Youth Autistic Traits and Brain Activation States During Socioemotional Processing

Abstract

Objective:

Different aspects of social-emotional processing are linked to unique whole-brain patterns of neural activity termed ‘activation states’. Autistic traits (ATs) in youth include impairments in socio-emotional processing that vary depending on age and co-morbidity. Characterizing differences in activation states associated with these impairments may explain why youth with high ATs process socio-emotional information differently than peers.

Methods:

This study examined the frequency of three previously identified activation states in 545 youth from the Healthy Brain Network (HBN) study as they watched a socially engaging movie (5-15 years, 39% female, 54% White, 68% non-Hispanic). Regressions examined associations between ATs (measured by the SRS-2) and overall time spent in activation states, and whether these associations varied with age or co-occurring ADHD and anxiety symptoms (measured by the SWAN and SCARED).

Results:

At younger ages (<9 years), children with more ATs spent more time in an activation state associated increased activation in somatomotor and visual networks. At older ages (> 14 years) children with more ATs spent more time in an activation state characterized by greater default mode and ventral attention network activation. Across all ages, children with more ADHD symptoms spent less time in an activation state associated with greater cingulo-opercular network activation.

Conclusions:

ATs are associated with differences in brain activation during socio-emotional processing that vary with age and co-occurring mental health symptoms. Youth with more ATs may experience an altered developmental trajectory of social-emotional processing compared to peers with fewer ATs. Results inform personalized interventions and progress monitoring for youth with social impairments.

Introduction

Autism spectrum disorder (ASD) includes impairments in socio-emotional processing that vary with age and co-occurring symptoms, yet the brain bases of these impairments are incompletely understood. Brain activity during socio-emotional processing can be characterized as entering into a limited set of “activation states”, specific patterns of increases and decreases in activity across the brain (1,2). Individuals vary in their tendency to enter specific brain activation states, likely reflecting individual differences in cognitive and emotional processing. Yet. it is not known how autistic traits (ATs, behaviors and personal qualities that are consistent with the phenotype of autism and are continuously distributed in the population even at sub-clinical levels), relate to individual differences in how much time is spent in different activation states, or how these relations vary with age or co-morbidity. Clarifying these relationships may identify mechanisms of impaired socio-emotional processing over different phases of development in children with ATs and provide new targets for treatments.

At a macroscale, whole-brain activity tends to fluctuate between distinct activation states during real-world socio-emotional processing. Each activation state includes increases in regional brain activity and is thought to reflect a particular type of socio-emotional or cognitive processing. For example, when viewing emotionally-intense content, individuals tend to shift into an activation state that includes increased activity in the default mode network (DMN) (3); when viewing scenes with more spoken dialogue, most individuals enter into an activation state that includes auditory and language networks (4). However, individuals can vary in when they enter into and transition between specific brain states. Furthermore, this variation aligns with individual differences in subjective experiences (2,4), and with objective changes in arousal (4,5). Examining patterns of activation states in individuals with elevated rates of psychopathology may therefore help uncover mechanisms of differences in socio-emotional processing.

Children with ASD and elevated ATs differ in behaviors and brain activity linked to socio-emotional processing compared to other children. ASD affects social communication and functioning, with early emerging (e.g., (6, 7)) and persistent differences in social visual attention and cognition, potentially influencing emergence of later symptoms, and altering or limiting trajectories of social attention and functioning later in development (i.e., canalization) (8). Neural differences, such as deviation in resting-state functional connectivity, emerge early (9, 10) with connectivity differences during social tasks identified in childhood and adolescence (11). The relations between ASD and brain and behavior, however, depend on co-morbid psychopathology (e.g., (12,13)) and developmental context. For example, attention deficit hyperactivity disorder (ADHD) and anxiety co-occur in ASD in 26-48% and 11-40% of cases, respectively (14–17), and may influence social attention and its underlying neural processes.

