Processing of emotional faces has unique functional and cytoarchitectural associations in those with an autism spectrum diagnosis

Abstract

Those with an autism spectrum diagnosis (ASD) have been found to process emotional faces differently than other populations. Processing of emotional faces requires engagement of temporal, frontal, occipital, and limbic brain regions. Functional activity has been shown to differ in those with an ASD, particularly in the amygdala, inferior frontal cortex (IFC), and temporal brain regions. However, the consistency and direction of these associations have been inconsistent across studies. Recent findings have demonstrated that measures of neuron density differ in those with an ASD. Some of these regional differences in cytoarchitecture coincide with regions important to emotional facial processing. Therefore, the interaction between cytoarchitecture and functional activity may be important in elucidating unique neurophysiology in ASD. The present study uses diffusion weighted imaging (DWI) and functional magnetic resonance imaging (fMRI) data from the Adolescent Brain Cognitive Development^sm (ABCD®) study to investigate the relationship between cytoarchitecture and functional activity during emotional face processing in those with an ASD. 149 individuals with a reported ASD and 10,146 individuals with no reported diagnosis of an ASD (nASD) were identified. The emotional n-back (EN-Back) task was administered during fMRI acquisition, activating regions of the brain associated with processing emotional faces. Neuron cell body density was positively correlated with functional activation in the left amygdala in the ASD group but not the nASD group. These findings suggest that a unique relationship may exist between neuron cell body density in the left amygdala and functional activity while processing emotional faces in those with an ASD.

Introduction

Differences in social communication and social interactions are among the defining characteristics in individuals who receive an autism spectrum diagnosis (ASD) [2]. Communicating emotions through facial expressions is critical to social interactions, and people with an ASD have been found to process emotional faces differently than other populations [31, 69, 70]. The behavioral presentation of abnormal facial processing in autism includes poor memory (Boucher et al., 1998), differences in eye-tracking (Sterling et al., 2008), deviations in perception of emotion (Ashwin et al., 2006), and difficulty evaluating trustworthiness (Adolphs et al., 2001). Despite the well-established association of ASD and atypical facial processing, the underlying functional-structural relationships potentially driving these behavioral manifestations are still poorly understood [73].

Human processing of faces encompasses a variety of steps that differentially engage neuroanatomical regions depending on task-specific demands [39]. The occipital face area and fusiform facilitate processing of visual stimuli related to the face [39, 40, 43]. The fusiform plays a particularly important role in simple face detection [66], and the amygdala is primarily involved with discrimination between emotional and neutral faces [60, 71]. Further higher-order processing of emotional faces engages the inferior frontal cortex (IFC) [22, 58], and the uncinate fasciculus which connects regions in the temporal lobe [18]. Together, these regions represent a macroscopic system of interdependent structures implicated in processing emotional faces. While clear and distinct tasks have helped elucidate key regions in the processing of emotional faces, deficits related to psychiatric and developmental disorders such as ASD appear to produce more subtle changes in neurophysiology.

Previous investigations found that simple processing of emotional faces in ASD was associated with greater functional activity in the amygdala [51] and decreased functional connectivity in the inferior frontal regions in the context of working memory tasks [58]. In contrast, more recent investigations have found decreased functional activation in the fusiform gyrus [53] and decreased functional activity and connectivity in the amygdala [33, 47] in relation to emotional faces. These discrepancies may be a result of overly general measures obscuring more discrete changes. Functional activity in the amygdala has been reported to be more distinctly related to fear than other emotional stimuli in ASD [50]. Other studies have identified altered habituation of the amygdala to emotional stimuli in ASD [68]. However, seeking novel approaches to characterize these neuroanatomical regions permits leverage of novel techniques without limiting the scope of behavioral responses being considered.

