The Role of the Brain's Pragmatic Language Network in Reading Comprehension in Autistic Children

ABSTRACT

One of the earliest and commonly reported symptoms of autism spectrum disorder (ASD) is a delay in language development. Such delay may sometimes accompany deficits which can have a long‐term impact on reading comprehension. It is frequently reported that autistic children exhibit significant difficulties in pragmatics, which is the communicative use of language. While the focus of most studies on reading has been on comprehension, some have proposed a positive correlation between reading and pragmatics. Nevertheless, the neural mechanisms that underpin pragmatic language in autism remain poorly understood. The objective of this functional MRI study is to examine the differences in the brain's Pragmatic Network (PN) during two levels of reading tasks in autistic and neurotypical (NT) children. The study included children aged 8–13 years (VA task = 26 ASD and 15 NT; MS task = 25 ASD and 15 NT). The results demonstrate that while both groups engaged the PN, the ASD participants exhibited additional recruitment of PN areas that overlapped with language processing, contextual integration of linguistic information, and theory of mind. Furthermore, the ASD group, but not the NT group, showed a correlation between the percentage of signal change and reading comprehension. In addition to underscoring the role of the PN in reading comprehension, these findings point to increased engagement of the PN in autism.

Summary.

• Children with autism have significant difficulties in pragmatics, which is the communicative use of language.

• This study shows that brain areas associated with pragmatic language processing also play a role in reading comprehension tasks in autistic and neurotypical children.

• However, the autistic participants seem to engage a wider network of brain areas to accomplish such tasks.

• These findings suggest the relationship between pragmatic language and reading comprehension.

1. Introduction

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by difficulties with social communication and restricted and repetitive behaviors and interests (American Psychiatric Association ²⁰¹³). Autistic children show delays in retrieving word meaning from lexical memory and computing the syntactic and semantic relations (i.e., language comprehension) as early as in the first 2 years of their life (Mody and Belliveau 2013). Processing syntactic and semantic relations is not enough in many linguistic contexts and requires one to integrate different levels of information to understand the larger level meaning of the message (Friederici 2011). Other times, one must consider the message in a specific social context (Egorova et al. 2016; Escandell Vidal ²⁰⁰⁶) to better understand the meaning. Pragmatic language involves the integration of linguistic and contextual cues to infer the speaker's intention (Scott‐Phillips ²⁰¹⁷). This includes, but is not limited to, using and understanding figurative speech like speech acts, metaphors, sarcasm, puns, idioms, and irony. Thus, inferring the speaker's intention and going beyond what is uttered can be a challenging aspect of interpersonal communication, especially in conditions like autism. Difficulty in processing social communicative language is a central characteristic of autism, especially in verbally fluent autistic individuals. Research suggests that these difficulties may underlie deficits in social cognitive processes, like theory of mind (ToM) (Bambini and Bara 2012; Pexman ²⁰⁰⁸), in many but not all autistic individuals (Marocchini 2023). ToM refers to the ability to attribute mental states to others (Premack and Woodruff 1978).

Furthermore, delays in language development also have an impact on reading comprehension, learning, and academic success. Converging evidence suggests that 65% of autistic children have difficulties with reading comprehension (Nation et al. 2006), a complex skill that requires decoding phonological and semantic information, syntactic processing, morphology, and pragmatics (Nation and Norbury 2005). While most studies of reading have focused on comprehension, some suggest that improvements in comprehension may also facilitate more successful communication (Hartung and Willems 2020; Mar et al. ²⁰⁰⁶). A meta‐analysis of the skills related to reading comprehension in autistic children revealed mixed results (Duncan et al. 2021), with some finding a relationship between reading comprehension and pragmatics (Knight 2016) and others not (Jacobs and Richdale 2013). Nevertheless, studies in neurotypical (NT) populations have identified a correlation between reading and other skills related to social interactions. For example, Mar et al. (2006) found that reading fiction was positively correlated with ToM and empathizing skills.

Additional behavioral evidence suggests that reading and comprehending fiction enhances cognitive processes associated with pragmatic abilities, such as ToM, empathy, and social perception (Kidd and Castano 2013). On the other hand, neuroimaging studies examining reading comprehension in NT populations have shown the recruitment of an extended network of brain regions (reading network) in the left hemisphere (i.e., the posterior inferior occipital gyrus [IOG], fusiform gyrus [FG], posterior superior temporal gyrus [pSTG], precentral gyrus, intraparietal sulcus [IPS], supplementary motor area [SMA], inferior frontal gyrus [IFG], middle frontal gyrus [MFG], and thalamus (Koyama et al. 2011). A relatively recent functional magnetic resonance imaging (fMRI) study of reading comprehension found that while there was no statistically significant group differences in functional activation between autistic and NT children, the autistic children showed decreased functional connectivity between the left IFG and the left IOG (Bednarz et al. 2017). Another fMRI study of reading fiction in NT adults found reading ability to be correlated with language and ToM network functional connectivity (Hartung and Willems 2020).

