Amount of fiction reading correlates with higher connectivity between cortical areas for language and mentalizing

Behavioral evidence suggests that engaging with fiction is positively correlated with social abilities. The rationale behind this link is that engaging with fiction and fictional characters may offer a ‘training mode’ for mentalizing and empathizing with sentient agents in the real world, analogous to a flight simulator for pilots. In this study, we explored the relationship between reading fiction and mentalizing by looking at brain network dynamics in 57 participants who varied on how much fiction they read in their daily lives. The hypothesis was that if reading fiction indeed trains mentalizing, a task that requires mentalizing –Like immersing in a fictional story and engaging with a protagonist-should elicit differences in brain network dynamics depending on how much people read. More specifically, more frequent readers should show increased connectivity within the theory of mind network (ToM) or between the ToM network and other brain networks. While brain activation was measured with fMRI, participants listened to two literary narratives. We computed time-course correlations between brain regions and compared the correlation values from listening to narratives to listening to an auditory baseline condition. The between-region correlations were then related to individual differences measures including the amount of fiction that participants consume in their daily lives. Our results show that there is a linear relationship between how much people read and the functional connectivity in areas known to be involved in language and mentalizing. This adds neurobiological credibility to the ‘fiction influences mentalizing abilities’ hypothesis as suggested on the basis of conceptual analysis.

1 connections between brain areas while people listened to literary short stories. While listening 2 and reading stories might train slightly different parts of the brain, higher order areas affected by 3 semantic and social factors should be invariant to input modality (29-32). We aim to provide 4 insight into the neurobiological mechanisms that links social cognition abilities and engagement 5 with narratives. We had a sample of healthy young participants (N=57)Listen to two literary 6 short stories while brain activation was measured with functional Magnetic Resonance Imaging 7 (fMRI). Participants also listened to a reversed speech version of the same stories, which served 8 as a low-level baseline. After the scanning session, participants filled out a questionnaire battery 9 including a brief questionnaire related to their reading habits. 10 In the analysis we extracted the time courses during listening to the narratives and listening to 11 reversed speech separately for each participant from brain regions spanning the whole cortical 12 sheet. We then computed correlations between all regions and computed the difference in 13 correlation values between listening to narratives versus listening to reversed speech. This 14 difference was correlated with individual differences in fiction reading across participants. 15 Individual differences in brain connectivity are more robust in connected stimulus presentation 16 such as movies or narratives compared to isolated event related stimuli or rest (33). 17 In our analysis, we targeted the whole cortical sheet in order to not exclude certain regions a 18 priori. We did however have hypotheses about certain networks in which we expected that 19 differences in reading habits would have an influence on connectivity with other regions. First, 20 these are the regions of the mentalizing or ToM network (in the remainder of the paper we will 21 refer to these as 'mentalizing network') encompassing the anterior part of the medial prefrontal 22 cortex, the temporo-parietal junctions bilaterally, and the anterior temporal cortex bilaterally 23 (22,34,35). Second, as regions involved in empathizing we define the anterior insulae, and the 1 anterior cingulate cortex (ACC) (11). If there is an influence of fiction reading on social 2 cognition abilities, we expect the density of functional connections of parts of the mentalizing 3 and empathizing brain networks to be positively related to the amount of fiction reading that our 4 participants report in their daily lives. 5 Since reading is related to language, we also expected that brain regions involved in language 6 comprehension are sensitive to how much people read (36), and might show covariation and 7 connection with the mentalizing regions. Among the regions we expected to be involved were 8 the inferior frontal cortex bilaterally, the left posterior middle temporal gyrus, and the left 9 angular gyrus (37-40). Some of these areas are also known to be involved in emotional language 10 processing and mentalizing (41). One reason to not target these networks directly, but instead 11 employ a whole brain analysis approach, was that a lot of previous research has not investigated 12 mentalizing or language processes at the discourse level (with notable exceptions, see Ferstl, 13 2010 for overview).Limiting our analysis to a priori defined networks would hence increase the 14 chance of overlooking areas that are not described in previous literature (42). 