Recent investigation of the fusiform found that predictive models trained on features derived from multiple modalities of structural and functional imaging, better predicted social outcomes in individuals with an ASD than models trained on features of a single modality [27]. Although models reducing many features derived from multiple modalities are difficult to interpret, their predictive accuracy may indicate that properly characterizing the relationship between neurophysiology and social outcomes requires multiple modalities. For example, administering a selective-serotonin reuptake inhibitor to those with an ASD shifts functional activity of serotonergic neurons in the ventromedial prefrontal cortex towards neurotypical individuals when viewing emotional faces [77]. Therefore, changes in functional activity may be tied to characteristics of specific neuron populations. Differences in neuron populations of those with ASD have been found in non-task associated analyses of cytoarchitecture [17]. Individuals with ASD have been found with a reduced number of amygdalar neurons [63], fewer cerebral Purkinje cells [41], smaller Purkinje cells in the cerebellum [25], patches of cortical disorganization [67], and a greater number of narrow minicolumns in the frontal and temporal lobes [10, 12, 13]. Although these findings directly implicate anomalous measures of gray matter in those with ASD, they do not provide a clear connection between multimodal research that has implicated alterations in gray matter with social-emotional outcomes in ASD. There remains a need to directly characterize the relationship between anomalies in gray matter of those with ASD and functional processes that support socio-emotional processing.

The present study sought to identify unique associations between functional activity and neuron density, building on previous findings in the Adolescent Brain and Cognitive Development^sm (ABCD®) study [4, 5, 14, 34]. In addition to selecting regions implicated in functional processing of emotional faces, the present study selected regional and cytoarchitectural measures based on notable differences that were previously detected in those with an ASD in the ABCD study cohort. Specifically, the present study focused on recent findings that demonstrated those with an ASD had increased restricted normalized directional diffusion (RND) within the inferior frontal and fusiform cortices, and increased restricted normalized isotropic diffusion (RNI) within the amygdala [17]. Our hypothesis was that the magnitude of these changes within individuals with ASD would correspond to changes in functional activity while processing emotional faces, when contrasted to those without an ASD (nASD).

Methods

Study population

All measures were pulled from the Adolescent Brain and Cognitive Development^sm (ABCD®) study 6.0 release [37]. The ABCD study consists of 11,873 children recruited between ages 9–10-year-old at baseline. Participants were recruited from 22 sites (21 active sites) across the US from public and private schools and selection was based on race and ethnicity, sex, socioeconomic status, and urbanicity, in an effort to match US population and demographics [28]. The ABCD study is currently in the process of following these individuals over the course of 10 years—https://abcdstudy.org/. The present study utilized imaging data collected at the ABCD study’s baseline, 2-year follow-up, 4-year follow-up, and 6-year follow-up. Parent consent and child assent were approved by the central Institutional Review Board at the University of California – San Diego. Study procedures were conducted in accordance with the tenets of the declaration of Helsinki.

Individuals were excluded if they were not able to complete all administered tasks or complete an MRI scan. This excluded individuals who lacked fluency in English or uncorrectable sensory deficits (e.g., legal blindness). Participants were also excluded if they had a psychiatric diagnosis that prevented attendance at regular classes in school, neurological issues (e.g., brain tumor, previous head injury with loss of consciousness > 30 min), or significant perinatal medical issues (e.g., gestational age < 28 weeks, birthweight < 1.2 kg, complications resulting in > 1 month of hospitalization following birth) [28]. Individuals were excluded if they did not pass quality control measures, which included both automated and manual methods such as findings of artifact or file corruption by a radiologist [15]. Given the focus of the current investigation, individuals that were missing pertinent imaging data (see Neuroimaging Data section) or had a task accuracy below 60% (see Emotional N-Back section) were excluded.

Diagnostic groups and cognitive assessments

The ABCD cohort was subdivided into those with a presumed diagnosis of ASD and those with no ASD diagnosis (nASD), as previously reported [17]. The brief social responsiveness scale (SRS) was used to overcome the absence of formal testing or clinical records to confirm a diagnosis of ASD. A T-score above 60 on the SRS has demonstrated good reliability and sensitivity for identifying ASD across various psychopathologies [52]. Participants who had a parent reported diagnosis of ASD and scored below this threshold on the SRS were removed from subsequent analysis to improve validity of diagnostic categories. There were 149 individuals meeting criteria for ASD across all timepoints and had necessary measures available. Representation of individuals at each timepoint varied, depending on completion of imaging that passed quality control (see Table 1).

Table 1.

