Behavioral and neural interrelationships of sensory difficulties, intolerance of uncertainty, and anxiety in autistic children

Abstract

Purpose:

Autistic children often experience higher levels of sensory difficulties and anxiety, compared to their non-autistic counterparts. Available evidence suggests that intolerance of uncertainty (IU) mediates the relationship between these factors. We aimed to further the examination of the interrelationship between these phenomena, as well as their underlying neural mechanisms, in children.

Methods:

30 autistic children (ages 6-11 years) and 26 age-matched neurotypical (NT) peers participated in resting-state fMRI, as well as various behavioral assessments of sensory processing, anxiety, and IU.

Results:

Between-groups comparisons of behavioral scores showed higher levels of sensory processing differences, anxiety, and IU in autistic children. Mediation analysis also revealed that IU was a significant mediator between sensory difficulties and anxiety in this group. Network connectivity findings showed that connectivity between sensory cortices and cerebellar, higher order sensory, and limbic regions (esp. amygdala) were significantly correlated with sensory processing, anxiety, and IU.

Conclusion:

We report novel behavioral and neurophysiologic findings concerning important interrelationships in autism. Our findings may be related to predictive coding accounts of autism. Our results have the potential to inform future clinical practice, in that addressing sensory and IU difficulties has the potential to lead to improvements in anxiety in autistic children.

Introduction

Ninety percent, or more, of autistic children are known to experience sensory differences (American Psychiatric Association [APA], 2013; Green & Hernandez, 2015; Marco et al., 2011; Rajendran & Mitchell, 2007). Sensory difficulties can affect any sensory modality and can present as hyper- or hypo-sensitivities and/or sensory seeking behaviors in autistic individuals (Ben-Sasson et al., 2007; Crane et al., 2009; Dunn, 1997; Green et al., 2015; Jassim et al., 2021; Marco et al., 2011; Ozsivadijan, Knott, & Magiati, 2012; Posar & Visconti, 2018; Tomcheck & Dunn, 2007). These variable responsivity patterns can occur simultaneously in the same individuals in reaction to different sensory stimuli (Balasco, Provenzano, & Bozzi., 2020). Such difficulties with sensory input can be significant and have a dramatic impact on an autistic person’s quality of life, often affecting an individual’s level of independence, social relationships, self-esteem, and participation, even in fundamental or preferred activities (van Heijst & Geurts, 2015). In some, these sensory processing differences can even lead to selfdirected aggression or other harmful behaviors (Ozsivadijan et al., 2012).

One of the most common mental health difficulties for autistic people is anxiety, with clinical anxiety levels present in approximately 40% of autistic youth, more than double that in neurotypical youth (van Steensel & Heeman, 2017). Additionally, symptoms of anxiety appear to be significantly more severe in autistic youth compared to those with only a clinical anxiety diagnosis and may have various qualitative differences when compared to non-autistic individuals (APA, 2013; Kristensen et al., 2014; van Steensel & Heeman, 2017). Autistic children also often describe their anxiety by citing highly specific fears and phobias, changes in their regular environment or schedule, loud noises and other sensory differences, meltdowns, less familiar social situations, and/or difficulties related to lack of rigid structure (Kerns et al., 2014; Ozsivadijan et al., 2012).

Behavioral connections between sensory processing and anxiety have been shown numerous times in the literature in both adults and children (Green et al., 2010; 2013; 2015; Neil et al., 2016; MacLennan et al., 2021; Normansell-Mossa et al., 2021 South & Rogers, 2017; Wigham et al., 2015). While the full characterization of this relationship is still elusive, one factor that appears to play an important role in it is intolerance of uncertainty (IU; see South and Rodgers, 2017 for a review). IU was originally researched as an explanation for anxiety symptoms in neurotypical individuals (Dugas et al., 1998). It refers to the “tendency to react negatively on an emotional, cognitive, and behavioral level to uncertain situations and events” (Boulter et al., 2014). People who show increased IU may have difficulty adjusting to changes in routine and may interpret ambiguous information as threatening, which can significantly contribute to anxiety (Einstein, 2014; McEvoy & Mahoney, 2012).

The available literature shows a fundamental connection between sensory processing, IU, and anxiety. However, the particulars of this three-way association varies between studies, which suggests a complex interrelationship. Most commonly, IU has proven to mediate the relationship between sensory processing and anxiety (MacLennan et al., 2021; Neil et al., 2016; Normansell-Mossa et al., 2021; Wigham et al., 2015). This point was supported recently in a study from our group, which showed that increased uncertainty brought about by the COVID-19 pandemic was linked to concomitant sensory difficulties and stress in autistic children (Cardon & Bradley, 2023).

Possible Neural Underpinnings of the Relationship between Sensory Processing, IU, and Anxiety

Another aspect of the above interrelationships that remains less explored is their neural mechanisms, though evidence from the literature may shed light on this topic. For instance, one prevailing theory of atypical sensory processing involves aberrant neural connectivity within and between sensory cortices (e.g., Marco et al., 2011). While EEG, MEG, MRI, and fMRI investigations, have borne out this idea, consensus is still yet to be established (e.g., Bruneau et al., 2003; Cardon, Hepburn, & Rojas, 2017; Cardon, Buckhannon, & Rojas, 2023; Del Casale et al., 2022; Gage et al., 2003; Hull et al., 2017; Monk et al., 2009; Pua et al., 2017; 2020; Oram Cardy et al., 2008; Villalobos et al., 2005; Wang et al., 2018). Other theories advocate the idea of imbalance of excitation and inhibition, likely mediated via the relative levels of glutamate and GABA throughout the brain in autistic individuals (Rubenstein & Merzenich, 2003). Furthermore, some have investigated the contribution of peripheral sensory receptors to sensory processing atypicalities in this population (REFS). Similar to functional connectivity accounts of sensory processing in autism, mixed results can be found throughout the literature concerning these theories. Ultimately, it may be that these factors combine in some way to lead to sensory differences. In the present study, we focus on functional connectivity.