Variation in brain activation states elicited in different contexts may explain differences in socio-emotional processing linked to ATs. These processing differences may vary with age and co-occurring symptomatology. In our prior work, we examined brain activation states in children ages 5-15 years from the Healthy Brain Network (HBN) cohort as they watched the movie Despicable Me, finding three unique activation states. We identified that children with high generalized anxiety symptoms were more likely than peers to enter into a state that included high activity in the ventral attention (VAN) and default mode network (DMN), an activation state we theorized to be involved in processing emotionally salient and negatively-valenced stimuli (18). We proposed that increased time in this brain state may contribute to hypervigilance in children with high anxiety and help explain cascading biases in information processing and subjective experience. Clarifying the activation state patterns of children with elevated ATs might similarly help explain how and why socio-emotional processing is impacted in children with elevated ATs, and how this varies with age and co-occurring symptoms. Such information would provide clear new targets for interventions, as activation states could be modified in a context-specific manner through therapy, cognitive training, or non-invasive neurostimulation.

The goals of the current study were to investigate the relationship between ATs and the distribution of participants’ time across three previously identified brain activation states during the viewing of a socially engaging movie. To achieve this aim, we built on our prior work described above, which had examined brain activation states as children (ages 5-15 years) in the HBN watched a movie. We examined associations between ATs and overall time spent in activation states, as well as whether the relations between activation states and ATs varied with age. Given the relatively high rates of psychopathology (e.g., 54.8% ADHD, 45.4% anxiety, 13.8% ASD (19)) and co-morbidities in the sample, we also assessed variation by co-occurring ADHD and anxiety symptoms. Results clarify how children with autistic symptoms process socio-emotional information over development and suggest new avenues for developing brainbased treatments.

Methods

Participants

Participants represent a sub-set of youth who participated in the HBN study (20) in the New York City metro area. Data were collected between Winter 2016 and Winter 2020, with all participants providing assent and caregivers providing consent. All procedures were approved by relevant site IRB Boards.

Brain activation state data derived from high quality fMRI data as described in (18) was available for 620 youth from the HBN’s first nine data releases (Initial Imaging Sample, Table 1). Final analyses included 545 5-to-15-year-olds who also had Social Responsiveness Scale, 2^nd edition (SRS) parent-report data (Final Imaging and Behavioral Sample, Table 1). Participants with SRS data were older than participants without SRS data (10.5 years vs. 9.71 years, p = .025). The two groups did not vary in time spent in any brain activation state (ps > .45), nor in ADHD or anxiety symptoms (ps > .61). See Table 1 for participant demographic information, and Supplementary Material for additional details regarding participant ethnicity (Table S2), pubertal stage (Table S3), and caregiver-reported medication use during fMRI collection (Table S4).

Table 1.

Descriptives for brain activation state and final analysis samples.