Demographics for the nASD and ASD groups across all timepoints where imaging was collected. Exact values are only reported where a subgrouping contained 20 or more individuals, reducing risk of identifying individuals and in accordance with the data user agreement for data from the ABCD study to respect participant anonymity [1]

	nASD	ASD	nASD	ASD	nASD	ASD	nASD	ASD
	Baseline		2-year follow-up		4-year follow-up		6-year follow-up
Sex
Female	3742	< 20	2897	< 20	2467	< 20	1656	< 20
Male	3829	87	3236	66	2755	65	1754	48
Race/Ethnicity
Black	924	< 20	758	< 20	727	< 20	352	< 20
Other	500	< 20	341	< 20	311	< 20	173	< 20
White	5325	78	4383	63	3592	55	2530	45
More than one	822	< 20	651	< 20	592	< 20	355	< 20
Income
< 25 k	783	< 20	538	< 20	419	< 20	201	< 20
25 k to 50 k	941	< 20	701	20	555	< 20	313	< 20
50 k to 75 k	965	< 20	795	< 20	558	< 20	306	< 20
75 k to 100 k	1080	< 20	825	< 20	643	< 20	396	< 20
100 k to 200 k	2352	25	2027	21	1749	< 20	1263	20
> 200 k	903	< 20	819	< 20	897	< 20	736	< 20
Don’t know	282	< 20	195	< 20	218	< 20	122	< 20
Decline to answer	265	< 20	233	< 20	182	< 20	73	< 20
Education
Up to high school (No diploma)	268	< 20	232	< 20	218	< 20	246	< 20
High School diploma/GED	564	< 20	464	< 20	435	< 20	246	< 20
Some college	1815	31	1928	30	1562	23	953	< 20
Bachelor's degree	2037	33	1357	< 20	1118	< 20	759	< 20
Graduate school or professional	2887	30	2152	20	1889	21	1327	< 20

Emotional N-Back

The emotional n-back (EN-Back) was used to engage processing of emotional faces across 160 trials [6]. A series of emotional faces (happy, fearful, neutral) were presented to children. There were two variants of the task where children pressed a button indicating a match for the image immediately preceding a given image (0-back) or the image two iterations ago (2-back). Four task blocks of each variant of the task were administered in approximately 5-min fMRI runs. Blocks were composed of trials consisting of a 2 s stimulus (image) followed by a 500-ms fixation cross. Individual faces were presented with only one unique facial expression at a time. 24 unique happy, fearful, and neutral faces were presented (72 in total). Children were familiarized with the task in a practice session prior to administration of the actual task. Similar to prior work on the N-Back task in the ABCD study, participants with a total accuracy below 60% were excluded [56].

Neuroimaging data

In-depth description of image acquisition, processing, and normalization are fully provided elsewhere [34]. Image protocols for the present study include 3D T1-weighted (T1w) acquisition (1 mm isotropic) magnetization-prepared rapid acquisition gradient echo scans, 3D T2-weighted (T2w) acquisition (1 mm isotropic) fast spin echo with variable flip angle scans, and diffusion weighted imaging (DWI) acquisition (1.7 mm isotropic) multiband EPI scans with 96 noncollinear gradient directions (6 directions at b = 500 s/mm², 15 directions at b = 1000 s/mm², 15 directions at b = 2000s/mm², and 60 directions at b = 3000 s/mm²) and seven b = 0 s/mm² frames. fMRI acquisition (2.4 mm isotropic, TR = 800 ms) employed multiband echo planar imaging (EPI). Images underwent standard preprocessing as previously described (including corrections for eddy currents, motion, field inhomogeneities as well as other potential sources of artifact where appropriate) [14], using the FMRIB Software Library (FSL) [36, 65]. Quality control by trained technicians is performed on all images, facilitating removal of data where artifact due to noise could not be appropriately corrected in preprocessing [34].

A restriction spectrum imaging (RSI) model of DWI data was used to characterize neuron density in vivo [75]. RSI provides a multicompartmental model in which a fraction of fluid diffusion is attributed to the intracellular (restricted) and a fraction attributed to extracellular (hindered) compartments of brain tissue. The present work focused on measures of restricted diffusion to better characterize the intracellular compartment, reflecting components of neuron cell bodies. RNI measures were used to characterize neuron cell bodies, where intracellular diffusion of fluid is largely isotropic. RND measures were used to characterize neurite projections from the neuron cell body (e.g., dendrites, axons), where intracellular diffusion of fluid is anisotropic. Therefore, RNI and RND measures are the fraction of fluid diffusion within neuron cell bodies and neurite projections, respectively. As a point of comparison, RND is similar to fractionally anisotropy (FA) derived from classic diffusion tensor imaging (DTI) models, except the use of a model considering multiple B-shells and anatomically inferred tissue composition makes it unaffected by fibers oriented in multiple directions within a voxel [76].

