Functional brain abnormalities associated with comorbid anxiety in autism spectrum disorder

Abstract

Anxiety disorders are common in autism spectrum disorder (ASD) and associated with social-communication and repetitive behavior symptoms. The neurobiology of anxiety in ASD is unknown, but amygdala dysfunction has been implicated in both ASD and anxiety disorders. Using resting-state fMRI, we compared amygdala-prefrontal and amygdala-striatal connections across three demographically matched groups studied in the Autism Brain Imaging Data Exchange (ABIDE): ASD with a comorbid anxiety disorder (N = 25; ASD+Anxiety), ASD without a comorbid disorder (N = 68; ASD−NoAnx), and typically developing controls (N = 139; TD). Relative to ASD−NoAnx and TD controls, ASD+Anxiety individuals had decreased connectivity between amygdala and dorsal/rostral anterior cingulate (dACC and rACC). Functional connectivity of these connections was not affected in ASD−NoAnx, and amygdala connectivity with ventral ACC/mPFC circuits was not different in ASD+Anxiety or ASD−NoAnx relative to TD. Decreased amygdala-dmPFC/rACC connectivity was associated with more severe social impairment in ASD+Anxiety; amygdala-striatal connectivity was associated with RRB symptom severity in ASD−NoAnx individuals. These findings suggest comorbid anxiety in ASD is associated with disrupted emotion monitoring processes supported by amygdala-dorsal ACC/mPFC pathways, whereas emotion regulation systems involving amygdala-vmPFC are relatively spared. Our results highlight the importance of accounting for comorbid anxiety for parsing ASD neurobiological heterogeneity.

Anxiety disorders, including generalized anxiety disorder, phobia, and panic disorder, as well as related obsessive-compulsive disorders, are among the most common comorbid conditions among individuals with autism spectrum disorder (ASD) (Lai et al., 2019). They disproportionately affect individuals with ASD with an estimated prevalence of 20% (anxiety) and 9% (obsessive-compulsive) (Lai et al., 2019) compared to prevalences in the wider population of 6% (anxiety) (Polanczyk et al., 2015) and < 1% (OCD) (Adam et al., 2012). Increased anxiety in ASD also is associated with more severe restricted, repetitive behaviors (RRBs) (Cashin & Yorke, 2018; Gotham et al., 2013; Rodgers et al., 2012) and social-communication abnormalities (Duvekot et al., 2018). However, comorbid anxiety in ASD is relatively understudied, due to challenges objectively evaluating anxiety among patients and the atypical nature of the clinical expression of anxiety in some individuals with ASD, including unusual phobias (e.g., toilets), behavioral distress in anticipation of change in routines or the environment, and frequent sensory fears (e.g., loud sounds). Treatments aimed at mitigating anxiety-related issues in ASD also appear to be less effective than in non-ASD populations highlighting the need to better understand neurobiological mechanisms of anxiety in ASD (B. H. King et al., 2009; Selles et al., 2015).

Resting-state functional magnetic resonance imaging (rs-fMRI) studies are well suited to objectively investigate neurophysiological processes associated with anxiety in ASD. ASD appears to be a disorder involving atypical brain connectivity (Kessler et al., 2016; Müller, 2007), and functional connectivity of rs-fMRI has been used in ASD to predict clinical outcomes using cross-validated prediction models (Abraham et al., 2017; Plitt et al., 2015). The task-free nature of rs-fMRI has advantages for quantifying the neurobiology of anxiety disorders as well, which are characterized by idiosyncratic responses to anxiety-related cues that can present challenges for task-based fMRI.

Findings from functional connectivity studies in ASD have been notably inconsistent (Müller et al., 2011; Picci et al., 2016), probably due to methodological and analytical variation (Müller et al., 2011; Nair et al., 2014) as well as clinical and neurobiological heterogeneity across affected individuals (J. B. King et al., 2019; Linke et al., 2017; Uddin et al., 2013). The prevalence of comorbid conditions in ASD, including anxiety, obsessive-compulsive disorder, mood disorders, ADHD, conduct disorders, and schizophrenia (Lai et al., 2019) likely represents a substantial source of neurobiological variability, though few studies have systematically compared brain connectivity between ASD individuals with and without comorbid psychiatric disorders. The advent of multisite ASD imaging databases with large sample sizes, including the EU-AIMS Longitudinal European Autism Project (LEAP) and the Autism Brain Imaging Data Exchange (ABIDE) (Di Martino et al., 2014, 2017), present an opportunity to separate distinct subgroups of individuals with ASD based on comorbid conditions, helping to resolve inconsistencies across studies of the neurobiology of ASD.

Preclinical and clinical studies of anxiety disorders have implicated dysfunction of the amygdala. The amygdala has dense bidirectional connections with anterior cingulate cortex (ACC) and medial prefrontal cortex (mPFC), (Kim et al., 2011) and heightened amygdala reactivity in patients with anxiety disorders has been shown to be associated with diminished top-down ACC/mPFC control of amygdala reactivity (Etkin, 2009; Jalbrzikowski et al., 2017; Qin et al., 2014; Swartz et al., 2014). ACC/mPFC targets of amygdala nuclei include dorsal and ventral divisions that serve distinct functions related to anxiety derived from their roles in emotion processing (Etkin, 2009). Emotion processing involves monitoring one’s internal emotional state and modifying the emotion being expressed to regulate one’s response based on context. The dorsal division, including dorsal anterior cingulate cortex (dACC) and the adjacent dorsomedial prefrontal cortex (dmPFC), is involved in emotion monitoring and evaluation (Etkin, 2009). The ability to resolve either ambiguous emotional cues (Simmons et al., 2008) or conflict from multiple incongruent cues (Etkin et al., 2006) involves dACC/dmPFC interactions with the amygdala, and in anxiety disorders these evaluations may be more likely to be biased to fear responses. In contrast, the ventral medial prefrontal division, which includes rostral anterior cingulate (rACC), subgenual anterior cingulate (sgACC), and ventromedial prefrontal cortex (vmPFC) is involved in emotional regulation (Etkin, 2009). After an amygdala-activating stimulus is determined to be non-threatening, the emotional state can be modified by these ventral areas via amygdala inhibition; activation in these three ventral areas is inversely correlated with amygdala activation during emotion processing indicating a regulatory role for ventral ACC/mPFC (Phelps et al., 2004), and this regulatory role is dysfunctional in many anxiety disorders. Amygdala-striatal networks also appear to play a significant role in regulating emotional responses to threatening stimuli. For example, amygdala-striatal function is associated with avoidance of novelty and uncertainty (Lago et al., 2017; Makovac et al., 2016).