	Initial Imaging Sample (N = 620)		Final Imaging and Behavioral Sample (N = 545)
Measure	N	%	N	%
Female	251	40%	210	38.5%
Right-handed	465	75%	409	75.1%
Household income
Less than $10,000	13	2%	12	2.2%
$10,000 to $19,999	13	2%	13	2.4%
$20,000 to $29,999	19	3%	18	3.5%
$30,000 to $39,999	28	5%	25	4.6%
$40,000 to $49,999	20	3%	20	3.7%
$50,000 to $59,999	11	2%	10	1.8%
$60,000 to $69,999	27	4%	25	4.6%
$70,000 to $79,999	13	2%	12	2.2%
$80,000 to $89,999	21	3%	21	3.9%
$90,000 to $99,999	22	4%	21	3.9%
$100,000 to $149,999	101	16%	98	18.0%
$150,000 or more	182	29%	182	33.4%
Choose not to disclose/ No income data	150	24%	88	16.1%
Race
Asian	15	2.4%	15	2.8%
Black/ African American	67	10.8%	64	10.3%
Hispanic or Latino	50	8.1%	47	8.6%
Native Hawaiian/ Other Pacific Islander	1	0.2%	1	0.2%
White/ Caucasian	298	48.1%	293	53.8%
Other	9	1.5%	9	1.7%
Two or more races	105	16.9%	102	18.7%
Unknown/ No race information available	75	12.1%	14	2.6%
Ethnicity
Hispanic or Latino	137	22.1%	132	24.2%
Not Hispanic or Latino	382	61.6%	373	68.4%
Decline to specify	30	4.8%	30	5.5%
No ethnicity information available	71	11.5%	10	1.8%
Measure	Mean	SD	Mean	SD
Age (years)	10.4	(2.81)	10.5	(2.78)
Psychopathology
SRS parent-report total t-score (n = 545)			56.3	(10.9)
SWAN average score (ns = 551 and 541)	0.37	(0.96)	0.37	(0.96)
SCARED self-report total score (n = 422)			21.9	(15.9)
Percentage of retained frames	97.2	(4.72)	97.3	(4.68)
Mean frame displacement (of retained frames)	0.360	(0.184)	0.357	(0.185)
Percentage of time in State 1	38.5	(6.68)	38.5	(6.63)
Percentage of time in State 2	26.4	(7.08)	26.4	(7.07)
Percentage of time in State 3	32.4	(3.99)	32.4	(3.95)

Note: Participants reporting ‘Hispanic or Latino’ race all reported “Hispanic or Latino” as ethnicity as well. ‘Hispanic or Latino’ was included in both racial and ethnic counts to reflect how information was reported by participants at intake.

Note: participant sex was dichotomously coded as “male” or “female” within HBN data.

Note: SRS = Social Responsiveness Scale, 2^nd edition, parent report; SWAN = Strengths and Weaknesses Assessment of ADHD and Normal Behavior; SCARED = Screen for Child Anxiety Related Disorders.

Measures

Autistic traits (ATs) were measured with total t-score from the parent-report version of the Social Responsiveness Scale, 2^nd Edition (SRS (21)), which includes items measuring social communication, cognition, awareness, motivation, and restricted/ repetitive behaviors (RRBs). It has been validated in children ages 4-18 years. ADHD symptoms were measured with the average parent-report score from the Strengths and Weaknesses Assessment of ADHD and Normal Behavior (SWAN (22)). Anxiety was measured with the total self-report (SR) score from the Screen for Child Anxiety Related Disorders (SCARED (23)), which was completed by children 7.7 years old and up (n = 422). Scores of 25 and greater may indicate an anxiety diagnosis. Self-report was specifically used for anxiety symptoms given past work suggesting stronger associations between naturalistic anxiety behaviors and child self-report than parent-report (24) and possible caregiver under-report of children’s internalizing symptoms specifically (e.g., (25), though see Tables S17–S19 for sensitivity analyses with parent-report of anxiety symptoms). All continuous measures were mean centered and z-scored for statistical analyses.

Clinician consensus diagnostic information was available for 367 of the 545 participants, and is summarized in the Supplementary Material, along with caregiver-reported participant diagnoses at study enrollment and parent-reported educational eligibility (Table S5). The diagnostic process in HBN (described in the HBN protocol paper (20) and additional documentation available via the HBN data portal (19)) integrates information from available questionnaires, assessments, and diagnostic interview with a clinician. See Supplementary Material for participant diagnoses as determined by the HBN diagnostic team (Table S5) and sub-scale scores for all measures (Table S6).

Brain activation states

Brain activation states were derived from time series of functional MRI data collected during passive viewing of a movie in-scanner via Hidden Markov modeling (HMM) approaches, as described in (18). fMRI data collection took place while participants watched a 10-minute clip from the movie Despicable Me, which includes scenes such as a tea party between Gru and the girls, minions engaging in hijinks, an argument between Gru and Dr. Nefario, and Gru preparing to steal the moon.