Converging evidence points to a significant overlap between the brain regions associated with pragmatic language processing and those involved in general language processing. Some of these areas include the IFG, middle temporal gyrus (MTG), pSTG, and angular gyrus (AG) (Bohrn et al. 2012; Duvall et al. ²⁰²³; Friederici ²⁰¹¹; Rapp et al. ²⁰¹²; Reyes‐Aguilar et al. ²⁰¹⁸). Brain regions that are primarily active in processing ToM have also been found to be commonly involved in pragmatic language processing (Duvall et al. 2023; Schurz et al. ²⁰¹⁴). These include the medial prefrontal cortex (MPFC), supramarginal gyrus (SMG) (Bohrn et al. 2012; Reyes‐Aguilar et al. ²⁰¹⁸), and cingulate cortex (Rapp et al. 2012; Reyes‐Aguilar et al. ²⁰¹⁸). Consequently, there is some overlap of regions between the Pragmatic Network (PN) and language network, as well as between PN and ToM network.

Studies have shown autistic and NT individuals demonstrate differences in the recruitment of brain areas associated with pragmatic language processing. For instance, autistic children, compared to NT children, have shown greater activation in the left precentral gyrus and the bilateral IFG (BA 45) when processing irony comprehension (Wang et al. 2006). Furthermore, studies examining the PN have found that autistic, relative to NT, participants had decreased activation in the left IFG (Duvall et al. 2023) and increased activity in the right IFG (Tesink et al. 2009) in adult and child samples. Altogether, evidence suggests reading comprehension and pragmatic language networks recruit similar brain areas, specifically those classically associated with language processing, such as the IFG and pSTG (Duvall et al. 2023; Koyama et al. ²⁰¹¹; Reyes‐Aguilar et al. ²⁰¹⁸). In other words, regions associated with reading comprehension and pragmatic language may be subsets of regions in the broader language network, with some overlap.

While most studies of reading have focused on decoding and comprehension, some have suggested a relationship between reading and pragmatics (Hartung and Willems 2020; Kidd and Castano ²⁰¹³; Mar et al. ²⁰⁰⁶). Despite autistic children's notable difficulties with pragmatic language processing (American Psychiatric Association ²⁰¹³), there is a significant gap in understanding the underlying neural mechanisms of pragmatics and its relationship with reading. To date, no studies have investigated the role of the PN during reading comprehension in autistic population. The aim of this fMRI study is to address this gap by investigating the differences in the brain's PN during two reading tasks in ASD and NT children. We hypothesized that both ASD and NT groups would recruit the PN, but the extent of this recruitment may differ. In particular, it is anticipated that the autistic group will exhibit a compensatory mechanism (Hogeveen et al. 2020; Xu et al. ²⁰²²), which will be evidenced by increased activation in areas related to the PN. The findings of this study will provide important insights into the relationship between reading comprehension and pragmatics in autistic children, which, in turn, would have significant implications for further understanding social communication difficulties that are central to autism.

2. Method

The objective of this study is to assess the brain areas of the PN that are recruited during two reading tasks of increasing complexity in autistic and NT children. The two tasks were the verbal absurdity (VA) task at the sentence‐level (e.g., When I want to play baseball, I grab a swimsuit and go to the pool), and the multisentence (MS) task involving two or more sentences (e.g., Little chirping sounds came from the tree. Mother robin searched for worms. The robin eggs had hatched; for details see Murdaugh et al. ²⁰¹⁷). These tasks were selected due to a few reasons. First, they were designed to progressively increase computational demands, with the VA task requiring sentence‐level processing and the MS task involving more complex MS integration. This progression provided an opportunity to investigate functional changes as participants transitioned from simpler to more complex tasks. Second, both tasks involve significant pragmatic demands, requiring participants to integrate multiple streams of information to derive meaning, thus serving as a proxy for cognitive load. Consequently, these tasks offered a unique platform for assessing the neural changes associated with varying levels of cognitive demand. To further elucidate the differential increase in activation in each group, the percentage of BOLD signal change in the entire PN during each task was extracted. Finally, the relationship between the percentage of BOLD signal in the PN during each task and tests of reading skills was assessed through stepwise regression.

3. Participants

The study included children aged 8–13 years. All participants were included if they were right‐handed and native English speakers with a Full‐Scale Intelligence Quotient (FSIQ) above 75 as measured by the Wechsler Abbreviated Scale of Intelligence (WASI). Additionally, the autistic participants had a current diagnosis of ASD by a licensed clinical psychologist, which was confirmed by the autism diagnostic observation schedule (Lord et al. 2000) or the autism diagnostic interview‐revised (Lord et al. 1994). Autistic children had average word decoding abilities—a score above the 37th percentile on the Slosson Oral Reading Test—Revised (SORT‐R)—but difficulties with reading comprehension as indexed by a reading comprehension score below the 37th percentile on the Gray Oral Reading Test—fourth edition (GORT‐4). NT children had no diagnosis of an ASD; parental reports were used to confirm the absence of any psychological disorders; participants with any reported diagnosis were excluded. Additionally, NT children did not have a language disorder and had average (greater than the 25th percentile) word decoding abilities and reading comprehension, as measured by the SORT‐R and GORT‐4 comprehension. Participants who did not meet any of the inclusion criteria and participants who were currently taking beta‐blockers or vasodilators, had a history of ferromagnetic material in the body or neurostimulators, were claustrophobic, or had a history of kidney disease, seizure disorder, diabetes, hypertension, anemia, or sickle cell disease were excluded from the study. In addition, participants' parents completed the Social Responsiveness Scale (SRS), which identifies social difficulties (see Figure S2A–C), and two reading tasks inside of the MRI scanner. The ASD participants were recruited from the Civitan–Sparks Clinic at University of Alabama at Birmingham (UAB), Mitchell's Place for Autism in Birmingham, the Autism Spectrum Disorders Clinic at the University of Alabama, the Autism Society of Alabama, and Lindamood–Bell Learning Processes centers across the country. The NT participants were recruited through advertisements in local newspapers and in UAB Reporter, as well as through flyers posted on the UAB campus. At the time of their imaging sessions, all participants were medication naive.