Estimates of functional activity attributed to processing of emotional faces was computed at the individual subject level, contrasting functional activity when observing neutral faces versus happy and fearful faces. This was modeled using a general linear model (GLM) implemented in Analysis of Functional NeuroImages (AFNI) software [19], that modeled activation specific to the EN-Back task cues and trial blocks as a square wave convolved with a two parameter gamma basis function. Additional regressors were used to account for baseline activity throughout task trials, quadratic estimates of nonlinear changes throughout the time-series, and estimates of motion [57]. Time points with significant frame displacement were removed [64] as well as frequencies due to oscillations from respiration and dynamic changes in magnetic susceptibility due to lung movement [24]. The derived outcome measures are the estimated functional activity while processing emotional versus neutral faces (EVNF).

T1w and T2w images were corrected for gradient nonlinearity distortions using scanner specific software [38, 72]. T2w images were registered to T1w image space using mutual information [74], and intensity inhomogeneity was corrected by applying sparse spatial smoothing of smoothly varying, estimated B1-bias fields [3]. Additional standard preprocessing and segmentation was applied to structural imaging using FreeSurfer [26]. Segmentation was based on the Desikan-Killiany brain atlas [45]. Structurally derived atlas-based segmentations were applied to image mapped estimates of RNI, RND, and modeled estimates of emotional-vs-neutral functional, averaging all voxels measures with regions of interest. Specifically, measures of functional activity attributable to processing emotional faces and RSI measures were sampled in the right and left hemisphere for the IFC, fusiform cortex, and amygdala. These regions were selected due to their implication in previous literature for emotional facial processing along with identification in prior investigation of neuron density in the ABCD cohort [17].

Statistical analysis

Mixed effects linear models were used to analyze the relationship between functional activity and neuron density, using covariates such as sex, highest parent education, collection site, or reported use of medications for ADHD (methylphenidate, Ritalin, Adderall, etc.). Continuous measures were centered and standardized across the entire study population before modeling. Statistical modeling was performed in two steps.

First, composition of a null model was derived excluding the variables of interest. This was performed by stepwise inclusion of each covariate into a model of functional activity as the outcome. Each model was compared to a prior iteration via likelihood ratio testing, using the MixedModels package [21]. This has been used in prior studies to ensure that the null model is not over parameterized prior to testing measures of interest [7, 16, 17].

The final null models were of the form:

where EVNF is the outcome variable derived from the GLM model at a particular region of interest, as described in "Neuroimaging data" section, and the right-hand side includes the intercept (i.e. 1) and sex as the only fixed effect terms in the null model. Additional right-hand side terms with the notation indicate inclusion of a random intercept for each group within variable. Groups for demographic variables of interest are available in Table 1.

Second, hypothesis testing was performed by inclusion of the term diagnosis*neuron density, modeling the main effects and interaction between neuron density (the mean RNI or RND within each area of interest) and diagnosis (ASD and nASD groups). The final model these are derived from takes the form:

The addition of the term indicates that each fixed effect term of interest, including interactions, included a random effect for each timepoint. Given each timepoint represents individuals within two-years of age of each other, these additional terms served as a control for age between ASD and nASD as well as controlling for potential age dependent changes in cytoarchitecture [8, 55]. for the final six fitted models are visualized in Fig. 1. Reported results characterize fixed effects using β coefficients and standard errors (SE) from their respective models. Terms comparing ASD and nASD groups used nASD as the reference group. Benjamini–Hochberg correction was used to correct p-values for multiple comparisons with statistical significance determined using an α of 0.05 [9].

Fig. 1.