In addition to sensory-specific cortices, supra-modal brain networks/regions have been implicated in sensory processing among autistics (i.e., brain regions/networks that are not tied to processing in one sensory modality, or another, but, which engage in multisensory or higher-order processing; Wang et al., 2018; Wang et al., 2021). Given the variability in the manifestation of sensory dysfunction, focusing on brain areas that connect to and have the ability modulate responsivity in all sensory systems is warranted. For instance, converging data from structural, functional, post-mortem, and animal studies suggest that disruption in the cerebellum from cellular to more macro levels contribute to the pathophysiology of autism (e.g., Anteraper et al., 2019; Cardon, 2018; Courchesne, 1991; 1997; Courchesne et al., 2001; Just et al., 2012; Kemper & Bauman, 1998; Overwalle et al., 2020; Zhao et al., 2018). The cerebellum forms reciprocal connections with all sensory systems and has a modulatory relationship with the cerebral cortex (Cakar et al., 20204; D’Mello & Stoodley, 2015; Kern, 2002; Mosconi et al., 2015). Furthermore, the cerebellum is a center of multisensory integration—a function regularly implicated in autism (Erickson et al., 2014; Molinari et al., 2009; Ronconi et al., 2016). Thus, connectivity between the cerebellum and sensory cortices (esp. higher order) is important for adaptive sensory processing. As such, decreased connectivity between these brain regions could be involved in autistic sensory difficulties, though evidence of this effect remains to be established.

Other supramodal regions may also be involved in sensory difficulties. For instance, the amygdala has been implicated in emotional aspects of sensory processing (Zald, 2003) and has also been heavily researched in association with sensory hypersensitivity and anxiety in autism (Kleinhans et al., 2009; Green et al., 2015; Herrington et al., 2016). For instance, Green et al. (2015) showed that increased neural activity in a network comprised of the amygdala and sensory cortices in autistic subjects was significantly associated with sensory overresponsivity. However, activation of the amygdala and its connection to behavior in autism is still being fully examined (Herrington et al., 2016; Kleinhans et al., 2009; Green et al., 2010; 2013; 2015).

South and Rodgers (2017) raised the likelihood of the importance of the integrative functions of areas such as the medial prefrontal cortex (mPFC), amygdala, and anterior cingulate cortex (ACC) in processing, development, and control of emotional aspects of sensory inputs, prediction, and anxiety. That is, it is thought that the MPFC (including ACC) controls emotional responses through strong connections to the amygdala (among other regions; Giustino & Maren, 2015), which, in turn, is a supramodal hub for sensory processing. Atypical function (esp. decreased connectivity) of this network could result in emotional regulation difficulties, which could impact sensory function (e.g., Shalom, 2009; Lecouvey et al., 2015; Smith et al., 2015; Tzur, 2015). Couto-Ovejero et al. (2024) recently also showed evidence of cerebellar involvement in processing of fear/threatening stimuli in autism, potentially linking the aforementioned points about cerebellar involvement in sensory processing and notions regarding emotional control.

Interestingly, recent reports have raised the possibility of the above brain regions forming a distributed “prediction network”, consisting of sensory areas, attention networks, the default mode network, cerebellum, and limbic regions (Bubic et al., 2010; Siman-Tov et al., 2019). If true, such function might implicate top-down processing of sensory information from multiple sensory modalities in both IU and anxiety. On the other hand, atypical bottom-up sensory processing (i.e., primarily afferent processing of sensory inputs) could be associated with maladaptive downstream function. Taken together, the above notions suggest that connectivity between and integration of signals from a distributed network of both sensory and supramodal brain regions is likely to underpin the complex interrelationship between sensory processing, IU, and anxiety.

Aims and Hypotheses

The present study aimed to examine the behavioral interrelationships between sensory processing, IU, and anxiety in school aged children and their neural underpinnings. We hypothesized that intolerance of uncertainty would be a mediating factor between behavioral sensory processing scores and measures of anxiety. Based on the above, we also hypothesized that decreased connectivity between sensory cortices and cerebellum and supramodal resting state networks (e.g., default mode network; DMN), and increased connectivity between these areas and the amygdala, would be correlated with behavioral sensory-, IU-, and anxiety-related behavior in autistic children.

Methods

Participants

30 school-aged children with a clinical diagnosis of Autism Spectrum Disorder (27 males, 3 females; mean age 9.16 years, SD = 1.79) and 26 neurotypical (NT) peers (19 males, 7 females; 9.44 years, SD = 1.55) participated in the current study (2017-2018). Autistic children were diagnosed according to the Autism Diagnostic Observation Scale-2 (ADOS-2) and criteria from the DSM-IV and V by clinicians in the greater Denver, CO area. All participants effectively used verbal language as evidenced by the observations of the research team, as well as testing with the Children’s Communication Checklist-2^nd Edition (CCC-2; Bishop, 2003)—i.e., autistic children had a mean General Composite Communication Score of 79.85 (SD=15.13; min=59.00; max=128.00), which is within normal limits, though lower than the NT group (m=108.00; SD=11.89). Additionally, 28/30 (93.3%) autistic and 24/26 (92.3%) NT participants were reported to be White and not Hispanic, while the remainder of participants for both groups reported that they were of Hispanic or Latino ethnicity. Participants were mainly recruited via previous participation in research studies at the University of Colorado Anschutz Medical Campus and via word of mouth and outreach to local childcare/educational agencies. All recruitment, consent, and testing procedures were approved by The Colorado Institutional Review Board (COMIRB) and in accordance with the Declaration of Helsinki.