Analyses used data from two HBN MRI sites, CitiGroup Cornell Brain Imaging Center and the Rutgers University Brain Imaging Center. Only participants with at least 80% low-motion data (frames <0.9 FD) were retained in analyses, a standard threshold for task-based fMRI analyses which has been demonstrated to balance maximizing power while minimizing noise from motion (26).

As described in (18), HMM approaches suggested that a 3-activation state model fit the data best based on AIC, BIC, and permutation-based p-values. States were characterized by relatively higher activation in specific networks. Briefly, State 1 was associated with higher activation in somatomotor and visual networks, and lower activation in cingulo-opercular (CON), default mode (DMN), dorsal attention (DAN), frontoparietal (FPN), ventral attention (VAN), and salience networks. State 2 was associated with higher activation in auditory, CON, somatomotor, and visual networks, with lower activation in DMN, DAN, FPN, salience, and VAN. State 3 was associated with higher activation in DMN, DAN, FPN, salience, and VAN networks, and lower activation in auditory, CON, somatomotor, and visual networks.

Activation states were assigned to each frame of each participant’s timeseries based on this 3-state model (see Figure S1 for examples of frame-level data, and Figure S2 for comparison of state assignment posterior probabilities as a function of ATs), and total time in each state (excluding high-motion volumes) was calculated for each participant. See Figure S2 for age-related differences in percentage of time participants spent in each activation state. As described in (18) and presented in Table S1, variables such as scene loudness, brightness, and emotional content were associated with participants’ likelihood of being in each activation state. Figure 1 also includes representative examples of frames when most participants are in each activation state. Supplemental analyses indicated that ATs did not relate to participants’ proportions of high-confidence frames of data (i.e., a probability <.05 or > .95 of being in a given state, p = .54, Figure S3).

Figure 1.

Network activation and movie features associated with three brain activation states.

Analytic Approach

For each of the three brain activation states, a similar series of analyses was performed. Five different models, representing five different potential theories to explain variation in the time spent in each activation state, were calculated. They included:

1. A baseline model controlling for participant age, sex, and level of motion (as measured by mean frame displacement). This model is consistent with the theory that youth psychopathology and ATs are not associated with differences in activation state occupation.

2. A model with all covariates from Model 1, in addition to a main effect of SRS t-score. This model is consistent with the theory that youth with higher ATs show consistent differences in state occupation across development

3. A model with an SRS x participant age interaction term added to Model 2. This model is consistent with the theory that the effects of ATs on state occupation vary across development.

4. A model with the main effect of SWAN scores added to Model 3. This model is consistent with the theory that ADHD symptoms help explain variation in state occupation (in addition to developmental effects of ATs).

5. A model with the main effect of SCARED scores added to Model 3. This model is consistent with the theory that anxiety symptoms help explain variation in state occupation (in addition to developmental effects of ATs).

We aimed to identify the model with the best fit and most plausibility using AICc (Akaike Information Criterion Corrected (27), which is better for quantifying model fits with smaller samples), and R². Differences in AICc (ΔAICc) of 2 or more indicate a significant difference in the plausibility of models (28). However, AICc can only compare fit of models across the same sample and sample size, and including variables of ADHD and anxiety symptoms in Models 4 and 5 reduced the available sample (to n = 541 and n = 422, respectively).

To avoid this reduction in sample size where possible, we elected to use p-values of our effects of interest (i.e., SRS, SRS x age, SWAN, and SCARED) in the five initial models above to identify a subset of “significant models” for each of the three activation states (Tables S8, S10, and S13). For each state, models identified as ‘significant’ were then fit in the same sample (reducing the sample size if Models 4 or 5 were chosen), and AICc values were then used to determine which ‘significant model’ most plausibly explained overall time in each activation state (Tables S9, S11, and S14). AICc weights indicate the probability that a given model is the best explanation for the given data. FDR was not performed in the initial assessment of model significance pre-AICc comparison, but see Tables S8, S10, and S13 for examination of significance after Benjamini-Hochberg (29) FDR correction.