The MRI data quality was evaluated using MRIQC. To assess head‐motion, we selected the framewise displacement measure from MRIQC (Esteban et al. 2017). Due to high head motion (> 3 mm), eight ASD and five NT participants were excluded from the VA task. Additionally, five ASD and one NT participants were excluded from the MS task. The final sample for VA was 26 autistic and 15 NT children; and for MS was 25 autistic and 15 NT children (see Table 1 for further information about the demographic breakdown). In the VA task autistic children (0.75 ± 0.68) exhibited slightly higher head‐motion compared to NT children (0.61 ± 0.66). However, these differences were not statistically significant (t(39) = 0.64, p = 0.52). Similarly, in the MS task, autistic children (0.80 ± 0.51) exhibited higher head‐motion compared to NT children (0.66 ± 0.43). However, these differences were not statistically significant (t(38) = 0.67, p = 0.51). In the autistic group, comorbidities included, including cases of attention‐deficit/hyperactivity disorder (ADHD) (n = 5), language disorders (n = 2), and a combination of a language disorder and ADHD (n = 2), childhood disintegrative disorder (1), and both ADHD and an anxiety or mood disorder (1). The procedures used in this study were in accordance with the Helsinki declaration and the University of Alabama at Birmingham Institutional Review Board for human subject protection approved the study. Participants' legal guardians signed a written informed consent, and participants gave written assent.

TABLE 1.

Demographics of participants by task.

	Verbal absurdity task (n = 41)
	ASD (n = 26)	NT (n = 15)	t	p
Sample size	5 F, 21 M	5 F, 10 M	0.40	0.52
Age	10.77 ± 1.39 (8–13)	10.62 ± 1.10 (8–14)	0.31	0.75
FSIQ	95.56 ± 11.88 (77–123)	98.40 ± 12.07 (78–118)	−0.75	0.47
GORT‐4	81 ± 12.08 (60–105)	108.33 ± 21.10 (80–170)	−4.59	0.0001***
SORT‐R	104.92 ± 6.89 (96–125)	109.53 ± 8.81 (91–116)	−1.74	0.095
SRS	74.92 ± 12.08 (41–90)	54.82 ± 19.31 (35–84)	3.18	0.05
Caucasian	15	8
Black	2	7
Asian	5	0
More than one	3	0
Hawaiian	1	0

	Multisentence task (n = 40)
	ASD (n = 25)	NT (n = 15)	t	p
Sample size	2 F, 23 M	3 F, 12 M	0.38	0.53
Age	10.84 ± 1.37 (8–13)	10.67 ± 1.47 (8–14)	0.37	0.72
FSIQ	92.96 ± 11.58 (77–123)	99.47 ± 11.29 (83–118)	−1.73	0.09
GORT‐4	79.38 ± 10.87 (60–95)	102.67 ± 14.13 (70–125)	−5.46	0.0001***
SORT‐R	105.46 ± 7.45 (95–122)	109.73 ± 7.81 (101–125)	−1.69	0.10
SRS	76.96 ± 11.96 (41–90)	49.91 ± 13.52 (35–83)	5.76	0.001***
Caucasian	12	9
Black	3	6
Asian	6	0
More than one	3	1
Hawaiian	1

Note: Statistics show the results of the t test for age, FSIQ, GORT, SORT, and SRD; and the results of the chi‐squared for gender, comparing autistic, and neurotypical children.

Abbreviations: FISQ = Full‐Scale Intelligence Quotient; GORT‐4 = Gray Oral Reading Test—fourth edition; SORT = Slosson Oran Reading Test—Revised; SRS = Social Responsiveness Scale.

^*** p < 0.001.

4. Experimental Paradigm

We examined brain activation during two fMRI tasks: VA and MS comprehension. All participants completed both tasks. Nevertheless, some participants' data from certain tasks were excluded due to high head motion. In the VA task, eight ASD and five NT were excluded; and in the MS task, five ASD and one NT were excluded. More details were provided in the exclusion criteria section. For each task, the stimuli were presented for 10 s, followed by a 3 s interstimulus interval. Participants were asked to relax while a 30 s fixation cross was presented at the beginning and end of the task to provide a baseline measure of brain activity. Each task was presented individually within a single run, with a total duration of 9 min. In order to minimize the impact of the stimuli, half of the participants were assigned to complete Version A of the task, while the other half were assigned to complete Version B of the task. Furthermore, to minimize the impact of the task itself, participants completed the VA and MS tasks in a different order (e.g., half of the participants were instructed to complete the MS first, while the other half were instructed to complete the VA task first). The total scanning time was 1 h, which included both tasks used for this experiment. The stimuli were presented in the scanner using E‐PRIME 1.2 facilitated by the Integrated Functional Imaging System (IFIS, In vivo Corporation, Orlando, FL). The vocabulary was familiar to the participants. We ensured this by presenting all the words before the testing to the participants to make sure they understood the meaning of individual words. Before the scan, participants practiced the tasks using a unique set of stimuli, which were not included in the version of the task presented during the scan.