Instrumentation

We implemented the following behavioral measures: the Short Sensory Profile (SSP; McIntosh, Miller, & Shyu, 1997), Intolerance of Uncertainty (IUS-12; Carleton et al., 2007), and the Screen for Child Related Anxiety Disorders (SCARED; Birmaher et al., 1999). The SSP is a 38-item caregiver questionnaire that is meant to assess areas of sensory processing, modulation, and behavioral and emotional responses. Though the SSP was intended for use in children ages 3–10, more recent, high-quality studies have successfully used this measure in children between the ages of 2-12 years (e.g., Williams et al., 2018; Hand et al., 2017; and Ausderau et al., 2014). Herein, we used the Total score of the SSP to measure general sensory difficulty—perhaps especially hypersensitivity—in our participants, even though some have questioned the similarity of the factor structure and subtests of the SSP when used in NT vs. autistic subjects (Williams et al., 2018). Each question of the SSP utilizes a Likert Scale of 1 (always) – 5 (never) to assess if a child’s sensory experience is “different” or typical, in which lower scores are associated with greater sensory differences, and vice versa. The SSP removes 27 items from the original 117 item Sensory Profile (SP) that are less related to sensory modulation and 60 others that proved to be less effective in discriminating the NT population from those with sensory differences (Williams et al., 2018). The SSP was chosen for our study due to its widespread use in assessing autistic children (e.g., Burns et al., 2017; Crasta et al., 2021; Glod et al., 2019; Simpson et al., 2019), including large-scale studies (Lajonchere et al., 2012; Charman et al., 2017), those in which it has been used to test the convergent validity of other sensory measures used in autism studies (Neil, Green, & Pellicano, 2017; Siper, et al., 2017; Tavassoli et al., 2016), neuroimaging studies (e.g., Hegarty et al., 2018; Lefebvre et al., 2022; Orekhova et al., 2012; ), and those focusing on phenotypic associations germane to the present study (Neil, Olsson, & Pellicano, 2016; Wigham, Rodgers et al., 2015).

The IUS-12 (Carleton et al., 2007) measures tolerance to different types of unpredictable scenarios, such as the future, ambiguity, and uncertainty as a whole. It consists of 12 items, taken from the 27-item original Intolerance of Uncertainty Scale. It employs a 5-point Likert scale for each item. The IUS-12 has excellent internal consistency in autistic subjects (a = 0.90; Boulter et al., 2014) and has been used extensively in this population (e.g., Neil, Olsson, & Pellicano, 2016; Wigham, Rodgers et al., 2015)

The Screen for Child Related Anxiety Disorders (SCARED) is a 41-item questionnaire that is completed by child (ages 8–18) or parent, rating common anxiety-inducing situations on a 3-point Likert scale (Birmaher et al., 1999). A recent large-scale study comparing the use of the SCARED in a younger (ages 5-8 years) to that in an older group (ages 9-12 years) showed virtually identical reliability between these groups (Sequeira et al., 2019). Stern et al., 2014 also used this instrument in autistic children between the ages of 7-14 years (see also van Steensel, Deutschman, & Bögels, 2012). Thus, our use of the SCARED in the children in the present study is in keeping with previously established practices. The questions require respondents to mark a 0, 1, or 2, with 0 indicating “not true or hardly ever true” and 2 indicating “very true or often true.” The SCARED has been shown to have generally high internal consistency (α=.91) as well as moderate sensitivity (0.71) and specificity (0.67; Hale et al., 2011). The total score of this instrument was chosen not just for its overall validity in identifying anxiety disorders, but also for its accuracy in identifying these disorders in autistic children (e.g., Stern et al., 2014).

Procedures

All data were collected at the Brain Imaging Center at the University of Colorado Anschutz Medical Center. In all, research appointments lasted two hours. Upon arrival, participants completed an MRI safety screening questionnaire, were briefed on the procedures, and were asked to change into hospital scrubs. During the remainder of the first hour, participants were allowed to tour the MRI facility and scanner, ask any questions they had, and underwent informed consent (parents) and assent (children) procedures. MRI scanning took place during the second hour of the appointments. Participants were allowed to watch a movie of their choice during their scans. They also were given a weighted blanket, pillows for their legs, and goggles and headphones for their comfort.

A 3T Siemens Skyra MR scanner was used to collect structural and functional fMRI data. All participants underwent a T1 weighted anatomical MRI (MP-RAGE) and a resting-state fMRI scan (rs-fMRI; 8 minutes): Whole-brain, blood oxygen level-dependent (BOLD) datasets consisted of 40 axial slices, 3.0 mm thick with a 0.5 mm gap, 220 mm² FOV (64 squared matrix = 3.0 mm²), repetition time = 2500 ms, echo time = 30 ms. During scanning, participants were asked to lie as still as possible despite any loud sounds the machine may make. Goggles and headphones aided in dampening the noise and allowing subjects to focus on the fixation cross during fMRI scanning. Subjects’ heads were secured with padding during scanning to reduce head motion. While the MRI scanning was performed, parents completed behavioral questionnaires.

Data Analysis

First, differences in behavioral scores between autistic and NT children on the SSP, IUS-12, and SCARED were calculated using between groups Mann-Whitney U tests. Additionally, Spearman rank order correlations were employed to investigate the correlations between all pairings of behavioral measures. We employed non-parametric statistics because data were not distributed normally, using the SPSS statistical software package (version 28.0.0; IBM Corp., 2021).