Abnormalities of the amygdala have been repeatedly implicated in ASD (Baron-Cohen et al., 2000). Sparse neuronal density in the amygdala has been observed in adolescents with ASD (Schumann & Amaral, 2006), and amygdala enlargement in early development has been linked to the severity of social behavior symptoms (Mosconi, Cody-Hazlett, et al., 2009; Nordahl, 2012). Amygdala-prefrontal-ventral striatal (VS) circuits have been implicated in social-communication abnormalities (Chevallier et al., 2012; Odriozola et al., 2019; Rausch et al., 2016; Shen et al., 2016), and prefrontal cortical-striatal circuit alterations appear to relate to the severity of RRBs (Agam et al., 2010; D’Cruz et al., 2016; Delmonte et al., 2013; Thakkar et al., 2008). Despite these advances in mechanistic understanding of anxiety and relevant brain circuitry, there have not been systematic investigations of functional connectivity of amygdala-ACC/mPFC and amygdala-striatal pathways and their relationship with anxiety in ASD. Comparing amygdala connectivity between individuals with ASD with and without comorbid anxiety is essential for determining both how ASD symptoms relate to trait anxiety, and the nature of emotion processing dysfunctions that lead to anxiety in individuals with ASD.

In the current study, we used rs-fMRI to examine functional connectivity of multiple discrete amygdala-ACC/mPFC and amygdala-striatal connections among individuals with ASD and comorbid anxiety (ASD+Anxiety) as well as individuals with ASD without a reported history of an anxiety disorder (ASD−NoAnx) and TD controls. Consistent with studies of individuals with anxiety disorders without ASD (Kim & Whalen, 2009; Roy et al., 2013), we hypothesized ASD+Anxiety individuals would show decreased functional connectivity between the amygdala and ventral prefrontal/striatal regions (rACC, sgACC, vmPFC, and nucleus accumbens) relative to ASD−NoAnx and typically developing (TD) individuals. We also expected ASD+Anxiety individuals to show decreased connectivity between the amygdala and dorsal prefrontal regions (dACC and dmPFC) relative to ASD−NoAnx and TD individuals consistent with alterations in cognitive appraisal and emotion monitoring functions (Etkin et al., 2009; Simmons et al., 2008). To determine the extent to which amygdala-ACC/mPFC and amygdala-striatal connectivity was associated with core clinical symptoms of ASD, we also examined the relationships between connectivity of target amygdala pathways and clinically rated social-communication and RRB symptoms.

Methods

Participants

Resting state fMRI data from a total of 232 participants from the multisite ABIDE (fcon_1000.projects.nitrc.org/indi/abide/) were analyzed (Di Martino et al., 2014, 2017). ABIDE is an open, anonymized neuroimaging database, in which all data was obtained following informed consent/assent procedures that were approved by the human subjects boards at each contributing institution. Three groups were identified from the database, matched on age (5–18 years), gender ratio, and IQ (Table 1, Figure 1): 1) the ASD+Anxiety group (N = 25) included individuals with a current diagnosis of ASD and a comorbid diagnosis of Generalized Anxiety Disorder (N = 12), phobia (N = 14), and/or Obsessive-Compulsive Disorder (N = 3) with no other comorbid psychiatric disorders; 2) the ASD−NoAnx group (N = 68) included individuals with a current diagnosis of ASD but no comorbid psychiatric diagnosis, and 3) the TD group (N = 139) included individuals with no history of or current psychiatric or developmental disorders. Participants were included from the six sites that reported an individual with ASD+Anxiety (Erasmus University Medical Center Rotterdam, Institut Pasteur, Kennedy Krieger Institute, New York University Langone Medical Center [two sites], and Oregon Health and Science University). Comorbid disorders were determined by the research team at each site as part of their standard diagnostic assessments, based on the Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS) (five sites: IP, KKI, NYU1&2, OHSU) or the Diagnostic Interview Schedule for Children – Young Child Version (DISC-YC) (one site, EMC). Psychotropic medication use was an exclusionary criteria for all groups, with the exception of stimulants that were withheld for 24–48 hours prior to the MRI scan (ASD+Anxiety N = 2, ASD−NoAnx N = 10). ABIDE subject IDs for each group are included in the supplementary material.

Table 1.

Clinical and demographic information for ASD+Anxiety individuals, ASD−NoAnxiety individuals, and typically developing (TD) controls.