Where appropriate, Johnson-Neyman intervals, which determine the moderator value at which a predictor become significant (30), were calculated for any relevant SRS x age interactions using the interactions R package (31). To keep models as parsimonious as possible, data collection site was not included as a predictor variable. Sensitivity analyses suggested that its inclusion in final models did not significantly alter effect sizes of variables of interest (i.e., age, ATs, and ADHD and anxiety symptoms, Figure S4).

To complement our main analyses, exploratory post-hoc regressions assessed associations between ATs inter-state transition rates at participant level and in frame-level data and (see Supplementary Material, Figures S5–S7). Supplementary analyses also examined the effect of categorical ASD diagnosis in a subsample of participants who had full diagnostic outcome information from HBN clinicians (n = 357; Figures S8–S13).

Results

Associations among autistic traits, ADHD symptoms, anxiety symptoms, and age

Autistic traits (ATs) measured with the SRS were linked to increased ADHD symptoms (SWAN, r(543) = .52, p < .001) and increased anxiety symptoms (SCARED, r(420) = .14, p = .003) but ATs were not linked to participant age (p = .23) or sex (p = .33). ADHD symptoms were lower in children that were older (r(539) = -.102, p = .017) and higher in males (r(539) = 0.201, p < .001). Anxiety symptoms were lower in males (r(420) = .155, p = .001). ADHD and anxiety symptoms were not correlated, p = .60. See Table S7 for pairwise correlations between symptoms and demographic variables, and diagnostic rates in sample pre-enrollment and post-study diagnostic process. See Figure S2 for a visualization of age-related differences in percentage of time in each state as a function of participant ATs.

Associations between autistic traits and total time in each brain state

State 1

State 1 is associated with higher activation in somatomotor and visual networks and is more likely to be elicited during scenes that are brighter, louder, and have fewer positive emotions. In comparing among models, the model with an SRS x age interaction (Model 3) was marginally more plausible than the baseline model (AICc weights of .588 vs .412, respectively) and explained more variance in R² (0.014 vs. 0.006; Table S9). In this model, children with more ATs spent more time in State 1 at younger ages, as indicated by a significant SRS x age interaction (β = −0.085, p = .049). For youth younger than 8.8 years old, ATs were significantly associated with more time in State 1 (Figure 2, Table S8). There was no main effect of ATs on time spent in State 1, indicating that relations between ATs and State 1 varied with development.

Figure 2.

More autistic traits are associated with more total time in State 1 in younger participants (β = −0.085, p = .049). Participants younger than 8.8 years old (non-gray region of plot) with higher SRS scores (more autistic traits; solid line) spend more time in State 1 (greater activation in somatomotor and visual networks) than their peers with lower SRS scores (dashed line).

Neither ADHD nor anxiety helped explain variation in total time spent in State 1 (ps > .18, Models 4 and 5). Supplementary post-hoc frame-level analyses suggested that scenes with significant age x SRS effects often included highly complex emotions and facial expressions, or scenes with high visual salience such as bright colors and glowing lights (Figure S5).

State 2

State 2 is associated with higher activation in auditory, CON, somatomotor, and visual networks, and is elicited during parts of the movie with less negative emotion and more motion, louder audio volume, and higher visual saliency. The most plausible model included the main effect of ADHD (Model 4, AICc weight = .746). It also explained the most variation based on R² values (0.099, Table S11). ADHD symptoms helped explain variance in time spent in State 2, β = −0.104, p = .036, such that participants with more ADHD symptoms spent less time in State 2 overall (Figure 3, Table S10). In supplemental analyses using diagnostic categories rather than dimensional symptom measures, ADHD diagnosis showed a similar effect on State 2 occupation as ADHD symptoms (SWAN average score), such that participants with ADHD spent less time in State 2 than participants without ADHD (Figures S10 and S11).

Figure 3.

A higher SWAN average score is associated with a smaller percentage of time in State 2 (β = − 0.104, P = .036). Participants with more ADHD symptoms spend less time in State 2 (greater activation in the cingulo-opercular, auditory, somatomotor, and visual networks).