Structural MRI and fMRI data were first imported to CONN toolbox (Whitfield-Gabrieli et al., 2012) within MATLAB 2020b (MathWorks) to perform both preprocessing and functional connectivity analyses. After pre-processing, consisting of motion correction, temporal high pass filtering, spatial Gaussian smoothing (6 mm³), co-registration, and motion correction using the ArtRepair toolbox within the CONN toolbox (Whitfield-Gabrieli & Nieto-Castanon, 2012), data were normalized to MNI space. Then, we performed group independent components analysis (ICA) across the autistic and NT children using Conn (Calhoun, Adali, Pearlson, & Pekar, 2001). Forty independent components (IC) were initially extracted. Any IC’s that were determined to be comprised of noise (i.e., activity in voxels outside of grey matter, motion, etc) were immediately excluded from future analyses. Following this initial selection, 18 ICs containing areas of activation corresponding to our hypotheses—i.e., sensory, cerebellum, amygdala, and other supramodal resting networks, such as the DMN—were selected. To further characterize and confirm these selections, we subjected these 18 networks of interest to a spatial match to template within Conn, using the 10 common ICs reported in Smith et al. (2009) and the 14 resting state functional networks reported in Shirer et al (2012). This step accomplished two goals: 1) to discriminate between true networks of interest and those containing spurious data; and 2) to assign a functional name to each network. IC networks that did not match well (minimum r=0.01) with the aforementioned network templates were also excluded from analysis. The final number of ICs included in our final analysis was 15.

Then, a number of distinct functional connectivity analyses were carried out. First, differences in within network connectivity were tested between the autistic and NT children in each of the retained IC networks via independent samples T-tests within Conn. Second, the functional connectivity between each IC network and all other brain voxels that was associated with SSP, IUS-12, and SCARED total scores was assessed within each participant group in Conn, controlling for age and sex (cluster-based Gaussian random field theory parametric statistics, 0.001 threshold; Worsley et al., 1996). Two-sided false discovery rate (FDR) corrections (p<0.05) were applied to the cluster sizes of connected voxels for each IC network. Significance of between groups comparisons and connectivity patterns for each IC-voxel-behavioral measure combination was determined through final Bonferroni multiple comparisons corrections for each of the above tests (corrected p=0.0034). Connectivity values were extracted for all participants in each of the above comparisons. These values were then imported into SPSS to determine the strength of the relationship (Pearson’s partial correlation, controlling for age and sex) between functional connectivity indices and behavioral performance.

Results

Between Group Behavioral Comparisons

Between groups comparison of behavioral scores revealed that autistic and neurotypical participants differed significantly in their scores on all behavioral measures included in the current study, such that the former group presented with greater sensory processing difficulty (U = 774.50; p<0.001), IU (U = 92.50; p<0.001), and anxiety (U = 185.00; p = 0.001; see Table 1).

Table 1.

Participant characteristics and between groups comparisons of behavioral total scores

	Autism Group (n=30) Mean (SD)	NT Group (n=26) Mean (SD)	Autistic vs. NT Children
			U	p	Eta- squared
Age (years)	9.16 (1.79)	9.44 (1.55)	427.00	0.54	0.01
Sex assigned at birth (M/F)	27/3	19/7	-	-	-
SRS-2 Total	74.64 (9.5)	45.46 (5.68)	11.00	<0.001	0.78
SSP Total	119.70 (24.87)	170.42 (15.38	774.50	<0.001	0.60
IUS-12 Total	36.33 (11.93)	19.68 (6.52)	92.50	<0.001	0.42
SCARED Total	31.23 (20.77)	13.30 (8.28)	185.00	<0.001	0.24

Within Group Behavioral Results

Within the group of autistic children, several significant correlations were observed between behavioral scores. For instance, scores on the IUS-12 (r = −0.48; p = 0.007) and SCARED (r = −0.45, p = 0.013) showed significant negative relationships with the SSP, suggesting that increased sensory difficulties were related to heightened IU and anxiety. Furthermore, the IUS-12 showed a significant positive correlation with the SCARED (r = 0.52, p = 0.003; see Table 2).

Table 2.

Within group correlations of behavioral total scores

	Autism Group			NT Group
	SSP	IUS	SCARED	SSP	IUS	SCARED
SSP	-	−0.48(0.007)	−0.41(0.03)	-	−0.59(0.002)	−0.03(0.89)
IUS	-	-	0.52(0.003)	-	-	0.11(0.59)
SCARED	-	-	-	-	-	-

The NT group exhibited some similar patterns of correlation to those in the autistic children, though others differed. For instance, The SSP and IUS-12 showed a significant negative correlation (r = −0.59, p = 0.002). On the other hand, neither the SSP nor the IUS-12 were significantly correlated with scores on the SCARED in this group.

Results of our behavioral mediation analysis revealed that the relationship between sensory processing and anxiety was mediated by IU in autistic children (see Figure 1). That is, the standardized regression coefficient between the SSP and IUS-12 was significant (ß = −0.24; p = 0.006). A similar pattern was also seen between IUS-12 and scores on the SCARED (ß = 0.84; p = 0.012). Though regression results for the autistic children showed a significant relationship between SSP and SCARED Total scores (r = 0.41; p = 0.03), the direct effect of SSP on SCARED in the mediation model was not significant (ß = −0.14; p = 0.36). Indirect effects of both models were tested using bootstrapping procedures (10,000 samples). The bootstrapped unstandardized indirect effect in the model was −0.198. Both values fell within 95% confidence intervals, suggesting that IU was a mediator of the relationship between sensory processing and anxiety. We observed no such relationships in NT individuals.