4.1. Experiment 1: VA

In the VA task, participants read sentences and decided, via button press, if the second part of the sentence was incongruent (e.g., When I want to play baseball, I grab a swimsuit and go to the pool) or congruent (e.g., The giraffe and the elephant ate grass together under the hot African sun) to the first part of the sentence. In total, there were 30 incongruent and 30 congruent trials. The second section of the sentence has been underlined for the purpose of differentiating between the two sections of the sentence. However, the participants read the sentences without this underline.

4.2. Experiment 2: MS

In the MS task, a series of three sentences were presented in which participants made judgments, via button press (yes or no), about whether or not the last sentence presented could be a logical conclusion given the context of the previous two sentences (e.g., It was a hot summer day. There was no school. Tom went snowboarding). In the above example, the third sentence is incongruent with the first two so the participant would press the button corresponding to no. In the following example, “little chirping sound came from the tree. Mother robin searched for worms. The robin eggs had hatched,” the third sentence is congruent and therefore the participant would press the button corresponding to yes. In total, there were 11 incongruent and 11 congruent trials.

5. fMRI Data Analysis

The MRI data were acquired on a 3 Tesla Siemens Allegra MRI scanner with a 64‐channel head coil. The anatomical acquisition parameters were TE = 0.03 s, TR = 2.3 s, flip angle = 12, matrix size = 256 × 256. The functional MRI acquisition parameters were TE = 0.03 s, TR = 1 s, flip angle = 60, matrix size = 64 × 64. The transformation of dicom to nifti images was performed using the dcm2niix software (v1.0.20190902) (Li et al. 2016). The fMRI data preprocessing was performed using fmriprep (Esteban et al. 2019). The T1w reference and the MNI's (Montreal Neurological Institute) unbiased standard MRI template for pediatric data, from the 4.5 to 18.5‐year age range (MNIPediatricAsym), were selected for spatial normalization using nonlinear registration with antsRegistration (ANTs 2.3.3).

5.1. General Linear Model (GLM)

To investigate the changes in activity of specific areas involved in pragmatic language processing, at the individual and group level analysis, we used a mask based on a previous activation likelihood examination (ALE) meta‐analysis examining pragmatic language processing (compared to literal) (Reyes‐Aguilar et al. ²⁰¹⁸). To assess the brain areas associated with language processing, a GLM was performed using the FEAT tool from FMRIB Software Library (FSL; v6.0.3) software (Woolrich et al. 2004). For the first‐level analysis, three independent regressors were used in each GLM to model the hemodynamic responses. For both tasks, the regressors were congruent, incongruent, and fixation. To assess processing during reading comprehension as a whole, all the stimuli were included, both correct and incorrect, in the analyses. This approach was taken to prevent the removal of stimuli that were more challenging for the participants. In addition, separate analysis would have resulted in skewed comparison due to differences in the correct and incorrect trials. Ten physiological regressors were used: the cerebrospinal fluid, white matter, dvars, framewise displacement, and six motion regressors (translation in x, y, z directions and rotation in pitch, roll, yaw angles). Since fmriprep does not perform spatial smoothing, a 6 mm smoothing kernel was applied at this level. To study the neural correlates of language processing in the PN, mixed‐effects GLM models (i.e., Flame 1) were created. To examine group‐level effects as well as group differences in autistic and NT children, a within‐group and between‐group analyses with no repeated measures were performed. The results from the first level and group analysis were cluster corrected (z = 2.3, p‐threshold < 0.05). To report the activity associated with language processing, the contrast that assessed both language conditions (i.e., congruent, incongruent), were compared to the control condition (fixation cross) and are reported in the following section (i.e., congruent + incongruent > fixation).

To assess the changes in the recruitment of the PN during reading, PN was used as a mask in the first‐level GLM and in the group‐level analyses. The contrast language (congruent + incongruent) > fixation cross was used and was corrected at the cluster level (z = 2.3, p‐threshold < 0.05).

6. Percent BOLD Signal Change

To examine the differential activation in the PN during language comprehension, the percent BOLD signal change was extracted from the contrast congruent + incongruent > fixation. The regions of interest (ROIs) were selected from a previous meta‐analysis that included studies on pragmatic language (Duvall et al. 2023; Reyes‐Aguilar et al. ²⁰¹⁸). The ROIs were examined, and those that did not overlap were selected (see Table S1). The maxima coordinate of the meta‐analysis did not report the mPFC. Nevertheless, mPFC was part of this network as an extension of the maxima coordinates and has been reported previously as part of the pragmatic language processing (Duvall et al. 2023). Therefore, this area was selected from the analysis irony > literal contrast reported in the meta‐analysis previously described. A total of 10 ROIs were included in this analysis (see Table S1 for details). The % of BOLD signal was extracted for each subject, task, and ROI using the featquery tool of FSL. An 8 mm sphere was created for each ROI using the MNI coordinates using fslmaths. Shapiro tests showed that not all the ROIs were normally distributed (p > 0.05), and the Levene's test showed that the variances between the statement categories were equal (p < 0.05).

To improve model performance, the data were normalized. Outliers were then removed per group, task, and ROI. For each task, a Welch's t test with FDR correction was performed to assess the differences between autistic and NT children. Finally, six stepwise AIC models were created to assess the variables associated with the % of BOLD signal change of the PN during the reading tasks. The variables selected were GORT‐4, SORT‐R, and SRS total T score, which assess reading comprehension, word decoding, and social difficulties associated with ASD, respectively. For this step, outliers were removed from each measure. For each task, models were created for each group individually and for both groups combined. Significance was set at p < 0.05, and a Bonferroni's correction was used to correct for multiple comparisons.