	ASD+Anxiety	ASD−NoAnx	TD Controls	Comparison
N (% Female)	25 (20.0%)	68 (17.6%)	139 (18.0%)
Age (years)	10.1 (3.2) CI [8.7, 11.4]	10.2 (3.2) CI [9.4, 11.0]	10.6 (2.5) CI [10.2, 11.0]	F(2,229) = 0.60 p = .547
IQ: Full	103.9 (11.4) CI [98.9, 108.9]	103.5 (14.0) CI [100.0, 107.0]	110.2 (11.8) CI [108.1, 112.2]	F(2,212) = 7.17 p < .001
IQ: Perf	105.0 (14.7) CI [98.4, 111.5]	104 (16.7) CI [99.5, 108.6]	106.9 (14.4) CI [104.2, 109.6]	F(2,185) = 0.73 p = .484
IQ: Verbal	102.3 (9.6) CI [97.7, 106.8]	101.3 (14.3) CI [97.2, 105.4]	112.8 (13) CI [110.3, 115.3]	F(2,169) = 15.28 p < .001
ADI Social	18.7 (5.2) CI [16.4, 20.9]	19.2 (5.7) CI [17.7, 20.8]	-	t(44.7) = 0.45 p = .656
ADI Comm.	16.3 (4.6) CI [14.3, 18.3]	15.4 (4.3) CI [14.3, 16.5]	-	t(38.0) = 0.84 p = .409
ADI RRB	7.0 (2.0) CI [6.1, 7.9]	5.5 (2.6) CI [4.8, 6.2]	-	t(52.0) = 2.85 p = .006

Note: Values include mean, standard deviation in parentheses, and 95% confidence intervals. Degrees of freedom for t-tests used the Satterthwaite approximation.

Figure 1.

Participant characteristics. The typically developing control group had higher full-scale IQs than the ASD+Anxiety and ASD−NoAnx groups, which was driven by higher Verbal IQs, but not Performance IQ. The two ASD groups overlapped in severity of social and communication impairments, but the ASD+Anxiety group had more severe RRBs. White dots and error bars indicate means and 95% confidence intervals.

IQ was quantified using the WASI (N = 105), WISC-IV (N = 85), DAS-II (N = 22), WAIS-IV (N = 1), WISC-III (N = 1), or WISC-V (N = 1), and was not available for 17 participants (Performance IQ was not reported for an 44 participants, and Verbal IQ was not reported for 60 participants). In the two ASD groups, severity of social impairment, communication, and RRBs were quantified using total scores (diagnostic algorithm) from the Autism Diagnostic Interview – Revised (Lord et al., 1994); ADI data was not available for 2 ASD+Anxiety and 11 ASD−NoAnx participants.

Data Analysis

Rs-fMRI preprocessing.

Scanner type and sequence parameters varied across sites and are detailed in the supplementary material. Resting state scan lengths varied across sites from 5:07 to 7:55 minutes (85–180 volumes). Estimated smoothness varied across sites (EMC: M = 10.42, SD = 0.85; IP: M = 7.45, SD = 0.27; KKI: M = 7.15, SD = 0.42; NYU1: M = 6.68, SD = 0.36; NYU2: M = 5.78, SD = 0.23; OHSU: M = 9.30, SD = 0.53); to account for this and other potential sources of variability, testing site was included as a covariate of non-interest in all analyses. Resting state fMRI data were analyzed using the Configurable Pipeline for the Analysis of Connectomes, C-PAC (Yan et al., 2013), which incorporates neuroimaging tools FSL (Jenkinson et al., 2012), AFNI (Cox, 1996), and ANTs (Avants et al., 2014). Anatomical data were conformed to RPI orientation, then registered to a 2mm MNI152 brain-only template by applying a non-linear transform to the skull-on images using ANTS. Images were then skull-stripped using AFNI’s 3dSkullStrip and segmented into WM, GM, and CSF using FSL’s fast tool. The resulting WM mask was multiplied by a WM prior map that was transformed into individual space using the inverse of the linear transforms previously calculated during the ANTs procedure. A CSF mask was multiplied by a ventricle map derived from the Harvard-Oxford atlas distributed with FSL. Skull-stripped images and grey matter tissue maps were transformed into MNI space at 2mm resolution.

Functional data were sampled to RPI orientation, the first four volumes were censored, and slice timing correction was applied. Motion correction was performed by first coregistering functional images to the mean image using AFNI 3dvolreg, and then calculating a new mean image and coregistering functional images to the new mean. Functional to anatomical registration was performed using FSL’s boundary-based registration and a 7 degree of freedom linear transform. Nuisance variable regression was applied to the motion-corrected data and included a 2nd order polynomial, a 24-parameter motion model (the three translation and three rotation parameters of the current and preceding volume, and their squared values), and five principal components from WM and CSF noise ROIs (CompCor) (Behzadi et al., 2007) to correct for physiological noise. Two preprocessing pipelines were run in parallel, one that included global signal regression, and one without (see Murphy & Fox, 2017). Residuals of the nuisance variable regression were bandpass filtered (0.001Hz, 0.1Hz), written into MNI space at 3mm resolution, and smoothed using a 6mm FWHM kernel.

Connectivity matrix.

Timecourses were extracted from preprocessed data in select regions of interest (ROI). Left and right amygdala ROIs (Figure 2a) were defined using the probabilistic, cytoarchitectonic, Juelich histological atlas (Amunts et al., 2005). Medial prefrontal cortex (Figure 2c) and basal ganglia (Figure 2b) ROIs were defined using the whole-brain functional atlas, Craddock 200 (Craddock et al., 2012), which uses spectral clustering to parcellate the brain into regions with homogeneous connectivity patterns. The five medial prefrontal ROIs (all bilateral) included dmPFC, dACC, rACC, sgACC, and vmPFC. Six basal ganglia ROIs included left and right head of the caudate, nucleus accumbens, and putamen. Further details on ROI definitions are included in supplementary material.

Figure 2.

Region of interest definitions. Amygdala ROIs (A, coronal slice at Y = −7) were defined using the Juelich histological atlas. Basal ganglia (B, coronal slice at Y = 8) and anterior cingulate/medial prefrontal cortex (C, saggital slice at X = 0) ROIs were defined using the Craddock 200 functional atlas.