Figure 1:

Results of the mediation analyses of the relationship between sensory processing (SSP) and anxiety (SCARED), as mediated by intolerance of uncertainty (IUS-12)

Resting-state fMRI Results

Independent t-tests between groups revealed networks in which there were significant differences—the default mode network (IC7), a cerebellar/brainstem network (IC14). The former showed patterns of connectivity that differed between the groups centered in the left lateral occipital cortex / angular gyrus (peak: t(52) = −4.15; p = 0.00012), while the latter exhibited differences in the left temporal pole (peak: t(52) = −5.55; p = 0.000001) and (peak: t(52) = −4.43; p = 0.00005). In each of these differences, connectivity was significantly decreased in autistic children (see Table 3 and Figure 2). None of these connectivity patterns was correlated with SSP, IUS-12, or SCARED scores in either group.

Table 3.

Functional Connectivity in Independent Components

		Between Groups	Within Groups
IC	IC Area	Autistic > NT	Autistic			NT
			SSP	IUS-12	SCARED	SSP	IUS-12	SCARED
1	Higher Order Visual	--	L FPole (0.000)	--	--	--	--	--
2	Auditory / Sensorimotor	--	R/L Planum Temp (0.000)	--	--	--	--	--
3	Ventral DMN	--	L Ling (0.001)	--	--	--	--	--
4	Cerebellum	--	--	L LOC (0.000)	--	--	--	--
5	Temporal Pole	--	L FPole (0.000)	--	--	--	--	--
6	Primary Visual	--	--	--	--	--	R Sup Front G (0.007)	--
7	DMN	L Sup LOC / AG (0.0014)	--	L FPole (0.000)	--	--	--	--
8	Executive Control	--	--	--	--	--	--	--
9	Cerebellum (vermis)	--	R Central operculu m/Heschel’s (0.000); R/L Lingual / Temp Occip Fus (0.000); L Central operculu m / L PoCG (0.001)	--	--	--	--	--
10	Sensorimotor	--	--	--	L amygdala/BG (0.0002)	--	--	--
11	Dorsal Attention	--	--	Precuneus (0.003)	--	L FPole (0.000)	--	L Ang G (0.000)
12	L Fronto-Parietal	--	--	--	--	--	--	--
13	Dorsal Attention (visuospatial)	--	--	Intracalc/lingual/precuneus (0.000)	--	--	--	--
14	Cerebellum (vermis)/Brainstem	L Temp Pole / FPole (0.0002)	--	R/L Sup LOC / S Par Lob (0.000)	R/L Sup LOC / S Par Lob (0.000)	--	--	--
15	Sensorimotor	--	--	--	--	--	--	--

^*corrected p < 0.003 – all p-values describe FWE corrected significance for cluster size

Note. L/R = left/right; FPole = frontal pole; poCG = Post-central gyrus; Planum Temp = Planum Temporale; mPFC = medial prefrontal cortex; LOC = lateral occipital cortex; S Par Lob = superior parietal lobe; Ling = lingual gyrus; DMN = default mode network; SFG = superior frontal gyrus; Temp Occip Fus = temporo-occipital fusiform; SMG = supramarginal gyrus; AG = angular gyrus; DAN = dorsal attention network; intracalc = intracalcarine sulcus; BG = basal ganglia

Figure 2:

Brain regions showing a significant difference in IC network connectivity between autistic and NT children (ASD>NT). Panel A shows ICs demonstrating significant differences between ASD and NT participants, while panel B presents the patterns of connectivity associated with these ICs.

Analysis of the associations between network connectivity and behavioral scores showed several connections significantly associated with sensory processing (i.e., SSP score; Figure 3). Most germane to the current hypothesis, the network represented in IC9 (cerebellum vermis) exhibited positive connectivity with central opercular (primary somatosensory cortex), superior/transverse temporal (primary and association auditory cortices), lingual gyrus (primary and higher order visual cortices), and cerebellar hemisphere regions (peak: 42, −9, 10; t(26) = 6.49; p = 0.000)—i.e., lower degrees of connectivity between these regions were associated with greater sensory difficulty in autistic children (r=0.87; p=0.00). Additionally, an auditory/sensorimotor network (IC2) exhibited negative connectivity with bilateral planum temporale (peak: −53, −28, 10; t(26) = −5.15; p = 0.000023), such that higher connectivity was correlated with heightened sensory differences (r=−0.82; p=0.00). Furthermore, higher order visual (IC1) and temporal polar (IC5) independent component networks were each linked to the left frontal pole region (peak: −30, 60, −5; t(26) = −5.49; p = 0.000009 & −14,64,10; t(26) = −5.33; p = 0.000014, respectively)—i.e., increased sensory differences were related to greater connectivity (IC1: r=−0.75; p=0.00; IC5: r=−0.86; p=0.00). Finally, the ventral DMN (IC3) showed a positive pattern of connectivity with the left lingual gyrus (peak: −17, −59, −2; t(26) = 5.10; p = 0.000026) that was correlated with the SSP (r=0.73; p=0.00).

Figure 3:

Independent component networks (A), network connectivity associated with IC activity and SSP Total score (B), and scatter plots of the correlation between SSP Total score and these patterns of connectivity in autistic children.