7. Results

7.1. VA

In the VA task, within‐group analyses revealed autistic participants showing increased activity in three clusters in the left hemisphere: (1) the fusiform, extending to the MTG and the AG; (2) the IFG pars opercularis, extending to the IFG orbitalis; and (3) the premotor + SMA, extending to the dmPFC, rmPFC, and the frontal eye fields (FEFs; see Figure 1A and Table 2). The NT group showed increased activity in the following two clusters: (1) the IFG pars opercularis, extending to IFG orbitalis and pars triangularis and (2) the MTG (see Figure 1B and Table 2). The group difference analysis (ASD > NT and NT > ASD) did not show significant activation differences at a cluster corrected statistical threshold (z = 2.3). Nevertheless, with an uncorrected threshold (p < 0.001) and a minimum cluster size of 50 contiguous voxels, the ASD > NT showed increased activation in the following five clusters. The cluster size was set at a minimum of 50 contiguous voxels to reduce the risk of the cluster occurring by chance. The results showed increased activation in (1) the left fusiform, extending to the MTG; (2) the right mPFC, extending to the dmPFC; (3) the left primary auditory, extending to the STG; (4) the left temporal pole; and (5) the left dlPFC (see Figure 1C and Table 2).

FIGURE 1.

Brain areas exhibiting activation during verbal absurdity task. Within results for ASD (A) and NT (B) are cluster corrected at 2.3. The between group (C) ASD > NT is uncorrected at p < 0.001 with minimum of 50 continuous voxels.

TABLE 2.

MNI coordinates of the regions showing significant activation in autistic and neurotypical children during the verbal absurdity task.

Cluster size	Region	Z	MNI coordinates			Label
			X	Y	Z
ASD
1202	LH Fusiform (BA 37)	5.93	−52	−40	−8
		5.35	−56	−28	−12	LH pMTG (BA 21)
		4.49	−52	−54	4	LH Angular gyrus (BA 39)
1039	LH IFG (BA 44)	5.01	−48	10	16
		4.35	−34	24	−8	LH IFG (BA 47)
768	LH Premotor + SMA (BA 6)	4.17	−4	2	54
		3.86	−6	44	44	LH dmPFC (BA 9)
		3.5	−6	20	56	LH FEF (BA 8)/SFG
NT
562	LH IFG (BA 44)	4.19	−50	8	6
		3.77	−42	30	−14	LH IFG (BA 47)
		2.9	−52	26	8	LH IFG (BA 45)
317	LH pMTG (BA 21)	3.69	−52	−38	−4
ASD > NT (uncorrected)
395	LH Fusiform (BA 37)	3.49	−52	−58	4
		3.36	−64	−40	−10	LH pMTG (BA 21
193	RH rmPFC (BA 10)	2.49	8	42	16
		2.35	8	52	24	RH dmPFC (BA 9)
63	LH Primary auditory (BA 41)	2.33	−50	−12	8
		1.87	−58	−14	−2	LH pSTG (BA 22)
		1.78	−64	−18	−2	LH pMTG (BA 21
53	LH Temporal pole (BA 38)	3.07	−56	4	−20
52	LH lateral dlPFC (BA 46)	2.56	−48	40	4
NT > ASD
ns

7.2. MS

The MS task within‐group analyses had the autistic participants showing increased activity in the following three clusters: (1) the left IFG, extending to the premotor + SMA; (2) the bilateral FEF; and (3) the fusiform, extending to the MTG (see Figure 2A and Table 3). The NT group showed increased activity in the left IFG and the premotor + SMA (see Figure 2B and Table 3). The direct comparison between the two groups did not show significant activity at a cluster corrected threshold (z = 2.3). Nevertheless, at an uncorrected threshold with a minimum of 50 contiguous voxels, the contrast ASD > NT showed increased activation in the following six clusters: (1) left fusiform, extending to the MTG, SMG, IFG pars triangularis, STG; (2) right FEF, extending dmPFC and mPFC; (3) right primary sensory, extending to the STG and the MTG; (4) the left STG, extending to the insula and primary auditory cortex; (5) left FEF; and (6) the ventral anterior cingulate cortex (ACC), extending to the dorsal ACC (see Figure 2C and Table 3).

FIGURE 2.

Brain areas exhibiting activation during multisentence task. Within‐group results for ASD (A) and NT (B) are cluster corrected at 2.3. The between‐group (C) ASD >NT is uncorrected at p < 0.001 with a minimum cluster size of 50 contiguous voxels.

TABLE 3.

MNI coordinates exhibiting activation in autistic and neurotypical children during the multisentence task.