Pairwise correlations were computed between left/right amygdala and each of the ACC, PFC and basal ganglia ROIs. Pearson’s correlation coefficients were normalized using the Fisher’s R to Z transform, and then converted to Z-scores by dividing by the square root of the variance. Degrees of freedom (i.e., number of time points) were corrected for autocorrelation in the BOLD signal using a whole-brain Bartlett Correction Factor individually for each subject (Fox et al., 2005; Van Dijk et al., 2010); implementation is described in the supplementary material. This correction was necessary due to the non-independence of time points in the BOLD signal and the varying run lengths across sites (85–180 volumes), but precluded scrubbing volumes based on motion, due to the need for a continuous timecourse in calculating autocorrelation in the BOLD signal. Motion was accounted for in analyses by adding mean framewise displacement as an individual-level nuisance predictor in all regression models.

Regression models.

Z-score connectivity values were analyzed with linear mixed effect regression models using the lme4 package (Bates et al., 2015) in R (R Core Team, 2018). Each of the five ACC and mPFC and three basal ganglia ROIs were examined in a separate model; test statistics were not adjusted for multiple comparisons across these eight models, as each ROI was chosen for its theoretical importance in ASD+Anxiety. Categorical fixed effects were contrast coded, and continuous fixed effects were transformed to Z-scores. Site was included as a nuisance predictor using dummy coding, with NYU1 (the most populous site) as the reference level. For clarity, individual site estimates are not included in figures of model fixed effect estimates, but can be found in the full model outputs in the supplementary material. Each model contained fixed effects of amygdala-hemisphere (Left −.5, Right +.5), age (linear and quadratic orthogonal polynomials), site (dummy coding), motion (mean framewise displacement), and random intercepts of subject. All models with basal ganglia ROIs included an additional fixed effect of crossing (ipsilateral −.5, contralateral +.5), referring to whether the amygdala ROI and basal ganglia ROI were in ipsilateral or contralateral hemispheres. Group was analyzed using Helmert coding. The first contrast compared the TD control group to the ASD−NoAnx and the ASD+Anxiety groups (TD vs ASD−NoAnx&ASD+Anx: TD −2/3, ASD−NoAnx +1/3, ASD+Anxiety +1/3), and the second contrast compared the two ASD groups to each other (TD 0, ASD−NoAnx −.5, ASD+Anxiety+.5); models also included interactions between each group contrast and age, hemisphere, and crossing. Models assessing the relationship between connectivity and ASD symptoms included only ASD−NoAnx and ASD+Anxiety participants, and thus included a single group fixed effect contrast (ASD−NoAnx −.5, ASD+Anxiety +.5). Clinical severity models also included fixed effects of Social impairment, Communication, and RRBs (ADI total scores, diagnostic algorithm), as well as interactions between each clinical measure and group. Because ADI scores referred to symptoms at age 4–5, ADI-age interactions were included as additional fixed effects. Parameter-specific p-values were calculated based on the Kenwood-Rogers approximation for degrees of freedom. Standardized beta coefficients were calculated by scaling coefficients by two standard deviations (Gelman, 2008) using the R package sjplot (Lüdecke, 2019). Individual group comparisons were performed by comparing estimated marginal means of each group from the full model using the R package emmeans (Lenth et al., 2019). Additional models included IQ (full-scale, performance, verbal) and group*IQ interactions as additional fixed effects. However, IQ did not improve model fit and had small effect sizes, and thus was not included in the models below in order to maximize participant inclusion (17 individuals were missing full-scale IQ data, 44 were missing performance IQ, and 60 were missing verbal IQ).

Data Availability

Raw data was obtained from the freely available ABIDE I and II databases. ABIDE subject IDs used in the current study are listed in the supplementary material.

Results

Amygdala – ACC and amygdala-mPFC connectivity

ASD+Anxiety individuals showed decreased connectivity between amygdala and dorsal ACC regions including dACC and rACC relative to both comparison groups: ASD−NoAnx vs ASD+Anxiety t(232) = 3.940_dACC, 3.010_rACC, p = .0001_dACC, .003_rACC, TD vs ASD+Anxiety t(232) = 3.396_dACC, 2.647_rACC, p = .001_dACC, .009_rACC (Table S4 and Figures 3, 5). In contrast, ASD+Anxiety individuals did not show any differences in amygdala connectivity with ventral ACC/PFC (sgACC and vmPFC) or dorsal PFC (dmPFC) relative to the ASD−NoAnx and TD control groups, ASD−NoAnx vs ASD+Anxiety t(232) = 0.826_sgACC, 0.528_vmPFC, 1.171_dmPFC, p = .410_sgACC, .598_vmPFC, .243_dmPFC, TD vs ASD+Anxiety t(232) = 0.403_sgACC, 0.741_vmPFC, 0.655_dmPFC, p = .687_sgACC, .460_vmPFC, .513_dmPFC. The ASD−NoAnx group did not show any differences in amygdala connectivity with dorsal or ventral ACC/PFC targets relative to the TD controls, t(232) = 0.0871_dACC, 0.786_dmPFC, 0.588_rACC, 0.643_sgACC, 0.309_vmPFC, p = .385_dACC, 433_dmPFC, .557_rACC, .521_sgACC, .758_vmPFC. A laterality effect was observed where sgACC and vmPFC had higher connectivity to the left amygdala than the right (Cohen’s d effect sizes: 0.58_sgACC, 0.41_vmPFC) but there were no interactions between amygdala hemisphere and group.

Figure 3.