We also observed neural connectivity linked to IU (Figure 4). For instance, IC4—cerebellar vermis—showed negative connectivity with left inferior lateral occipital cortices (i.e., higher order visual and object recognition; peak: −47,−76,−12; t(26) = −5.24; p = 0.000018). In this instance, lower connectivity was associated with elevated IU (r=−0.77; p=0.00). IC14—cerebellar vermis / brainstem—exhibited positive connectivity with superior lateral occipital/parietal lobule (i.e., higher order visual, dorsal attention, and multisensory integration) regions bilaterally that was associated with decreased IU in the autism group (peak: −22,−73,45; t(26) = 6.14; p = 0.000002; r=0.79; p=0.00). Also, a negative connectivity pattern between IC7 (DMN) and the left frontal pole suggested lower degrees of connectivity were related to intensified IU (peak: −30,52,−12; t(26) = −6.42; p = 0.000001; r=−0.82; p=0.00). Finally, two separate, but complimentary IC networks contributing to the dorsal attention network (DAN; IC’s 11 and 13) showed connectivity with the precuneus (peak: −6,−65,30 t(26) = 5.43; p = 0.000011) and intracalcarine / lingual cortices (peak: −13,−77,10; t(26) = −5.48; p = 0.00001), respectively. IC11 presented with positive connectivity (r=0.71; p=0.00), while IC13 exhibited the opposite pattern (r=−0.71; p=0.00).

Figure 4:

Independent component networks (A), network connectivity associated with IC activity and IUS-12 Total score (B), and scatter plots of the correlation between IUS-12 Total score and these patterns of connectivity in autistic children.

Connectivity between IC14 (cerebellum vermis/brainstem) and lateral occipital/inferior temporal regions (peak: −46,−63,−19; t(26) = 5.10; p = 0.000026) was similarly related to SCARED scores, with lower connectivity associated with lower anxiety (r=0.77; p=0.00). Lastly and notably, IC10 (sensorimotor)—left basal ganglia / amygdala connectivity (peak: −19,5,1; t(26) = −5.02; p = 0.000032) was also significantly correlated with anxiety (SCARED), such that lower degrees of connectivity among these brain areas was associated with higher anxiety (r=−0.78; p=0.00; see Figure 5).

Figure 5:

Independent component networks (A), network connectivity associated with IC activity and SCARED Total score (B), and scatter plots of the correlation between SCARED Total score and these patterns of connectivity in autistic children.

Discussion

This study examined the behavioral relationships between sensory processing, IU, and anxiety in autistic and NT individuals, and neurophysiologic underpinnings of the same. Overall, we found: i) those on the autism spectrum presented with elevated degrees of all measured behavioral characteristics, compared to their NT counterparts; ii) IU had a mediating relationship between sensory processing and anxiety in autistic children; iii) sensory processing, IU, and anxiety were underpinned by connectivity patterns between sensory cortices and other sensory and supramodal brain regions—esp. cerebellum, DMN, attention networks, and the amygdala—that, in several instances, showed intriguing overlaps, suggesting common mechanisms.

Overall increases in sensory processing differences, IU, and anxiety, in autistic compared to NT individuals, are in keeping with previous literature (e.g., sensory—Green et al., 2015; Marco et al., 2011; Rajendran & Mitchell, 2007; IU—Boulter et al, 2014; Chamberlain et al, 2013; Jensen et al., 2016; anxiety/fear—van Steensel & Heeman, 2017). In addition to absolute differences in the above factors, we observed differences in the relationships between these variables between autistic and NT individuals. For instance, neither sensory processing nor IU were correlated with anxiety in NT individuals, as they were in our autistic sample, perhaps due to lower/absent levels of anxiety in the former. In contrast, we also noticed similarities between the autistic and NT groups. Specifically, the association between sensory processing and IU was significant in both autistic and NT children, suggesting that some behavioral relationships common to autism may span diagnostic boundaries (see Cuthbert & Insel, 2010; Abu-Akel et al., 2019; Kim et al., 2019; Kotov et al., 2017; Mayer, 2016). Our findings suggest that there is a fundamental relationship between sensory processing and IU, regardless of diagnosis (see also Cardon, Buckhannon, & Rojas, 2023; Cardon & Bradley, 2023).

Our most notable behavioral finding was the mediating influence of IU on the relationship between sensory processing and anxiety in autistic children. This result supports the notion that this interrelationship exists in autistic children, as it does in adults (Boulter et al., 2014; South & Rodgers, 2017; Neil et al., 2016; Wigham et al., 2015; MacLennan et al., 2021; Normansell-Mossa et al., 2021; Cardon, McQuarrie, & Gabrielsen, 2023). Though rooted in the anxiety literature originally, IU may have ties to other constructs that are less explored. For instance, many advocate the predictive coding theory of autism (Friston & Kiebel, 2009). That is, autistic individuals may exhibit difficulties in formulating predictions about upcoming inputs or scenarios based on their past sensory experiences (Sinha et al., 2014; van Boxtel & Lu, 2013; Van de Cruys et al., 2014). Making such predictions is a fundamental cognitive function that allows for smooth responses to ever-changing environmental stimuli. When a mismatch occurs between a prediction and the actual sensory stimuli a person experiences, it is often unpleasant, startling, or even painful (e.g., Van de Cruys et al., 2014). As such, people for whom prediction is difficult often find uncertain scenarios threatening (Pellicano & Burr, 2012). Over time, if prediction mismatches occur frequently enough, IU could certainly be developed (Pellicano & Burr, 2012; van Boxtel & Lu, 2013; Van de Cruys et al., 2014; Normansell-Mossa et al., 2021). In turn, increased IU might be associated with elevated anxiety. Additionally, because predictive functions influence processing of sensory inputs in a top-down manner (e.g., through focusing of attention, modulation of sensory brain regions, connecting sensory information to previous knowledge, reward, and motor functions, and, overall, comparison of inputs to previously generated predictions to evaluate the accuracy of one’s brain’s generative model of the world; Van de Cruys et al., 2014; Bubic et al., 2010), predictive atypicalities could be associated with sensory difficulties (Van de Cruys et al., 2019). Therefore, predictive coding may offer a parsimonious theoretical framework that explains of the complex interrelationship between sensory processing, IU, and anxiety. Because IU seems to be linked to prediction (Cannon et al., 2021; Thillay et al., 2016), our mediatory finding may suggest that predictive coding accounts for some portion of the relationship between sensory processing and anxiety. As such, increasing predictability/reducing uncertainty (e.g., in home, clinic, educational settings) and/or helping individuals enhance their abilities to adaptively deal with uncertainty are viable targets for supports designed to decrease challenges associated with sensory processing and anxiety in autism (Cannon et al., 2021; Cox et al., 2020; Hallett et al., 2020). Future work that combines such findings with clinical expertise could be fruitful.