From top to bottom, rows correspond to the IFC, fusiform, and amygdala with the left and right sides corresponding to the left and right hemisphere. The conditional means across the random effect of timepoint, contributing to a given fixed effect are plotted as red dots with black bars extending by 1 standard deviation in each direction. Fixed effect terms plotted for each region include the intercept for the nASD group (“Intercept”), the intercept for the ASD group (“diagnosis: ASD”), the slope for the nASD group (“RSI” or “RND”), and the slope for the ASD group (“RSI & diagnosis: ASD” or “RND & diagnosis ASD”). Estimates were derived from the random effects for timepoint used in the fitted models predicting functional activity while processing emotional faces vs neutral faces, whose fixed effects are included in the presented results. IFC: For both the left and right IFC, conditional mean effects of the intercept and slope of RND in both ASD and nASD varied from estimates of fixed effects greatest at baseline. However, standard deviation from these estimates at each timepoint overlapped. Fusiform: For the left fusiform, conditional means of the intercept and slope of RND in both ASD and nASD varied greatly from estimated fixed effects at year 4. For the right fusiform, conditional means of the intercept and slope of RND in both ASD and nASD was pronounced at year 2 and year 4. Amygdala: For the left amygdala, conditional means of the intercept increased in the nASD group and decreased in the ASD group across each subsequent timepoint. Estimates of the slope for RNI in the left amygdala increased across each timepoint in both groups. For the right amygdala, conditional mean effects of the intercept and slope of RND in both ASD and nASD varied greatly from estimates of fixed effects at baseline

Results

Functional activity

The ASD group had lower EVNF IFC bilaterally that was not statistically significant (left: β = −0.102119, SE = 0.104189, p-value = 0.3270; right: β = −0.024506, SE = 0.111295, p-value = 0.8257). The ASD group had lower EVNF in the left fusiform (β = −0.0499036, SE = 0.108879, p-value = 0.6467) and higher EVNF in the right fusiform (β = 0.039266, SE = 0.114924, p-value = 0.7326). However, neither were found to be statistically significant changes in EVNF. The ASD group had lower EVNF in the amygdala bilaterally (left: −0.332826, SE = 0.106454, p-value = 0.0018; right: β = −0.0328482, SE = 0.128134, p-value = 0.7977). Therefore, there was a significant reduction in functional activity when processing emotional faces compared to neutral faces in the ASD group.

Sex was a fixed effect in the fitted mixed linear model to account for potential bias due to overrepresentation of males in the sample. When compared to females, males had lower EVNF measures in the left IFC (β = −0.00108806, SE = 0.00467593, p-value = 0.8160) and higher in the right IFC (β = 0.000224535, SE = 0.00478785, p-value = 0.9626). Similarly, males had lower EVNF in the left fusiform (β = −0.0499036, SE = 0.108879, p-value = 0.6467) and higher in the right fusiform (β = 0.00407826, SE = 0.00497943, p-value = 0.4128). Males had higher EVNF bilaterally in the amygdala (left: β = 0.00167443, SE = 0.00530967, p-value = 0.7525; right: β = −0.0328482, SE = 0.128134, p-value = 0.7977). No findings related to sex remained significant when controlling for nuisance variables.

Neuron density and function activity

Although the linear relationship between left IFC RND and EVNF appears to have a negative association when fit to a simple linear model (see Fig. 2 top-left), inclusion of demographic control variables resulted in a non-significant positive relationship in the ASD group (β = 0.13276, SE = 0.499362, p-value = 0.7903). The apparent positive association between left IFC RND and EVNF in the ASD group (see Fig. 2 top-right) was not statistically significant (β = 0.396676, SE = 0.499215, p-value = 0.4268). The negative association between right IFC RND and EVNF (see Fig. 2 top-right) was non-significant in both the nASD group (β = −0.117065, SE = 0.529851, p-value = 0.8251) and the ASD group (β = −0.0186091, SE = 0.529754, p-value = 0.9720).

Fig. 2.

Along the x-axis are raw measures of RNI and RND, reflecting the fraction of DWI signal attributed to neuron cell bodies and dendrites, respectively. Beta values estimating the functional signal attributed to processing of emotional faces contrasted with processing of neutral faces are visualized along the y-axis. For each region of interest, a simple linear regression between neuron density measures and functional activity is plotted immediately left of a violin plot, visualizing the distribution of function measures. Black lines within violin plots represent median values within each population. Blue and yellow are used to represent the ASD and nASD groups, respectively. These relationships represent the raw linear relationship prior to inclusion of hierarchical control variables in the linear model reported in the results. These relationships represent the raw data without controlling for hierarchical control variables included in reported results, with x-axis aligned across each column and y-axis aligned across each row, providing a comparative reference of raw measures. Shaded regions indicate a confidence interval of 95%. All columns are aligned along. Measures greater than 3 standard deviations were not included in the plot so that the majority of data could reasonably viewed at scale. IFC: The ASD group had an increase in RND associated with an increase in functional activity related to emotional processing of faces in the left IFC, whereas the nASD group had a negative association between RND and functional activity. Both groups had a decrease in functional activity related to RND in the right IFC. Fusiform: There was a modest negative association between RND and functional activity in the left fusiform for the nASD group and a modest positive association for the ASD group. There was a modest positive relationship between RND and functional activity in the right fusiform for the nASD group and a negative relationship between the ASD group. Amygdala: There was a positive relationship between RNI and functional activity in the amygdala for both groups. However, this relationship was only modest in the nASD group with a greater correlation in the ASD group, particularly in in the left amygdala