Cluster size	Region	Z	MNI coordinates			Label
			X	Y	Z
ASD
1501	LH IFG (BA 44)	5.06	−52	16	12
		4.53	−52	22	4	LH IFG (BA 45)
		4.25	−46	10	28	LH Premotor + SMA (BA 6)
		4.15	−40	18	−4	LH Insula (BA 13)
1068	LH FEF (BA 8)	4.67	−6	22	56
		3.94	6	26	42	RH FEF (BA 8)
770	LH Fusiform (BA 37)	4.47	−52	−40	−8
		4.28	−58	−40	−8	LH pMTG (BA 21)
NT
812	LH IFG (BA 45)	4.89	−52	18	0
		4.51	−56	14	24	LH IFG (BA 44)
272	LH Premotor + SMA (BA 6)	4.73	−2	12	54
ASD > NT (uncorrected)
718	LH Fusiform (BA 37)	3.08	−56	−60	−2
		2.86	−66	−26	−2	LH pMTG (BA 21)
		2.84	−60	−30	20	LH SMG (BA 40)
		2.79	−54	−28	8	LH IFG (BA 45)
		2.66	−62	−18	2	LH pSTG (BA 22)
405	RH FEF (BA 8)	3.19	8	36	40
		3.15	10	48	22	RH dmPFC (BA 9)
		2.68	12	60	22	RH rmPFC (BA 10)
286	RH Primary sensory (BA 1)	2.85	48	−12	10
		2.4	52	−16	−4	RH pSTG (BA 22)
		1.98	60	−20	−14	RH pMTG (BA 21)
223	LH pSTG (BA 22)	2.77	−48	2	−10
		2.73	−44	−8	4	LH Insula (BA 13)
		2.44	−46	−14	6	LH Primary auditory (BA 41)
96	LH FEF (BA 8)	2.52	−2	34	52
59	LH ventral AC (BA 24)	2.34	−4	4	34
		1.82	−6	14	36	LH dorsal ACC (BA 32)
NT > ASD
n.s.

8. Percent of BOLD Signal Change

The results showed that autistic children, compared to NT children, had significantly higher percent BOLD signal change in the PN during both the VA (t(297) = 2.14, p < 0.05; see Figure 3A) and MS tasks (t(302.23) = 3.35, p < 0.0001; see Figure 3B).

FIGURE 3.

Differences in percent BOLD signal change between autistic and neurotypical children during verbal absurdity (A) and multisentence (B) tasks. The plots show the density curves, and the box plots show the mean (red circle and thick line), interquartile range (rectangle), and the lower/upper adjacent values (black lines stretched from the rectangle), and scatter plot. The significant differences between statement categories are reported with the p‐value.

9. Relationship Between Reading Comprehension and % of BOLD Signal Change in the PN

In models including ASD and NT participants combined, the % of BOLD signal change in the PN during the VA (F(2,31) = 3.8, p > 0.05, adjusted‐R ² = 0.14) and MS (F(2,31) = 4.40, p < 0.01, adjusted‐R ² = 0.17) tasks was predicted by SRS scores (for VA: β = 0.24, t = 2.14, p < 0.05; MS: β = 0.27, t = 2.23, p < 0.05) and the SORT‐R word decoding abilities (for VA: β = 0.27, t = 2.29, p < 0.05; MS: β = 0.29, t = 2.30, p < 0.05). However, these results did not survive Bonferroni correction (see Figure 2A,B). The model with autistic children only showed that the % of BOLD signal change during MS (F(1,21) = 16.97, p > 0.01, adjusted‐R ² = 0.42) was predicted by SORT‐R word decoding (β = 0.54, t = 4.12, p < 0.01, see Figure 4). The % of BOLD signal change during VA was not statistically significant for autistic and NT children, nor was it significant for MS in NT children. Finally, reading comprehension, as measured by the GORT‐4, did not emerge as a significant predictor in these models.

FIGURE 4.

Relationship between reading comprehension with % of bold signal change in the Pragmatic Network during multisentence task in autistic children.

10. Discussion

This study examined differential recruitment of the brain's PN during reading tasks in ASD and NT children. The GLM within‐group analysis found increased activation of the PN in the left hemisphere, specifically in the IFG, posterior MTG, and premotor + SMA, in both groups. The ASD group also recruited additional PN areas in the left hemisphere, such as the FG, FEF, pSTG, AG, insula, SMG, ACC, and mPFC. Moreover, the between‐group analysis revealed significant differences when the uncorrected threshold was used, which showed that the autistic group recruited additional areas from the PN in both hemispheres, including the mPFC, left MTG, and FFG. It is noteworthy that even with an uncorrected threshold, the NT group did not show an increased activation in comparison to the autistic group. In addition, the % of signal change showed a positive relationship with word decoding in the ASD group for the MS task.

During both reading tasks, the autistic and NT groups showed frontotemporal activation of areas associated with the PN and reading comprehension (i.e., FG, SMA) (Koyama et al. 2011; Reyes‐Aguilar et al. ²⁰¹⁸). These areas, in conjunction with the pMTG, have also been found to be associated with language processing (Friederici 2011). Evidence suggests that the IFG, the premotor + SMA, and the pMTG are heavily involved in semantic processing (Bašnáková et al. 2014; Chee et al. ¹⁹⁹⁹; Moore‐Parks et al. ²⁰¹⁰; Reyes‐Aguilar et al. ²⁰¹⁸; Ulrich et al. ²⁰¹³; Yang et al. ²⁰⁰⁹). The SMA has also been shown to have increased activity when the load of semantic processing increases (Yang et al. 2009). Conversely, the IFG and premotor + SMA have been shown to be involved during automatic semantic processing (Ulrich et al. 2013; Yang et al. ²⁰⁰⁹), making semantic judgments (Bašnáková et al. 2014; Chee et al. ¹⁹⁹⁹), and selecting meaning (Moore‐Parks et al. ²⁰¹⁰; Reyes‐Aguilar et al. ²⁰¹⁸). The pMTG has been implicated in controlled semantic retrieval and has also been shown to be functionally correlated with regions, including the IFG, implicated in control‐demanding semantic tasks (Davey et al. 2016). Consistent with previous evidence, our results suggest that these areas are actively engaged when the language processing load or demand is increased, perhaps suggesting the specific sensitivity of these regions to linguistic judgments in autistic and NT children.