Amygdala – PFC connectivity. Comorbid anxiety was associated with decreased connectivity between the amygdala and dorsal/rostral ACC. Developmental effects on amygdala connectivity were observed throughout the ACC, with age-related decreases in amygdala-dACC/rACC connectivity, and a U-shaped pattern of amygdala-sgACC connectivity, with lower levels in early adolescence compared to childhood or later adolescence. Amygdala connectivity was left lateralized for sgACC and vmPFC, but did not interact with group. Points are standardized fixed effect estimates with 95% confidence intervals. Predictors are contrast coded: Hemisphere (Left −.5, Right +.5), TD vs ASD−NoAnx&ASD+Anxiety (TD −2/3, ASD−NoAnx+1/3, ASD+Anxiety +1/3), ASD−NoAnx vs ASD+Anxiety (TD −1/3, ASD−NoAnx −1/3, ASD+Anxiety +2/3). Age and Age² are orthogonal polynomials centered in the range 5–18 years. Subjects’ mean framewise displacement (FD) is Z-transformed. *p < .05, **p < .01, ***p < .001.

Figure 5:

Amygdala – PFC connectivity. Top row: Comorbid ASD+Anxiety (dark magenta) was associated with decreased connectivity between bilateral amygdala and dorsal/rostral emotional attention/regulation areas (dACC, and rACC) compared to ASD−NoAnx and TD control groups. Error bars represent 95% confidence intervals. Age-related changes in amygdala connectivity (second row) were observed throughout the ACC (age was modeled with orthogonal linear and quadratic components). Age-associated connectivity differences were characterized by linear decreases in dACC/rACC, and in sgACC by a U-shaped curve, with connectivity higher in childhood, decreasing in early adolescence and then beginning to increase in adulthood. Greater social impairment scores on the ADI (third row) were associated with decreased connectivity between amygdala and dmPFC/rACC in the ASD+Anxiety group (dark magenta), but not the ASD−NoAnx group (light blue). Greater communication impairment was associated with increased connectivity between amygdala and rACC in the ASD+Anxiety group, and decreased amygdala-dACC connectivity in the ASD−NoAnx group. Lines in rows 2–5 represent predicted model estimates when all other predictors were set to 0 (i.e., mean values).

Amygdala – Basal Ganglia Connectivity

ASD+Anxiety individuals showed decreased connectivity between the amygdala and the nucleus accumbens compared to ASD−NoAnx individuals: t(232) = 2.150_NucAcc, p = .033_NucAcc, but no differences were observed between groups in amygdala – caudate or amygdala – putamen connectivity (Table S6, Figures 6, 8). Strong laterality effects were observed for amygdala – putamen connectivity, with greater connectivity from the right versus the left amygdala, and for ipsilateral versus contralateral amygdala – putamen connections. No group by side interactions were seen for amygdala connectivity with basal ganglia nuclei.

Figure 6.

Amygdala – Basal ganglia connectivity. Comorbid ASD+Anxiety was associated with decreased connectivity between the amygdala and the nucleus accumbens compared to ASD−NoAnx individuals. A developmental effect of amygdala-nucleus accumbens connectivity was observed, with linear decreases in connectivity with age. Amygdala-putamen connectivity was right lateralized and greater for ipsilateral connections but did not interact with group. Points are standardized fixed effect estimates with 95% confidence intervals. Predictors are contrast coded: Hemisphere (Left −.5, Right +.5), Crossing (ipsilateral −.5, contralateral +.5), TD vs ASD−NoAnx&ASD+Anxiety (TD −2/3, ASD−NoAnx+1/3, ASD+Anxiety +1/3), ASDNo vs ASDAnx (TD −1/3, ASD−NoAnx −1/3, ASD+Anxiety +2/3). Age and Age² are orthogonal polynomials centered in the range 5–18 years. Subjects’ mean framewise displacement (FD) is Z-transformed. *p < .05, **p < .01, ***p < .001.

Figure 8.

Amygdala – basal ganglia connectivity. Comorbid ASD+Anxiety was associated with decreased connectivity between the amygdala and the nucleus accumbens compared to ASD−NoAnx individuals. Increased RRB severity was associated with increased connectivity between the amygdala and the Caudate/Putamen only for ASD−NoAnx individuals. A linear effect of age was observed for the nucleus accumbens, where amygdala connectivity decreased across development; no interactions with ASD or anxiety were observed. Top row: individual connectivity scores and group means with 95% confidence intervals. Second row: curves represent model fits for each group; age was modelled with orthogonal linear and quadratic components. Third – fifth rows: predicted model estimates across the ADI social, communication, and RRB total score ranges when each other predictor was set to 0 (i.e., mean values).

Clinical and demographic associations.

Age was associated with connectivity between the amygdala and all three ACC regions, but not with PFC connectivity. Amygdala-sgACC connectivity across participants followed a U-shaped pattern with higher connectivity in childhood and early adulthood compared to adolescence (Figure 5; effect size: 0.36_{linear_age}, 0.43_{quad_age}). In dACC and rACC, linear decreases in age-related amygdala connectivity were observed (effect size: 0.28_{linear_age_dACC}, 0.34_{linear_age _rACC}). No diagnostic group differences were seen in the relationships between age and amygdala-ACC/PFC connectivity. In the basal ganglia, increased age was linearly associated with lower amygdala – nucleus accumbens connectivity, but no group by age interactions were observed for amygdala – nucleus accumbens connectivity, and age was not associated with amygdala-caudate or amygdala-putamen connectivity.

