; Gustave ISC = 0.135; Dmitri ISC = 0.199, parcel 59: Sherlock ISC = 0.307; Mycroft ISC = 0.477; Gustave ISC = 0.482; Dmitri ISC = 0.633 all Ps = .0002 unless otherwise noted) (Fig. 6). Since the OFC (parcel 42) was no longer significant in the present analysis, we focus on interpreting the remaining three parcels (10, 43, 59) that were consistent across the two approaches.

Figure 6.

Parcels that showed distinctive synchronization pattern based on narrative roles when the continuous nature of the narrative was retained (Bonferroni-corrected P < .0005). Darker colors represent greater ISC values. Across both datasets, parcels 16, 43, and 53 showed greater synchronization in response to the protagonists, while parcels 10, 11, and 59 showed greater synchronization in response to the antagonists. IFG = inferior frontal gyrus, LOC = lateral occipital cortex.

No differences in low-level audiovisual features between protagonists versus antagonists

Comparison of low-level audiovisual features provided further evidence that the ISC results are not likely to due to simple differences in luminance and audio amplitude between protagonists and antagonists. Although some differences in luminance and audio amplitude were observed (audio amplitude in Sherlock and The Grand Budapest Hotel and luminance in Sherlock, Supplementary Fig. S1), the directions of these differences were not consistent (e.g. audio amplitude was greater for the protagonist in Sherlock, and the antagonist in The Grand Budapest Hotel), therefore eliminating the possibility of a systematic impact of such low-level audiovisual features on our main findings.

Discussion

Here, we sought to identify brain regions that represent information about narrative roles by comparing protagonists and antagonists from two movie-watching fMRI datasets. Neural synchronization was increased in a network of brain regions including the inferior frontal gyrus (IFG) in response to protagonists. Meanwhile, activity of the bilateral auditory cortex and its neighboring regions showed greater neural synchronization during the appearance of the antagonists. Notably, these findings were generalized across two independent datasets.

We hypothesized that the DMN would exhibit differential neural synchronization patterns during the appearance of protagonists and antagonists. This was based on recent studies highlighting the DMN’s role in narrative engagement (Song et al. 2021, Ohad and Yeshurun 2023), social categorization by narrative roles (Ron et al. 2022), narrative comprehension (Nguyen et al. 2019), self-referential thinking (Philippi et al. 2015), and semantic information integration (Yang et al. 2023). We note that parcel 43—the IFG region that showed increased synchronization to protagonists vs. antagonists—was categorized as the DMN according to the Schaefer-Yeo cortical atlas (Schaefer et al. 2018). However, we are careful not to claim that the DMN hypothesis was supported based on this finding, as the IFG is not considered to be a traditional component of the DMN.

The IFG showed greater neural synchronization in response to protagonists. The IFG is essential to semantic processing. The IFG is associated with the level of immersion in a fictional narrative (Metz-Lutz et al. 2010), suggesting that entering the constructed reality in a play depends on verbal processing. The IFG also unifies conflicting prior world knowledge and given information (Hagoort et al. 2004, Yeshurun et al. 2017). Relatedly, based on the previous research showing that IFG activation increased in high-context conditions than in low-context conditions (Keidel et al. 2018), it can be inferred that the IFG may be involved in the integration of previous and new information (Thompson-Schill et al. 1997, Greenberg et al. 2005). Since the narrative revolves around the protagonists, the audience has more information about them than the antagonists, leading to higher involvement of information integration process.

Furthermore, the IFG has been suggested to be crucial in empathy and social behavior. Indeed, the IFG was involved while reading about protagonists and inferring their actions and intentions (Mason and Just 2011). Interestingly, IFG function is known to be more relevant in processing emotional empathy than cognitive empathy (Oliver et al. 2018). Other studies have found that IFG lesions lead to poor performance in emotional empathy tasks (Shamay-Tsoory et al. 2009) or emotion inference tasks (Dal Monte et al. 2014). Thus, supposing that people show greater emotional empathy towards protagonists than toward antagonists, our IFG results may reflect the differences in emotional empathy based on opposing narrative roles.

Conversely, regions that include mainly the bilateral auditory cortex [Heschl’s gyrus, planum temporale, superior temporal gyrus (STG)], the parietal operculum, and temporal pole showed higher synchronization during the appearance of the antagonists. The STG is critical to auditory processing and speech perception (Chang et al. 2010, Mesgarani et al. 2014), but some studies have found it to support functions related to the perception of emotional content during audiovisual processing (Phillips et al. 1998) and emotional learning (Grosso et al. 2015). As for the bilateral parietal operculum and temporal pole, an interesting interpretation is possible considering the appearance of the antagonist typically entails threatening situations to the protagonist. The parietal operculum has been reported to exhibit increased activity when experiencing threatening stimuli (Straube and Miltner 2011) and to demonstrate distinct brain patterns based on the emotions elicited by external stimuli (Sachs et al. 2018). Similarly, the temporal pole is often involved in selecting the most appropriate behavior in social situations. According to Frith et al. (2003), the temporal pole helps generate a broader semantic and emotional context for given information from past experiences. In social situations, this context can be referred to as a “script,” a sequence of behavior that is considered proper and typical in a specific, well-known context (Schank and Abelson 1917). As such, the STG and its surrounding regions could elicit an emotional response and infer the intention, computing the possible sequence of actions the protagonists might take under the antagonist’s threat.