Sensory processing, IU, and anxiety also showed overlaps in their neural correlates. For instance, each of these constructs was correlated with connectivity between the cerebellum and sensory cortices (primary and higher order) in autistic participants. In one notable example, decreased connectivity between the cerebellum (IC9) and auditory, visual, and somatosensory cortices was associated with greater overall sensory difficulty in autistic children (Figure 3). Also, we found a similar pattern between cerebellum/vermis—inferior lateral occipital cortex connectivity and IU—i.e., decreased connectivity was correlated with elevated IU (Figure 4). Furthermore, increased connectivity between cerebellum/vermis and superior lateral occipital cortices was related to heightened IU (Figure 4). Finally, there was a positive relationship between cerebellum/vermis/brainstem—inferior lateral occipital/temporal regions and anxiety (Figure 5). Taken together, these converging findings underscore the notion, and support our hypothesis, that cerebellar—sensory cortical connectivity is important among the underlying neural mechanisms of sensory processing, IU, and anxiety.

Numerous previous reports, using an array of methodology, propose the cerebellum as an area of high interest in autism (e.g., Courchesne et al., 1991; Fatemi et al., 2002; Allen & Courchesne, 2003; Olivito et al., 2016). Among other functions, predictive coding may be subserved heavily by the cerebellum (Courchesne & Allen, 1997; Friston, 2002, 2005, 2010; Popa et al., 2019; Siman-Tov, 2019). The cerebellum plays a prominent role both in processing and integrating sensory information, as well as preparing the necessary systems (esp. motor) to make appropriate responses to sensory inputs (Courchesne & Allen, 1997). Within the framework of the predictive coding theory, on one hand, an individual’s sensory processing differences could influence prediction formation. Alternatively, or additionally, if the brain regions involved in making predictions presented with functional differences, predictions about environmental events might also be altered, potentially leading to hyper-/hypo-sensitivities, and/or other sensory differences. Finally, if the connectivity between sensory and predictive areas were atypical, predictive function and sensory processing could both be impacted (Ward, 2019; Sapey-Triomphe et al., 2019; Van de Cruys et al., 2019). Since we did not employ a direct measure of prediction, we cannot conclusively say whether this cerebellar—sensory cortex relationship is due to prediction. However, taken together, the findings concerning IU and the underlying connectivity patterns involving the cerebellum, a brain region known to be involved in predictive coding, may support the notion of sensory processing—predictive coding associations in autism. More recent research has also found the cerebellar—cortical connections to also be involved in cognition, social processing, and emotion and aggression (McDougle et al., 2016; Van Overwalle et al., 2014; Schmahmann & Caplan, 2006), including social differences related to predicting the actions and emotions of others in autism (Keysers, Silani, & Gazzola, 2024). Thus, there may be alternative ways to interpret the present data. At the very least, the present results should prompt future investigations into cerebellar—cortical connectivity in autism, perhaps using more direct indices of prediction and examining predictive coding at various time scales, and/or considering other possible cerebellar contributions to atypical sensory processing in autism.

We also observed associations between sensory and other supramodal cerebral regions that were related to both sensory processing and IU. The most common of these supramodal networks were the DMN and dorsal attention networks (DAN). For example, positive connectivity between the ventral DMN (IC3) and visual cortices (lingual gyrus) was correlated with sensory differences. The DMN (IC7) also showed connectivity with the left frontal pole that was associated with IU. IU was also linked with connectivity between the DAN and precuneus, a prominent node in the DMN. Thus, it seems that the behavioral relationship between sensory processing and IU was underpinned by complex connectivity between sensory, DMN, and attention networks (Benkarim et al., 2021). Several have implicated the DMN and attention networks in autism (e.g., Casale et al., 2022; Funakoshi et al, 2016; Huang et al., 2021; Keehn et al., 2013; Neufeld et al, 2017; Paul et al., 2021), including our own group (Cardon, Buckhannon, & Rojas, 2023). The precuneus, though involved in many functions, may be instrumental in switching between task-negative (DMN) to task-positive states (e.g., Cavanna & Trimble, 2006; Utevsky, Smith, & Huettel, 2014). For instance, Utevsky, Smith, and Huettel (2014) showed increased functional connectivity between the precuneus and the right frontal-parietal (cognitive control) network during task engagement and enhanced connectivity between the precuneus and the DMN during rest. These findings suggest that the precuneus contributes to both task-positive and task-negative states. Additionally, given the DAN’s role in effortful control of externally oriented attention, which is counter to the interoceptive/task-negative state of the brain (i.e., focus on internal thoughts and body signals) when DMN-dominated, fluid transition between these states would be important for typical function (perhaps esp. sensory processing) and is a cognitively demanding proposition. Also germane to the above, previous studies have shown that the confluence of DMN, attentional and sensory network processing to be highly related to prediction—i.e., developing models of temporal sequences of upcoming events to reduce the cognitive costs of switching by focusing attention on desired or salient stimuli, while ignoring other aspects of the environment (e.g., Kurtin et al., 2023). Thus, the DMN—attention—sensory connectivity related to both behavioral IU and sensory processing reported herein is likely connected to predictive brain function.