More severe ADI-rated social abnormalities were associated with decreased connectivity between the amygdala and dmPFC/rACC for individuals with ASD+Anxiety (Table S5, Figure 5; effect size = 0.57_dmPFC, 0.40_rACC) (dmPFC Estimate = −0.77, SE = 0.30, t(80) = −2.58, p = .012; rACC Estimate = −0.55, SE = 0.28, t(80) = −1.97, p = .052), but these connections were not associated with social interaction scores in the ASD−NoAnx group (dmPFC Estimate = 0.08, SE = 0.17, t(80) = 0.50, p = .618; rACC Estimate = 0.01, SE = 0.16, t(80) = 0.04, p = .968). Social abnormalities were not associated with any other amygdala-PFC connections. More severe ADI-rated communication impairments were associated with increased amygdala rACC connectivity for individuals with ASD+Anxiety (Table S5, Figure 5; effect size = 0.56) (Estimate = 0.61, SE = 0.23, t(80) = 2.61, p = .011), but this relationship was not significant for the ASD−NoAnx group (Estimate = −0.08, SE = 0.16, t(80) = −0.50, p = .618). Instead, increased communication impairments were associated with decreases in amygdala – dACC connectivity in the ASD−NoAnx group (effect size = 0.28) (Estimate = −0.32, SE = 0.16, t(80) = −1.99, p = .050); this association was not observed in ASD+Anxiety individuals (Estimate = 0.03, SE = 0.24, t(80) = 0.12, p = 0.904.

More severe RRB symptoms on the ADI were associated with increases in connectivity between the amygdala and the Caudate/Putamen for ASD−NoAnx individuals (Table S7, Figures 7, 8; effect size =) (Caudate Estimate = 0.31, SE = 0.12, t(80) = 2.65, p = .010; Putamen Estimate = 0.30, SE = 0.13, t(80) = 2.31, p = .023), but these associations were not observed in the ASD+Anxiety group (Caudate Estimate = 0.02, SE = 0.22, t(80) = 0.10, p = .921; Putamen Estimate = 0.27, SE = 0.24, t(80) = 1.14, p = 0.26). Neither Social nor Communication scores were associated with amygdala – basal ganglia connectivity in either ASD group.

Figure 7.

Amygdala – Basal ganglia connectivity and clinical ratings in individuals with ASD. Increases in RRB symptom severity were associated with amygdala – Caudate/Putamen connectivity. Points are standardized fixed effect estimates with 95% confidence intervals. Predictors are contrast coded: Hemisphere (Left −.5, Right +.5), Crossing (Ipsilateral −.5, Contralateral +.5), ASD−NoAnx vs ASD+Anxiety (ASD−NoAnx −.5, ASD+Anxiety +.5). Age and Age² are orthogonal polynomials centered in the range 5–18 years. Subjects’ mean framewise displacement (FD) is Z-transformed. Social, Comm., and RRB are ADI Total scores for Social, Communication, and Restricted and Repetitive Behaviors (Diagnostic Algorithm); values were scaled by converting to Z-scores using sample mean and SD. *p < .05, **p < .01, ***p < .001.

Discussion

The current study provides new evidence for disrupted functional connectivity between amygdala and the dorsal division of mPFC in ASD that is specific to patients with a comorbid anxiety disorder. In contrast, amygdala-vmPFC connectivity was relatively unaffected in both ASD subgroups. Based on the ACC/PFC regions examined, these results indicate that clinical anxiety in ASD may predominantly involve disruption of amygdala connections to dorsal ACC/PFC that are involved in emotion monitoring and appraisal functions. Importantly, we did not see any differences in amygdala-ACC/PFC connectivity between ASD−NoAnx individuals and TD controls suggesting that alterations of resting-state amygdala-ACC/PFC functional connectivity in ASD are unique to those individuals with a co-occurring anxiety disorder. Consistent with prior work, ASD+Anxiety individuals were rated as having more severe RRBs than ASD−NoAnx individuals, suggesting that anxiety may play a role in the development of RRBs (Cashin & Yorke, 2018; Gotham et al., 2013; Rodgers et al., 2012). More severe social impairment in ASD+Anxiety individuals was associated with decreased amygdala-dmPFC and amygdala-rACC connectivity; communication skills were associated with amygdala-rACC connectivity in individuals with ASD+Anxiety but with amygdala-dACC connectivity in individuals with ASD without anxiety; and RRB symptoms were associated with amygdala-striatal connectivity in ASD−NoAnx individuals, indicating that atypical development of these brain systems may contribute to a broad range of clinical impairments in ASD.

Amygdala – mPFC connectivity in ASD

Dorsal and ventral divisions of mPFC have been shown to exert distinct regulatory roles on amygdala reactivity to emotional stimuli. Dorsal ACC/PFC, including dACC and dmPFC, monitor and appraise emotional cues and integrate lateral prefrontal cognitive control processes (Etkin, 2009). Ventral regions, including rACC, sgACC, and vmPFC, integrate contextual cues and down- or up-regulate amygdala responses according to the perceived threat level based on the context (Etkin, 2009). Decreased ventral ACC/PFC-amygdala connectivity, especially within vmPFC networks, is a common feature of anxiety disorders which results in insufficient downregulation of amygdala reactivity in non-threatening situations (Kim & Whalen, 2009; Roy et al., 2013).

In contrast to prior findings on individuals with anxiety disorders but no ASD (Kim & Whalen, 2009; Roy et al., 2013), our results demonstrate that ventral ACC/PFC-amygdala connections are relatively unaffected in individuals with ASD and comorbid anxiety, though there was some evidence that amygdala-rACC connectivity was disrupted. Though rACC and the more ventral sgACC and vmPFC have overlapping roles, they are recruited for different types of emotion regulation. rACC is innervated by emotion monitoring circuits of dACC and dmPFC, and it plays a role in emotional regulation when resolving ambiguous or conflicting emotional inputs (Etkin et al., 2006). More ventral sgACC and vmPFC systems integrate inputs from sensory processing and memory circuits of temporal and parietal cortex and are preferentially involved in emotion regulation in response to aversive or threatening cues (Indovina et al., 2011; Motzkin et al., 2015).