The autistic group also showed increased activation in additional areas of the PN, such as the FG, FEF, AG, pSTG, ACC, and insula (Duvall et al. 2023; Kotila et al. ²⁰²⁰; Reyes‐Aguilar et al. ²⁰¹⁸). The pSTG and FG have been associated with reading comprehension (Koyama et al. 2011), and the AG and pSTG with language processing (Friederici 2011). While the FEF is involved in eye‐selective activity, recent research suggests its role in executive processes, especially during reading. An increased activation in the FEF was associated with reduced reading ability, suggesting that poorer readers may rely more on executive processes to achieve adequate comprehension rather than reading‐specific processes (Wang et al. 2019). Given the cognitive demands of the reading tasks, it is possible that the ASD group may have had to work harder and engage more executive processes to complete the tasks than the NT group.

Interestingly, both groups showed increased activation in areas primarily associated with integration (i.e., pMTG) (Davey et al. 2016; Jung‐Beeman ²⁰⁰⁵; Petrides ²⁰²³). Nevertheless, autistic children showed recruitment of additional areas (i.e., the pSTG, AG, SMG, and ACC). The SMG has been found to be associated with activating and integrating semantic knowledge (Chow et al. 2014; Frith and Frith ²⁰⁰⁶; Jung‐Beeman ²⁰⁰⁵) during language comprehension (Binder et al. 2009). On the other hand, the pSTG and the ACC have been associated with processes related to social communication, such as the integration of information with the context (Egorova et al. 2016; Lavin et al. ²⁰¹³). Similarly, the AG has been shown to have an integrative role in comprehension and reasoning while manipulating conceptual knowledge (Seghier 2013). The AG and SMG have also been associated with episodic memory retrieval (Vincent et al. 2006) and the integration of knowledge (Benedek et al. 2014). Finally, the pMTG has been proposed to allow semantic retrieval to be “shaped” to suit a task or context (Davey et al. 2016). The tasks in the current study involved integrating semantic meaning across different sentences. In line with previous evidence, the results suggest these regions play a role in integration. The recruitment of additional areas in autistic children may suggest that they may have had to work harder and engage more resources to integrate the linguistic information into comprehending the meaning of the whole message during both tasks.

Only the ASD group showed increased activity in the mPFC (dorsal and rostral), AG, and pSTG. These areas have been associated with pragmatic language processing (Bohrn et al. 2012; Reyes‐Aguilar et al. ²⁰¹⁸), ToM (Amodio and Frith 2006; Gallagher et al. ²⁰⁰⁰; Kana et al. ^{2009, 2015}; Mar et al. ²⁰⁰⁸; Schurz et al. ²⁰¹⁴), and overall social cognition (Tranel et al. 2002). In addition, these areas have been proposed to work in conjunction with the ACC as part of the social cognition network, which supports the comprehension of pragmatic language (Reyes‐Aguilar et al. ²⁰¹⁸). Interestingly, the ACC and the insula are considered part of the salience network (SN), which has been linked to attributing salience to the perceived external events (Seeley et al. 2007), and consequently may have a role in pragmatic understanding (Kotila et al. 2020). In addition, SN has been associated with switching abilities, error detection (Twait et al. 2018), and with ToM (Kim et al. 2016). Behavioral studies have shown positive relationships between reading and ToM abilities (Kidd and Castano 2013; Mar et al. ²⁰⁰⁶). Altogether, these results suggest that reading tasks that require a greater demand to understand the message elicit activation in areas associated with PN and ToM. In autistic children, given some difficulties in these skills, they may need to recruit these additional areas to help complete these tasks.

Furthermore, the increased activation in areas related to the PN in autistic children, compared to NT children, was also seen in the comparison. These findings may suggest that autistic children may utilize compensatory mechanisms to read and comprehend. Previous studies have highlighted such mechanisms in autistic individuals. For instance, Xu et al. (2022) found that increased connectivity between the amygdala and the mirror neuron system may functionally compensate for amygdala dysfunction and social deficits. Similarly, Hogeveen et al. (2020) observed increased recruitment of the hippocampus in autistic individuals, which may compensate for reduced connectivity between the medial temporal lobes and the posterior medial network, helping to preserve episodic memory. However, it is important to note that these results were obtained using an uncorrected threshold, and therefore should be interpreted with caution. Further studies could explore this relationship in more depth. In addition, results of the linear models show a positive relationship between word decoding and the PN during the MS task, but not during the VA task. The MS task has been shown to be more challenging than the VA task, as there are more sentences and more information that the reader must consider before making a choice (Murdaugh et al. 2017). Given that this relationship was only found during the more challenging task, our findings support the idea that autistic children may use an extension of the PN during reading comprehension, especially as the computational demand increases. On the other hand, in the NT group there was no significant relationship between decoding abilities and in the GLM when compared to ASD, suggesting that both tasks may have been relatively easier and therefore did not require increased effort. Taken together, results of the current study are consistent with previous findings demonstrating a relationship between reading and processes involved during social interactions, including pragmatics and ToM (Hartung and Willems 2020; Jacobs and Richdale ²⁰¹³; Kidd and Castano ²⁰¹³; Knight ²⁰¹⁶).