Another supramodal brain region implicated in the constructs of the current study is the amygdala. Amygdala hyperreactivity has previously been associated with both anxiety and sensory processing (Green et al., 2015). The amygdala also plays a significant role in processing emotional reactions to sensory stimuli, especially fear and anxiety, and is highly connected to cortical sensory regions (Belmonte et al., 2004; Zald, 2013). In the present study, a resting-state network comprised of somatosensory and primary motor cortical regions exhibited negative connectivity with the left amygdala and basal ganglia, which was significantly related to SCARED total scores in autistic children. This association suggests a close link between brain activity in sensory and limbic regions and the interrelationship of behaviors pertinent to the present study. Exploratory analysis of the connection (partial correlation controlling for age and sex) between this connectivity pattern and IUS-12 scores in autistic participants revealed a significant negative correlation (r = −0.61; p<0.001). Though this relationship was weaker than that shown for SCARED scores, it suggests that the sensory—amygdala connectivity may also be related to IU.

In addition to the prediction aspect of sensory processing, various reports have cited multisensory processing and sensorimotor integration as particularly difficult for many autistic individuals (e.g., Beckerson, May, & Kana, 2022; Marco et al., 2011). Our results lend evidence to these notions. For instance, a negative connectivity pattern between IC2 (auditory / sensorimotor) and bilateral planum temporale was significantly related to SSP scores, pointing to covariation of connectivity among a combination of sensory-related areas and sensory difficulties. Such a finding may suggest that differences in connectivity between sensory and sensory—motor brain areas subserve multisensory and sensorimotor difficulties in autism (reminiscent of Chen et al., 2021).

When one steps back, our data show that there are multiple processes and patterns of neural connectivity that underpin the interrelationships between sensory processing, IU, and anxiety. This notion is to be expected, given the complexity of these behaviors. However, there may be unifying principles that provide a framework for understanding the interrelationship of these phenomena. For instance, the combination of sensory cortical and supramodal networks in our results, might suggest a role for the nexus of bottom-up and top-down neural processing in the behaviors we examined (e.g., Coll et al, 2020; Król & Król, 2020; Seymour et al, 2019; see Van Eylen et al, 2013 for an opposing viewpoint). Many of the brain regions discussed herein are situated optimally to mediate bottom-up and top-down processing of sensory input—e.g., cerebellum, amygdala, attention—because they receive manifold sensory inputs and can modulate sensory brain regions. Prediction, attention, and their role in top-down influence of bottom-up sensory inputs, may fit into this framework, in that accurate predictions are intertwined with adaptive multisensory processing, memory formation/retrieval, internal model generation, experience-dependent model refinement and updating, transfer of this information to influence incoming stimulus processing, as well as action/reaction to these stimuli—processes that may be mediated by a “predictive network” in the brain (Siman-Tov et al., 2019). Such a network is likely comprised by many of the regions discussed in the current study (e.g., sensory, DMN, cerebellum, attention, amygdala) and the connectivity between them. Thus, connectivity between supramodal (top-down) and sensory brain regions (bottom-up), as well as connections between sensory areas themselves, may underlie the complex constellation of sensory differences in autism. Activity and connectivity between these regions also seem to contribute to IU and anxiety in this population. Just as these behaviors are interrelated, so may their neural correlates be.

Limitations and Future Directions

One of the main limitations of our study is its relatively small sample size, which may decrease the ability of these results to be generalized across the autism population, especially given its heterogeneity. Also, the autistic participants in the current study were all highly verbal and able to undergo fMRI scanning, which is not characteristic of a large portion of the autistic population as a whole. Additionally, participant groups were not matched according to IQ or gender. Perhaps future investigations will build on the present preliminary findings in larger samples in the interest of improving generalizability. Additionally, based on our hypotheses, we chose to focus primarily on the cerebellum, amygdala, hippocampus, MPFC, and sensory brain areas. While activity and connectivity in several of these regions of interest were significantly associated with sensory processing, IU, and anxiety, others were not. Furthermore, because these behaviors are highly complex and interrelated, it is almost guaranteed that other brain regions underlie them. Future studies should seek to both replicate and expand on the current neurophysiological results.

Conclusion

The purpose of the current study was to examine the behavioral relationship between sensory processing, IU, and anxiety, and brain activity related to these behavioral interactions in autistic children. We posited that IU would be a mediating factor between sensory processing and anxiety, and that connectivity among sensory cortices, medial prefrontal cortex, hippocampus, amygdala and cerebellum would be significantly associated with the interrelationship between these behavioral constructs. Consistent with our hypotheses, IU played a mediating role between sensory differences and anxiety in autistic, but not NT, children. Furthermore, connectivity patterns in autistic children between sensory cortices, the amygdala, and the cerebellum, were significantly correlated with behavioral measures. Our results are consistent with recent literature that points to the notion that prediction difficulties may be central to sensory processing, as well as related to IU and anxiety. These data have the potential to guide future supports in various therapeutic areas.

HIGHLIGHTS.

• Autistic children showed greater behavioral sensory difficulties, anxiety, and IU

• IU mediated the relationship between sensory difficulties and anxiety

• Sensory difficulty, IU, and anxiety were linked to cerebellum—sensory connectivity

• Brain-behavior connections also involved frontal, DMN, and attention networks