Selective impairment in rACC is consistent with a pattern of disrupted amygdala connectivity driven by the more dorsal dACC and dmPFC. Together, atypical connectivity of amygdala and dACC, dmPFC and rACC suggest that in ASD+Anxiety, emotional monitoring, appraisal, and conflict resolution may be especially impacted. Atypical regulation of subcortical circuits by dorsal ACC/PFC has been shown to disrupt multiple affective and cognitive processes in ASD, including behavioral flexibility and inhibitory control (D’Cruz et al., 2016; Mosconi, Kay, et al., 2009; Voorhies et al., 2018). The possibility of overlapping etiologies for these deficits involving dysfunctional dorsal ACC/PFC highlights the need to integrate anxiety and cognitive control functions of dorsal ACC/PFC in neurodevelopmental models of ASD.

Amygdala – prefrontal and amygdala-striatal connectivity did not differ between individuals with ASD without anxiety and TD controls suggesting that amygdala-ACC/PFC disturbances may be specific to patients with comorbid anxiety conditions. Results of whole-brain functional connectivity studies of ASD have been notably inconsistent (Müller et al., 2011; Picci et al., 2016), and outcomes vary based on analytical approaches (Müller et al., 2011; Nair et al., 2014) and sample variability (Linke et al., 2017). Amygdala connectivity has been repeatedly investigated in ASD, with many findings of amygdala hypo-connectivity with frontal, striatal, and temporal targets compared to TD (Iidaka et al., 2019; Rausch et al., 2016, 2018; Shen et al., 2016), but also examples of amygdala hyper-connectivity in ASD (Kleinhans et al., 2016; E. R. Murphy et al., 2012). Notably, when we controlled for common sources of variability in our ASD−NoAnx group, including comorbid conditions and medication use, we did not observe differences in amygdala connectivity compared to TD, but did replicate general findings of amygdala-PFC and amygdala ACC hypo-connectivity in our ASD+Anxiety individuals.

Amygdala -Striatal connectivity

The amygdala sends efferent projections to ventral and dorsal striatum, and we anticipated that these connections would be altered in both ASD groups (Chevallier et al., 2012; Lago et al., 2017; Makovac et al., 2016). We found evidence for disrupted connectivity between the amygdala and the nucleus accumbens in individuals with ASD+Anxiety, which is consistent with prior reports of amygdala-striatum hypoconnectivity in anxiety disorders (Göttlich et al., 2014; Roy et al., 2013). Structural and functional abnormalities of the striatum and of fronto-striatal connectivity have been associated with RRBs in ASD (Abbott et al., 2018; Delmonte et al., 2013; Langen et al., 2014; Thakkar et al., 2008), and we found some evidence for associations between RRB severity and amygdala-caudate/putamen connectivity, but only in individuals with ASD without anxiety. There are multiple overlapping mechanisms that can contribute to RRBs, such as reduced behavioral flexibility (D’Cruz et al., 2016) or sensorimotor disruptions (Unruh et al., 2019), and our findings suggest differences in some of the systems that contribute to RRBs in individuals with ASD with and without comorbid anxiety.

ASD clinical correlations

We found that reduced amygdala – dmPFC and amygdala – rACC connectivity were associated with more severe ADI-rated social abnormalities in individuals with ASD+Anxiety, and more severe communication impairments were associated with increased amygdala – rACC connectivity. This suggests that dysfunctional amygdala-prefrontal connectivity is a potential key neural mechanism of impaired social-communication processing in individuals with ASD who also have comorbid anxiety. In individuals with ASD without anxiety, amygdala-dACC connectivity was inversely associated with communication impairments. Thus, anxiety related to alteration in amygdala-ACC/PFC circuitry may represent an important contributing factor to social-communication challenges in ASD.

Similar symptom profiles may develop from distinct underlying brain processes in individuals with ASD based on their overall clinical presentation, including comorbid disorders. Our results highlight a distinct mechanism of social-communication impairment in a subgroup of ASD with a comorbid anxiety disorder suggesting that multiple neurodevelopmental disruptions may contribute to social-communication symptoms in ASD. The pattern of aberrant dorsal/rostral mPFC connectivity with the amygdala suggests that issues with threat evaluation (Simmons et al., 2008) or anticipatory worrying (Barker et al., 2018; Makovac et al., 2016) may contribute to impaired social interactions in individuals with ASD+Anxiety, which necessitates a different approach than social issues that arise from, for example, issues with social motivation or reward (Chevallier et al., 2012; Delmonte et al., 2012), though the exact relationship is likely to vary between different types of anxiety/obessessive-compulsive disorders.

Limitations

Comorbid anxiety in the current study was broadly characterized to include individuals with diagnoses of GAD, phobia, or OCD. Although these disorders share common neural features, including heightened amygdala reactivity (Etkin & Wager, 2007), they also have distinct neural underpinnings (Blackford & Pine, 2012; Goodwin, 2015) that need to be considered in the context of comorbid ASD. As a result, this study investigates the consequences of common neural features across anxiety disorders, but is not equipped to identify the effects of individual disorders, such as striatal effects on RRBs in ASD+OCD, or associations between ASD+social anxiety and ventral ACC/mPFC. A further limitation in our sample is the lack of specificity in quantifying both anxiety and ASD clinical impairment. Dimensional measures of anxiety were not available, and thus anxiety could only be examined categorically (based on research team evaluations at each site using K-SADS or DISC-YC assessments). Several robust ASD-specific dimensional anxiety assessments recently have been developed, including the ASC-ASD (Rodgers et al., 2016), ADIS/ASD (Kerns et al., 2017), and the PRAS-ASD (Scahill et al., 2019), providing new opportunities to clarify relationships between amygdala-PFC connectivity and anxiety across a broader range of comorbid psychopathology. ASD severity was quantified using the ADI, a diagnostic tool, which was available for most but not all participants. In addition, because individuals from multiple testing sites were combined in order to accumulate a sufficient number of ASD+Anxiety individuals, variability across sites was introduced, owing to differences in scanner parameters or resting state task instructions, though site distributions were generally balanced across groups. Our sample excludes for psychotropic medication use, which are known to have wide-ranging impacts on functional connectivity in ASD that would be difficult to interpret in the context of this multi-site study and without detailed information on dosing and duration (Linke et al., 2017). However, this decision limits the generalizability of our results to more severely impaired individuals with ASD and more severe anxiety cases.