15 Methods and Materials 16 Participants 17 Sixty participants took part in the experiment. Data from three participants were removed from 18 subsequent analysis. One participant had a data acquisition artifact in one of the runs, and two 19 other participants had more than 15% of data corrected in at least one of the runs in the motion 20 detection procedure (see below). All analyses are hence performed on data of 57 participants (31 21 women, 26 men; mean age 22.7 years; s.d. 3.64,Range 18-35; 9Left-handed by self-report, see 22 Willems et al. (43) for reasons to not exclude left-handed participants). All participants were 1 healthy volunteers with Dutch as their native language, and without a history of neurological or 2 psychiatric illness. The study was approved by an external ethics committee (CMO Committee 3 on research Involving Human Subjects, Arnhem-Nijmegen, Netherlands, protocol number 4 2001/095), and participants gave informed consent in accordance with the Declaration of 5 Helsinki. Participants were paid for participation. 6 Stimuli 7 Literary Stories and baseline 8 Stimuli were audio recordings of two literary short stories, and a reversed speech baseline of 9 parts of the same stories (Table 1). There were two minimally different versions of both stories 10 and each participant listened to one version of each story. The two different versions only 11 differed in the personal pronouns referring to the protagonist of the story. The original study for 12 which this data was recorded showed that the switching of pronouns has no effect on processing 13 or activated brain regions (44). The reversed speech baseline was created by taking the first half 14 of one and the second half of the other story played backwards, rendering it unintelligible.
15 Natural breaks in the content of the stories (paragraph end) were used to define the cut off, which 16 results in the reversed speech recording being slightly longer than both stories. Since reversed 17 speech is unintelligible, we only used half of each recording for the reversed speech baseline to 18 save time on the scanning protocol. The reasoning for including this low-level baseline was to a) 19 be able to subtract activations linked to low level processes like perception in addition to 20 individual intrinsic connectivity, and b) to have a 'task' that does not involve introspective or 21 narrative processing to get a reliable estimate of individual resting state connectivity in cortical 22 networks involved in processing discourse and mentalizing. 23 1 3 incorrect answers from the sum of all correct answers (total score can vary between -12 to 30). 4 We included both the questions regarding current reading behavior and the ART because they 5 target slightly different aspects of reading. The ART measures lifetime print exposure, whereas 6 our questions regarding reading targeted more current habits. Since the hypotheses we tested are 7 agnostic to this aspect, we decided to test for both lifetime exposure and current habits.
8 In addition, participants filled out the 40 item version of the Empathy Quotient (EQ) 9 questionnaire (47). The EQ questionnaire is a self-report measure of trait empathy consisting of 10 40 questions testing how sensitive individuals are to their own and others' emotions and is 11 widely used as a diagnostic tool for social and developmental disorders. Individual scores can 12 range between 0 and 80, and a score of 32 or lower indicates atypical performance. For this 13 study, we used the EQ to check the distribution of EQ in our sample. We also conducted a 14 control analysis to test the sensitivity of our analysis to differences in brain connectivity 15 depending on scores on individual differences measures. People who score higher on the EQ 16 should show a modulation of brain connectivity in brain areas related to social cognition 17 depending on their individual EQ scores if these types of correlations are sensible. This tests 18 whether our fiction reading measures and their modulation of brain connectivity is sensible. Stimuli were presented to participants through MR-compatible ear plugs which additionally 5 served to reduce the noise from the scanner. All participants could hear the stimuli well, as 6 assessed by a volume test with an excerpt from another audio story spoken by the same speaker.
7 The scanner was collecting images during the volume test (and hence making loud noises).
8 Participants indicated whether volume should be increased or decreased during the volume test, 9 to be able to comfortably listen to the materials. None of the participants asked for the volume to 10 be increased to the maximum level, indicating that they could all hear the stories comfortably. 3 se, was the fact that correlation estimates tend to be unstable with sample sizes as employed here.
4 By combining correlation coefficients in node degree, we only look at regions which have higher 5 correlation values for many region-region connections.