Similar to those findings in the left IFC, the left fusiform appears to have a negative association in the nASD group and a positive association in the ASD group when fit to a simple linear model (see Fig. 2 middle-left). However, once controlling for confounding variables there was only a modest positive association in the nASD group (β = 0.0327871, SE = 0.490262, p-value = 0.9467) and the ASD group (β = 0.12821, SE = 0.487922, p-value = 0.7927). In contrast, the right fusiform had a modest negative association between RND and EVNF in the nASD group (β = −0.226399, SE = 0.514337, p-value = 0.6598) and ASD group (β = −0.337724, SE = 0.514183, p-value = 0.5113), failing to detect an interaction between RND in the fusiform and diagnostic groups.

RNI was used to assess density of neuron cell bodies at the amygdala. There was a positive association between EVNF and density of neuron cell bodies in the amygdala bilaterally (left: β = 1.26596, SE = 0.413497, p-value = 0.0022; right: β = 0.055502, SE = 0.506337, p-value = 0.9127). There was a significant interaction between the ASD group and RNI on EVNF, demonstrating a greater rate of increase in EVNF with increased neuron cell body density in the left amygdala (left: β = 1.23408, SE = 0.412041, p-value = 0.0027; right: β = 0.0524766, SE = 0.506349, p-value = 0.9175).

Discussion

To gain insight into the historically inconsistent functional imaging findings related to the processing of emotional faces in ASD, the present study incorporated recently identified cytoarchitectural measures that differed in those with ASD. The use of imaging from the ABCD study [4, 5, 14, 34], a previously identified population of individuals with ASD, and relevant cytoarchitectural measures [17] allowed for a more focused investigation than previously possible. Specifically, the relationship between RNI in the amygdala, RND in the fusiform, and RND in the IFC was evaluated in the context of processing emotional faces during fMRI scanning. Although these regions have been found to differ functionally in ASD in previous studies [51, 58], the present study did not identify significant findings in the fusiform or IFC. However, the ASD group had a significant reduction in EVN. Furthermore, the ASD group were found to differ in the relationship between EVNF and cytoarchitecture in the left amygdala, potentially demonstrating a differential dependence on cytoarchitecture in ASD for socio-emotional functioning.

The significant reduction in left amygdalar EVNF in the ASD group demonstrates a comparative decrease in activity when processing emotional faces compared to neutral faces. This is inconsistent with prior research where increased functional activity was found in the amygdala while processing emotional faces [51]. The current design had the benefit of collecting imaging data across four timepoints spanning a 6-year period, with multiple trials of the EN-Back task. The significant reduction in functional activity identified in the ASD group potentially represents a more robust characterization of decreased mean functional activity than previous studies. However, functional activity is also dependent on temporal factors such as habituation [46, 68] or functional connectivity [33]. Therefore, additional characterization of functional activity during processing of emotional faces in ASD is necessary to fully elucidate this relationship.

In addition to mean EVNF, the relationship between emotional recognition and the amygdala may be related to changes in structural connectivity throughout development [61]. The present study found a significant positive correlation between EVNF and RNI in the left amygdala as well as a significant interaction in the ASD group, demonstrating that corresponding increases in neuron cell body density and functional activity attributable to processing emotional faces were significantly greater in the ASD group (see Fig. 2 left-bottom). The amygdala is of particular interest in ASD as other studies have posited that ASD is characterized by periods of premature maturation of the amygdala [32, 54]. More broadly, measures of neuron density have been found to vary with age throughout the general population (Qian et al., 2020). Therefore, statistical models fitted diagnostic group, measures of neuron density, and their interaction conditional on timepoint, in addition to other nuisance variable discussed the Statistical analysis section. Random effects for timepoint increased for estimates of fixed effects for RSI in the left amygdala for both the ASD and nASD groups (see Fig. 2). Overlap between predicted intervals across timepoints for these estimates and consistent trends in both groups is reassuring that the ASD group has an actual increased rate of functional activity attributable to processing emotional faces with increases in neuron cell body density in the left amygdala. However, this study only captures 4 time points between the ages of 9 and 16, limiting characterization of developmental changes throughout adolescence. Characterization of cytoarchitecture and task specific functionality across development will require a greater age range and longitudinal sampling to appropriate characterize in future studies.