Our findings build upon recent studies that highlight the neural representations of narrative roles (Ron et al. 2022, Ohad and Yeshurun 2023) or fictional characters (Broom et al. 2021). However, important as they are, these studies either utilized static visual stimuli or characters from a single narrative. To increase the generalizability of the findings and ecological validity, we used two independent datasets to identify and validate brain regions that show the same response based on narrative roles. As observed in our results, neural synchronization patterns during movie-watching vary substantially depending on the movie’s characteristics. Since the involvement of many brain regions would likely be dependent on the movie per se, future movie-watching fMRI studies would benefit from validating with multiple different narratives, especially when dealing with high-level social processes.

Our study converges with and critically extends the work by Ohad and Yeshurun (2023). Both studies share a similar approach in that narrative-driven naturalistic stimuli are used to examine fMRI-based neural synchronization patterns. It is worth noting that there are several key differences between studies, such as the use of audiovisual versus audio-only narratives, examination of the degree of engagement with the narrative, and inclusion of multiple narrative stimuli. Ohad and Yeshurun (2023) observed a dissociation in neural synchronization patterns during the appearance of positively and negatively engaging characters. Our study, on the other hand, focused on identifying neural synchronization differences based on the narrative role of the characters. As people generally tend to be more positively engaged toward the protagonists and negatively engaged toward the antagonists, the present findings may be interpreted in the context of narrative engagement. Moreover, although we conducted an exploratory investigation to identify brain regions with different synchronization patterns based on the narrative roles, beyond the likability or valence of each character, some people may be inclined to root for villains (Keen et al. 2012), and some narratives might have protagonists who rarely earn empathy. While we left this possibility untested, future studies considering such factors would help clarify the underlying mechanism for different narrative roles to engage different neural synchronization patterns.

As we sought to test the hypothesis that the DMN would show differential patterns of neural synchronization during the appearance protagonists and antagonists, we decided to focus on cortical parcellation. One caveat of using a cortical parcellation atlas (Schaefer et al. 2018) is omitting subcortical areas. Our additional analyses of the amygdala yielded no significant differences in synchronization, suggesting that the amygdala is not sensitive to information about narrative roles. As we did not survey to entirety of the subcortical regions, future studies may aim to examine the potential subcortical contributions in processing narrative roles. In addition, our strategy for computing ISCs for the protagonists and antagonists involved carefully annotating and selecting the TRs during the appearance of each character and concatenating them. Strictly speaking, this is not a natural timeseries but rather a string of neural signals sampled in order at multiple discrete time points. The fact that we observe consistent results across 1000 iterations of resampling TRs suggests that these character-specific timeseries are not completely artificial and likely carry meaningful signals. We also add that this strategy was adopted from a recent fMRI study that successfully delineated character-specific brain responses in a similar manner (Ohad and Yeshurun 2023). Also, the present analysis was unable to address whether the dissociation in synchronization between the protagonists and antagonists might be due to emotional perception or emotional experience (Lindquist et al. 2012, Saarimäki 2021). For instance, in the Sherlock dataset, scenes with the protagonist were rated as more positive than scenes with the antagonist (Supplementary Tables S8 and S9). Therefore, the present study cannot rule out the possibility that the differences in neural synchronization levels in the IFG might be explained by positive and negative affect or attitude toward the characters, regardless of their narrative roles.

Finally, during the 1000 iterations of resampling TRs, we found a set of brain regions whose ISCs were consistently different between protagonists and antagonists, yet the direction of these differences was inconsistent across the two datasets. Some were a part of the DMN, such as the angular gyrus and precuneus, all of which showed increased ISCs to the antagonist of The Grand Budapest Hotel but not to the antagonist of Sherlock. Accordingly, we refrain from claiming that the entirety of the DMN is engaged more to the protagonists versus antagonists. Interestingly, the majority of such brain regions were a part of the visual network. Other than the speculation that some visual feature of the characters might be driving this outcome, pinpointing the root of these consistent yet opposite effects in the visual network is challenging. Future movie-watching fMRI studies designed to carefully delineate narrative roles would better address these remaining questions.

In conclusion, we found that the IFG is conditionally involved in processing opposing narrative roles. These findings support the view that information integration and emotional empathy are automatically employed depending on a given character’s narrative role during passive movie watching. Taken together, this study adds to the current literature that different narrative roles involve high-order social cognitive processing and may facilitate a better understanding of person perception in more naturalistic settings.

Supplementary data

Supplementary data is available at SCAN online.

Conflict of interest

None declared.

Funding

This research was supported by the National Research Foundation of Korea (NRF-2021R1F1A1045988). This research was also supported by the Sungkyunkwan University and the BK21 FOUR (Graduate School Innovation) funded by the Ministry of Education (Korea) and the National Research Foundation of Korea.

Data availability

The datasets are publicly available at https://openneuro.org/datasets/ds001132/versions/00003 (Sherlock) and https://openneuro.org/datasets/ds003017/versions/1.0.3 (The Grand Budapest Hotel). The codes for the main analysis are accessible at https://doi.org/10.5281/zenodo.12783652.