6 A randomization approach was taken to assess which number of connections constitutes a 7 statistically significant effect. The critical value for node degree was established by calculating 8 node degree after randomizing the individual difference scores for each individual difference 9 score separately. This procedure was repeated 2,500 times, and the maximal node degree across 10 regions for each randomization was used to build a reference distribution. The critical value for 11 node degree was the value in the randomization distribution compared to which only 2.5% of 12 randomization produced a more extreme value (corresponding to p<0.05 two-sided, family-wise 13 error rate). In this way, we explicitly tested which regions had a statistically significant number 14 of connections with other regions influenced by a given individual difference score. The analysis 15 showed that 6Regions or more was the maximal critical value across individual difference 16 measures, and we adopted this value to correct for multiple comparisons. Behavioral results 3 There was considerable spread in reading behavior as reported by participants ( Fig. 2A).  12 not feel inhibited to report that they hardly ever read (see Fig. 2A). While the reading habits 13 questions target more current lifestyle habits, the ART questionnaire captures lifetime exposure 14 to fiction better, but does not necessarily reflect current habits. Given that the ART also has its 15 shortcomings (one can score very high on the ART without actually reading books), we had no 16 reason to value the ART higher above the self-report measures, or vice versa. The EQ 17 questionnaire did not correlate significantly with any of the other measures (all rho<|0.13|, all 18 p>0.3). 19 Most important for the present individual differences approach is that there was considerable 20 spread in all measures, and most notably in those measuring amount of fiction reading. 6 and orbitofrontal cortex, temporo-parietal junction, and the right frontal pole (see Table 2, Figure   7 3).
8    Figure 6A). In these regions, the number of region-region by individual difference correlations 9 was statistically higher than expected by chance. This effect was present in a set of regions 10 including (among others) bilateral inferior frontal sulci, bilateral anterior middle temporal 11 regions, anterior medial prefrontal cortex (MPFC),Right posterior supramarginal gyrus, the 12 lingual gyri bilaterally, and the middle frontal cortex bilaterally (see Table 3, Fig 4, Figure 6A).
13 The effect was also present in several regions in sensory and motor cortices. 14 1 For an illustration of the location of the regions see Figure 1A, and Supplementary   7 Another set of areas was found to have higher node degree depending on the score on the ART 8 (Table 4, Fig. 5, Figure 6B). This network partially overlapped with the regions showing 9 increased connectivity depending on how much the participant reported to read and included 10 (among others) the bilateral inferior frontal sulci, the anterior medial prefrontal cortex 11 (MPFC),Left angular gyrus, and left temporal pole (see Table 4, Fig. 5, Figure 6B).
12  Figure 1A, and  21 Several of these regions are not only parts of the mentalizing and language networks but also part 22 of the DMN. The DMN seems to be involved in aspects of cognition that are linked to aesthetic 1 experience, self-relevance and self-referential thinking, episodic thinking (28), and non-directed 2 attention (e.g. mind wandering, flow). It is thus not surprising that several studies found it to be 3 involved in narrative processing (6,25,26). A fundamental aspect of engaging with literature is the 4 aesthetic appraisal of the language used and the plot built (e.g. suspense and twists), but also the 5 cultural, social, and self-referential search for greater meaning. Narrative processing naturally 6 involves episodic structuring of events in a story. Immersing in a story also involves decrease in 7 executive control and directed attention, as participants allow the story to lead their attention 8 instead of actively seeking information.

(node degree, ordered from Largest to smallest). For an illustration of the Location of the Regions see
9 Before we discuss the most important regions which showed sensitivity to the amount of fiction 10 reading, an important point needs to be made. In our analysis, we establish which regions have 11 more functional connections with other regions during listening to a literary narrative, depending 12 on how much participants read fiction. The hypothesis (as discussed above) is that this is a training 13 effect: through repeated exposure to fiction, these areas are activated together more frequently, 14 and hence form a network that is more easily activated as compared to people who have less 15 exposure to fiction. While this is a viable hypothesis, an alternative hypothesis is that people with 16 stronger connections in these networks enjoy reading more and hence are more likely to be avid 17 readers. These two hypotheses are not mutually exclusive, and it is important to point out that our 18 experimental design is agnostic to the directionality of this relationship. 19 All regions that show differences depending on how much people read are regions that are typically 20 found in tasks related to social cognition and language processing, including sensory and motor 21 areas that are typically involved in processing semantic information about events and characters.