There were no significant findings specific to the fusiform in the ASD group. This is inconsistent with prior research that found individuals with ASD demonstrated reduced EVNF in the fusiform when processing faces [53]. However, a significant positive correlation between RND and EVNF in the right fusiform for the nASD group was found. The absence of a similar relationship in the ASD group may be due to an overall shift toward lower RND values (see Fig. 2 row 2), requiring a greater sample size to compensate for the absence of higher values driving a correlation. Alternatively, the complex relationship between the fusiform and other structures involved in processing emotional faces may require a more nuanced model informed by more modalities [27].

Prior literature has identified decreased functional activity in the IFC of individuals with ASD when performing tasks that required interpretation of emotional faces [44] or working memory of emotional faces [58]. Although the present study identified decreased functional activity in the IFC of those in the ASD group when processing emotional faces, these findings were not significant when accounting for additional confounding variables. The current study sought to characterize the relationship between each region individually. However, recent analyses have demonstrated that the IFC may play an important role in mediating function of brain regions, particularly in relationship to mental health [49]. Therefore, it is important that future s model the relationship between regions of the brain when processing emotional faces.

A meaningful difference in EVNF activity based on sex was not observed in the present study. These findings may be influenced by the small number of females in the ASD group (less than 20 females identified, see Table 1). This overrepresentation of males is consistent with the sex ratios observed in ASD across the global population [23]. However, males with ASD have been found to have greater connectivity between functionally homotopic voxels in the left fusiform and right inferior frontal gyrus, when compared to females with ASD [48]. Sex-specific differences in brain structure in ASD have also been identified (e.g., brain symmetry [20], frontal and superior temporal cortices [59]). Although, sex-specific differences within the ASD population are frequently observed, the typical course of early neurodevelopment in nASD has been found to differ between males and females [29], with some of these typical developmental differences also identified in ASD [59]. The absence of findings specific to sex are reassuring in that the reported results are not confounded by a disproportionate representation of males in the ASD group but does not constitute a formal characterization of neurophysiological outcomes as a function of sex and ASD. Future investigation will require careful consideration of biopsychosocial factors related to sex and throughout adolescent development.

The ability to account for multiple fluid compartments occupying a common space makes RSI optimal for characterizing regions densely populated by neuron cell bodies. RNI and RND are approximations of neuron cell body and neurite density, respectively. The total neuron density may be approximated from the sum of RNI and RND, which is comparable to the widely used neurite density index (NDI) measure derived from the neurite orientation dispersion and diffusion (NODDI) multicompartment model [55]. Measures derived from multicompartment models of diffusion have demonstrated promise in modeling the association between neurobehavioral outcomes [11] and regional variation in histology [62]. In addition to histological comparison used in the model’s original validation, RSI has demonstrated empiric gains over DTI with in vivo applications such as neurooncology [76]. However, measures derived from RSI and NODDI models cannot simply be interpreted as better DTI measures. Additionally, RSI and NODDI measures are not strictly interchangeable despite being highly correlated [8]. Therefore, discussion of results in terms of neuron density represents the current best understanding of these models but will continue to be refined with ongoing research.

Given the novel investigation of in vivo estimates of neuron density and its association with functional activation during the processing of emotional faces, there are some important limitations to be considered. The current study only considers individuals between 9 and 16 years of age. Although narrowing the age-range avoids obscuring findings with changes unique to specific neurodevelopmental stages in ASD [42], it limits generalizability of findings. Furthermore, statistical models included control variables previously identified in the ABCD study [4, 16, 34], but were unable to explicitly account for comorbidities common to ASD or neuropsychiatric disorders superimposed on ASD. Therefore, future investigation should further explore these findings across a greater span of development while monitoring their stability in cohorts with known comorbidities.