The findings of this study may provide insight into the involvement of PN during reading tasks in autistic and NT children. When compared to the NT group, the ASD group demonstrated increased activation in the PN during the VA and MS tasks, more so in MS. As previously stated, these results were obtained at an uncorrected threshold and require further investigation. This suggests that autistic children with poorer reading comprehension skills recruited more neural resources (additional areas of the PN) to accomplish specific reading tasks. Finally, there was a relationship between the PN and reading, such that the increase in PN activation during the MS task was associated with word decoding abilities. These findings contribute to the existing body of literature that demonstrates the relationship between reading and elements involved in social interactions, including ToM and language (Hartung and Willems 2020; Kidd and Castano ²⁰¹³; Mar et al. ²⁰⁰⁶). These findings offer insights into potential avenues for pragmatic skills intervention, such as targeting pragmatic language deficits through reading. For instance, autistic individuals or those experiencing pragmatic difficulties may address these issues through targeted reading comprehension interventions or through engagement in everyday activities such as participation in book clubs or frequent reading of books, particularly those of a fictional content. Moreover, the enhancement of pragmatic skills can be achieved by directing the focus toward brain regions implicated in PN, which have been shown to exhibit elevated levels of activation as a compensatory mechanism. This approach is predicated on the premise that it will yield therapeutic benefits for autistic individuals.

11. Limitations

Further studies could investigate the direct impact of reading on pragmatics in autistic populations. This could be achieved by employing formal reading comprehension interventions or incorporating reading into everyday activities, such as book clubs or other informal reading activities. The use of fiction books may be particularly beneficial in this context. For further studies it would be beneficial to conduct research using different contexts and a wider range of text types. For instance, behavioral studies have indicated that reading expository text is associated with growth in reading comprehension, which is predicted by executive functions (Wu et al. 2020), and narrative text enhances pragmatic abilities (Kidd and Castano 2013). In addition, given the ASD group had a different reading profile (e.g., discrepant poor comprehender reading profile with average or above average word reading and poor comprehension) than the NT group, it is possible these results may be due to the differing reading profiles rather than their diagnostic group. Future studies can address this limitation by adding an additional group of NT children with the discrepant reading profile, which our lab plans to do. Another limitation of this study is that a direct measure of pragmatic language was not used. Additionally, no cognitive measures associated with pragmatic language were used. For instance, it has been reported that pragmatic language comprehension is associated with executive functions (Caillies et al. 2014; Pexman ²⁰⁰⁸), social cognition (Caillies et al. 2014; Spotorno et al. ²⁰¹²), and processing style (Booth and Happé 2010; Martin and McDonald ²⁰⁰³). Further studies could assess a direct measure of pragmatic abilities and cognitive resources, which would enhance our understanding of the role of the PN during reading and the differences between autistic and NT children. Regarding the tasks used for this study, although explicit estimates of reliability and validity for the tasks used in this study are not available, future research will seek to validate these tasks through additional psychometric testing. We recommend further evaluation of their reliability and validity in diverse populations to enhance the generalizability of the findings and support the robustness of the task measures. Another limitation of the study is the inclusion of all task stimuli, regardless of whether participants responded correctly or incorrectly to specific stimulus items, in the analysis. This approach was adopted to ensure that potentially challenging stimuli, which may engage critical cognitive processes, were not excluded from the analysis. In addition, it is possible that the process that goes into thinking about a given response (whether correct or incorrect) perhaps elicits similar pattern of brain activity. Yet another reason for adopting a combined approach was that splitting the responses would have caused power differences and made the comparison between correct and incorrect responses skewed. Nevertheless, future research could address this limitation by systematically analyzing the neural mechanisms underlying reading in separate conditions for correct and incorrect responses or examining their combined effects. Such an approach could provide deeper insights into how accuracy influences cognitive and neural processes during reading.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1. Overlapping of the Pragmatic Network (red) with Reding Network (blue) and theory of mind (yellow). The figure illustrates a significant degree of overlap between the three networks in the left hemisphere, specifically within the inferior frontal gyrus and the posterior superior temporal gyrus.

Figure S2. Differences by group for (A) Gray Oral Reading Test, (B), Slosson Oral Reading Test, and (C) Social Responsiveness Scale.

Figure S3. Relationship between SRS and SORT‐R with % of BOLD signal change in the Pragmatic Network in autistic and neurotypical group. (A) showed the relationship of the Verbal Absurdity task with (A.1) SRS and (A.2) SORT‐R; (B) of the Multisentence with (B.1) SRS and (B.2) SORT‐R in autistic children. BOLD = blood oxygen level dependent; SORT‐R = Slosson Oral Reading Test‐Revised; SRS = Social Responsiveness Scale.

Table S1. The MNI coordinates associated with Pragmatic language network and results from normality test.

Valles‐Capetillo, E. , Kurtz M. R., and Kana R. K.. 2025. “The Role of the Brain's Pragmatic Language Network in Reading Comprehension in Autistic Children.” Autism Research 18, no. 8: 1550–1562. 10.1002/aur.70076.

Data Availability Statement

The data that support the findings of this study are not publicly available now but because we have several ongoing data analyses from this dataset. However, are available from the corresponding author on reasonable request.