22 The left and right inferior frontal regions are the first notable regions whose number of connections 23 showed a sensitivity to amount of fiction reading. The involvement of these regions as part of the 1 language network is well-established (40), and especially right inferior frontal gyrus (RIFG) has 2 been implicated in discourse comprehension (54,55). In the present results the RIFG formed 3 connections with the left inferior frontal cortex, the lingual gyri, and the anterior part of the medial 4 frontal cortex -a prominent region in the mentalizing network (Fig. 7, Fig. 8). This suggests that 5 the RIFG could be an interfacing region between the language network and parts of the mentalizing 6 network.
7 The region to region connectivity pattern that was most affected by how much participants read 8 was the connectivity of the left and right inferior frontal gyri to the left and right lingual gyri. Left 9 IFG is a prominent region in language processing, linked to semantic selection and coordination, 10 whereas right IFG seems to be crucial for discourse processing (see discussion above). The 11 connectivity profile of the left lingual gyrus (one of the regions with the highest node degree) was 12 such that it connects with inferior frontal cortices bilaterally, the right angular gyrus, as well as the 13 precuneus bilaterally (Fig. 8). Both, the precuneus and right temporo-parietal regions are part of 14 the extended language network (54). The lingual gyrus is not typically considered part of the 15 language network (and hence was not hypothesized to be implicated in this study). However, 16 several fMRI studies report that the lingual gyri are activated during language comprehension (56-17 58). It has also been shown that the cortical thickness of this region co-varies with lateral temporal, 18 inferior parietal and inferior frontal regions often implicated in language, thus indicating that these 19 regions experience common activation, presumably related to language comprehension (59). A 20 full discussion of this region and its potential role in language comprehension is beyond the scope 21 of this paper (42). It does seem however, that it may be an area which has been somewhat 22 overlooked by the field. 1 Binder and colleagues labeled as the angular gyrus could functionally be the posterior 2 supramarginal gyrus in the current study. These regions are also adjacent to the temporo-parietal 3 junction, which is (bilaterally) considered central to the ToM network. This spatial proximity 4 might support an interface between cognitive systems for the integration of semantic and social 5 information.
6 In discussing the results, we have somewhat artificially distinguished between neural areas 7 important for language and neural areas important for social cognition. This distinction is well- 18 While the evidence of a relation between lifetime fiction exposure and differences in functional 19 network connectivity during listening to literary stories is purely correlational, our study confirms 20 the link between the neural processes involved in reading and in social cognition. We show that 21 reading fiction is linked to increased number of functional connections within and between 22 language and mentalizing regions and to other parts of the brain. Apart from established language-1 related regions like inferior frontal areas, regions recognized as the extended language network 2 seem to play an important role in the brain network processing narratives.
3 Interestingly, we found no correlation between trait empathy (EQ) and reading in our sample.
4 Previous studies also failed to find a link between self-reported trait measures for social skills and 5 reading behavior measures (12), while self-reported happiness in social relationships was found to 6 be related to fiction reading in the same sample. It is possible that trait empathy is in fact not related 7 to reading, not trainable, or that the social cognition skills that might be trained by reading are not 8 the same as measured by the EQ questionnaire. The EQ was originally designed as a diagnostic 9 tool for people with clinically relevant impairments and not to be a sensitive measure to scale 10 individual differences in trait empathy in healthy adults. While the EQ is still a widely used 11 measure for individual differences in social skills, it might in fact be insensitive for measuring 12 behaviorally relevant differences in healthy adults. In this study, we used the EQ as a control 13 analysis to test if our analyses are sensitive to region-region connectivity differences in social 14 cognition networks depending on one individual difference score. 15 The causal link between exposure to and initial interest in fiction is interesting, but far beyond the 16 scope of this paper. What is important to note is that the link between a life-time exposure to fiction 17 and social cognition seems to be more robust than brief exposure to fiction, as was shown by two 18 recent large-scale studies (14,15) which found that there is no immediate effect of social cognition 19 performance improvement after a short exposure to fiction, but long-term fiction exposure was 20 positively related to social cognition skills.
21 It is possible that there are other individual differences overlapping with how much people read 22 that could explain the results. For example, IQ is related to language skills that in turn are related