Despite finding that ASD+Anxiety individuals had more severe RRBs than ASD−NoAnx, we did not find any robust associations between RRB severity and amygdala connectivity in ASD+Anxiety individuals, and instead only found amygdala-striatal associtations with RRB severity in individuals with ASD without anxiety. This lack of an association in ASD+Anxiety individuals was unexpected, given that repetitive behaviors have been associated with anxiety, and the amygdala is structurally connected to striatal and PFC regions that have been implicated in RRBs in ASD. However, there are limitations in how RRBs were quantified in the current study that should be considered in the context of this null anxiety effect. Due to the nature of the sample, RRBs were quantified by ADI diagnostic score totals, which are a retrospective account of symptom severity at age 4–5. Retrospective reports involve inherent biases that would be mitigated by contemporaneous measurement of RRBs. In addition, participants’ ages ranged from 5–18 years, whereas diagnostic scores refer to symptom severity at age 4–5, and thus the discrepancy between scanning age and symptom measurement age varies across individuals and approaches 14 years. This discrepancy is particularly important because RRBs tend to shift over development, with more lower-order RRBs (e.g., sensorimotor mannerisms)observed at younger ages and more higher-order RRBs (e.g., an insistence on sameness) at older ages (Esbensen et al., 2009), and the higher-order RRBs, especially insistence on sameness, are most frequently found to be associated with anxiety in ASD (Cashin & Yorke, 2018; Gotham et al., 2013; Rodgers et al., 2012). Our models incorporated moderating effects of age on the associations between ADI scores and amygdala connectivity, but these age effects are likely to vary across individuals, and it will be important to examine how amygdala-PFC and amygdala-striatal circuits relate to current RRB severity across development for distinct subtypes of RRBs.

Conclusions

The present study demonstrates that comorbid anxiety with ASD is associated with decreases in functional connectivity between amygdala and dorsal/rostral medial ACC/PFC areas involved in emotional monitoring and regulation, and decreases in amygdala – nucleus accumbens connectivity. Importantly, our measures of amygdala connectivity were associated with social-communication impairment in individuals with ASD plus anxiety, and to a lesser degree in individuals with ASD without anxiety. This suggests that different neural mechanisms may contribute to similar manifestations of core ASD symptomology in different clusters of individuals with ASD. These results have broad implications for the role of anxiety and other comorbid psychiatric conditions in unraveling neurobiological heterogeneity in ASD, and suggests that comorbid conditions are an important factor that should be considered in studies of functional connectivity in ASD. Future directions should examine ASD+Anxiety in more detail, as well as the broader impact of other comorbid conditions on ASD heterogeneity. Our observations of anxiety-specific associations between amygdala connectivity and broadly defined ASD symptom severity provides a strong rationale for examining detailed associations between quantifiable levels of trait anxiety and current ASD social, communication, and RRB severity. In addition, expanding the scope beyond the amygdala will be necessary to provide a clearer view of amygdala, ACC, mPFC, and striatal interactions within a broader network of cognitive control, sensory, and default mode areas in ASD+Anxiety. Beyond anxiety, associations between core ASD symptoms and other commonly comorbid conditions, including ADHD, conduct disorders, and depression may be useful in understanding the distinct neural mechanisms that can contribute to different phenotypic presentations among individuals with ASD. The different underlying mechanisms have treatment implications as well. Cognitive behavioral therapies that emphasize cognitive control strategies to mitigate anxious symptoms may be especially effective in individuals with dysfunctional amygdala – dmPFC connectivity (Etkin et al., 2009), and responsivity to pharmacological treatments for anxiety may be related to amygdala-mPFC functional connectivity alterations (Faria et al., 2014; Whalen et al., 2008). Understanding comorbid conditions in ASD provides essential knowledge towards mechanistic models of ASD behavior that go beyond the symptoms of the world’s many autisms.

Figure 4.

Relationships between Amygdala – PFC connectivity and clinical ratings in individuals with ASD. Social impairment interacted with group, such that decreased amygdala-dmPFC/rACC connectivity were associated with more severe social impairment only in individuals with ASD+Anxiety. Lower communication skills were associated with increased amygdala-rACC connectivity in individuals with ASD+Anxiety, and decreased amygdala-dACC connectivity in individuals with ASD−NoAnx. Points are standardized fixed effect estimates with 95% confidence intervals. Predictors are contrast coded: Hemisphere (Left −.5, Right +.5), ASD−NoAnx vs ASD+Anxiety (ASD−NoAnx −.5, ASD+Anxiety +.5). Age and Age² are orthogonal polynomials centered in the range 5–18 years. Subjects’ mean framewise displacement (FD) is Z-transformed. Social, Comm., and RRB are ADI Total scores for Social, Communication, and Restricted and Repetitive Behaviors (Diagnostic Algorithm); values were scaled by converting to Z-scores using sample mean and SD. *p < .05, **p < .01, ***p < .001.