Controlling for environmental variables that can potentially influence neurodevelopment has become an increasing point of concern in research. The effect of environmental variables across time has previously been considered specifically in the ABCD population [30], with additional recommendations for management across the six year timespan [35]. There remains additional complexity when incorporating relatively novel biomarkers, such as those characterizing in vivo neuron density. Although smaller studies often employ case-matching for each ASD participant, large cross-sectional studies must account for multiple collection sites and greater demographic variability. Where age is one of the most critical factors used to match ASD participants, incorporation of study timepoint was used as a proxy for age across effects of interest in statistical modeling, along with inclusion of multiple demographic variables (see "Statistical analysis" section). Although this serves a similar purpose as case-matching by age, it does not explicitly model longitudinal changes. Given neurodevelopmental differences across time are of particular interest in ASD [32, 54], future research will need to continue clarifying the extent to which neurodevelopmental differences are consistent throughout adolescence as well as uniquely identifiable within specific stages of development.

Conclusion

In summary, the present study evaluated functional activity while processing emotional faces while controlling for processing neutral faces, finding a relative decrease in functional activity in the left amygdala of the ASD group compared to the nASD group. An additional interaction between ASD group and RNI was found, demonstrating a unique positive correlation between cell body density in the left amygdala and functional activity in the ASD group. In contrast, the IFC and fusiform had no significant findings between functional activity and the ASD group. Although the lack of findings in other regions is inconsistent with prior investigations, the extent to which these findings may differ in a more heterogeneous representation of ASD is still unknown. Finally, further investigation should be pursued to determine how functional activity in the left amygdala is related to the unique development of cell bodies in ASD and its potential application to clinical interventions.

Abbreviations

ASD

Autism spectrum diagnosis/disorder

nASD

No reported diagnosis of ASD

EN-Back

Emotional n-back

fMRI

Function magnetic resonance imaging

MRI

Magnetic resonance imaging

DWI

Diffusion weighted imaging

IFC

Inferior frontal cortex

RSI

Restriction spectrum imaging

RND

Restricted normalized direction diffusion

RNI

Restricted normalized isotropic diffusion

Authors’ contributions

ZPC performed the primary data analyses and statistical tests, created the illustrations, and wrote the first substantial draft of the manuscript. JJF and EGF are site PIs for the ABCD consortium at the University of Rochester and coordinated data collection and project management. ZPC, JJF and EGF collectively conceived of the study and consulted regularly during the development of the analyses and data representations. JJF and EGF provided editorial input to ZPC during multiple drafts of the manuscript. All authors approve of this final version for publication, attest to the accuracy of the work reported, and agree to be fully accountable for all aspects of the work.

Funding

The ABCD Study® is supported by the National Institutes of Health and additional federal partners under award numbers U01DA041048, U01DA050989, U01DA051016, U01DA041022, U01DA051018, U01DA051037, U01DA050987, U01DA041174, U01DA041106, U01DA041117, U01DA041028, U01DA041134, U01DA050988, U01DA051039, U01DA041156, U01DA041025, U01DA041120, U01DA051038, U01DA041148, U01DA041093, U01DA041089, U24DA041123, U24DA041147. Neuroimaging at the University of Rochester ABCD site is conducted through the Translational Neuroimaging and Neurophysiology Core of the University of Rochester Intellectual and Developmental Disabilities Research Center (UR-IDDRC) which is supported by a center grant from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (P50 HD103536 – to JJF). A full list of supporters is available at https://abcdstudy.org/federal-partners.html. A listing of participating sites and a complete listing of the study investigators can be found at https://abcdstudy.org/consortium_members/. ABCD consortium investigators designed and implemented the study and/or provided data but did not necessarily participate in the analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or ABCD consortium investigators. ZPC’s work on this project was supported by the University of Rochester CTSA award number TL1 TR002000 from the National Center for Advancing Translational Sciences of the National Institutes of Health. Adolescent Brain and Cognitive Development study is a service mark of the U.S. Department of Health and Human Services.

Data availability

The ABCD data repository grows and changes over time. The ABCD data used in this report came from https://doi.org/10.15154/z563-zd24. Code used in analysis is available by request from the authors.

Declarations

Ethics approval and consent to participate

Parental consent and child assent were obtained at each ABCD Study visit. Parent consent and child assent were approved by the central Institutional Review Board at the University of California – San Diego. Study procedures were conducted in accordance with the tenets of the declaration of Helsinki.

Competing interests

The authors declare no competing interests.