ATs were not associated with percentage of time spent in State 2 (p = .051) in Model 2 (Table S10). The SRS x age interaction was also not significant, p = .83 (Model 3), though the main effect of age was significant in all models (βs = .142-.189, ps < .001), such that older participants spent more time in State 2 than younger participants. Supplementary post-hoc frame-level analyses suggested in some scenes with high visual salience (where younger youth with higher ATs were more likely to be in State 1), older youth with higher ATs were more likely to be in State 2 (Figure S6).

State 3

State 3 is associated with higher activation in DMN, DAN, FPN, salience, and VAN networks, and was more likely to be elicited during parts of the video with lower audio volume, more negative emotions, and less motion. Of the pre-planned analyses, the model with an SRS x age interaction provided the most plausible model for the data (Model 3, AICc weight = 0.473) and explained the most variation based on R² values (0.052, Table S14). In this model, children with more ATs spent more time in State 3 at older ages, as indicated by a significant SRS x age interaction (β = 0.087, p = .041; Table S13). Participants 14.3 years and older with more ATs spent more time in State 3 than their peers with fewer ATs (Figure 4). The main effect of ATs was not significant. (ps > .54).

Figure 4.

Higher SRS total t score (more autistic traits) is associated with more time in state 3 (greater activation in the default mode, dorsal attention, fronto-parietal, salience, and ventral attention networks), but only in older youths (β=0.087, , p=0.042). Youths 14.3 years old and older (non-gray region of plot) with more autistic traits (solid line) spend more time in state 3 than their peers with fewer autistic traits (dashed line). AUD, auditory network; CON, cingulo-opercular network; DMN, default mode network; DAN, dorsal attention network; FPN, frontoparietal network; SMD, dorsal somatomotor network; SML, lateral somatomotor network; SAL, salience network; SRS, Social Responsiveness Scale, 2nd ed.; VAN, ventral attention network; VIS, visual network.

The main effects of anxiety and ADHD symptoms did not help explain percent of time spent in State 3 (ps > .33), though supplemental analyses confirmed that significant effects of generalized anxiety identified in (18) persisted in a model with the SRS x age interaction and suggested a most plausible model with both significant main effects (Tables S15 and S16). Supplementary post-hoc frame-level analyses suggested that age x AT effects were most significant during emotionally complex or intense scenes (Figure S7).

Discussion

Summary of findings

Results from the current study reveal that autistic traits (ATs) are associated with age- and co-morbidity-dependent differences in brain activation states linked to socio-emotional processing. At younger ages (<9 years), children with more ATs spend more time in an activation state with increased activation in the somatomotor and visual networks, which may be associated with somatomotor processing and integration (State 1). Across all ages, children with more ADHD symptoms (commonly co-morbid with ATs) spend less time in a state with increased activation in the CON, which may be associated with attention shifting and maintenance (State 2). At older ages (> 14 years), children with more ATs spend more time in an activation state with increased activation in the DAN and VAN, which may be associated with social processing and stimulus-driven attention (State 3).

Developmental cascades

This overall pattern of age-related results suggests that children high in ATs may follow a specific trajectory of socio-emotional information processing over development that differs from children lower in ATs. The specific differences in socio-emotional processing vary by age, suggesting that elevated ATs may be associated with an altered neurodevelopmental cascade that relates to age-specific challenges and enduring functional differences.

Before 9 years, children with more ATs are more likely to be in a state we theorize to be related to somatomotor processing and integration (State 1). This is consistent with previous work suggesting that, even in infancy, autistic individuals attend to low-level salience in their visual environment over socially relevant content (6, 27), with differences persisting into childhood (33) and cascading across development, impacting social functioning (34).

Across all ages in the current study, children with more ADHD symptoms are less likely to be in a state we theorize may be related to maintenance of focus and arousal (State 2). Given the high rates of ADHD in children with elevated ATs, many children with autism would thus experience reduced time in State 2 relative to peers. ADHD symptoms predicting decreased time in this state is consistent with the high rates of autistic children presenting with inattention, hyperactivity, and executive function struggles (35).

At 14 and older, children with more ATs spend more time in an activation state we theorize may be related to social processing and stimulus-driven attention (State 3). This suggests that in the teenage years, ‘decoding’ social interactions maybe more effortful for youth with elevated ATs than their peers. The more exaggerated social cues in animated interactions may allow older youth with more ATs to more readily engage with social-emotional content that is less subtle – but potentially less engaging for their peers with fewer ATs. These results are also consistent with previous research identifying different connectivity and activation patterns in autistic adolescents specifically during social processing tasks (11).

Overall, the current results build upon previous findings and describe a cascade-like developmental trajectory of differences in socio-emotional processing linked to ATs. Interestingly, the age-dependent effects in these results mirror varied presentations and complaints related to ASD across childhood and adolescence (36), providing additional face validity. Early concerns about ASD often arise based on differences in eye contact, joint attention, and repetitive gross motor movements such as rocking and flapping (37) (potentially related to the visual and somatomotor networks prominent in State 1). Similarly, autistic and neurodiverse adolescents are likely to report struggles with processing and responding to complex social cues (associated with the DMN and State 3) as a major struggle motivating an ASD assessment and diagnosis (38). Across childhood and adolescence, autistic individuals often present with concerns about attention, hyperactivity, and executive function, highlighting how both ASD and ADHD symptoms can impact an individual’s attention especially in a social context (related to the CON, prevalent in State 2). It will be important to replicate the present findings in longitudinal samples to validate that age-related differences are also present within individuals sampled across development.

Consideration of common ASD comorbidities and overlapping biological etiology

More broadly, examination of co-occurring symptomatology is not only important in a sample like the HBN, with elevated prevalence rates for most diagnoses (e.g., 54.8% ADHD, 45.4% anxiety, 13.8% ASD, (19)), but for understanding the effects of ASD generally. Diagnoses commonly co-presenting with ASD such as ADHD and anxiety further exacerbate qualitative differences in how autistic youth perceive, process, and experience social contexts (39) and worsen social and functional outcomes (40). ADHD, ASD, and anxiety co-occur frequently and are all implicated in early and persistent differences in neurodevelopment in overlapping functional brain networks, such as the DMN (41–43), suggesting shared neurobiological etiology. Taking ASD out of the context of these other symptoms may lead to a fundamental misunderstanding of the biological and neural mechanisms underlying its etiology and presentation, as exemplified in the findings of ADHD symptoms predicting time in State 2. Similarly, longitudinal study may be important to identifying both separable effects of ATs, anxiety, and ADHD on network activation as well as where their overlapping symptomatology creates disorder non-specific impacts on processing and resulting functioning.

Developmentally tailoring interventions

While the presented analyses utilize cross-sectional data, they identify specific variations and shifts in developmental trajectories of brain activation associated with elevated ATs, which can inform more individually tailored and developmentally appropriate interventions. Younger children may benefit more from interventions that increase motivation and interest in the social aspects of the environment (i.e., decreasing time in a state related to sensory aspects including somatomotor and visual, State 1), thereby improving opportunities to gather important social information to aid in social cognition and functioning. Older youth may benefit more from interventions that support ‘decoding’ layers of social information and meaning, thereby building skills to support more complex social interactions and functioning, which may potentially decrease time spent in a state with higher activation of the VAN and DMN, State 3. Youth of all ages can benefit from interventions (whether behavioral, pharmacological, or other) that improve the maintenance of attention to the social content of their environment (i.e., increasing time in State 2, which is associated with increased activation of the CON). Activation states could also be used to personalize and monitor interventions, such as identifying if an individual tends to be in a less-optimal activation state during specific aspects of social engagement and interaction (e.g., maintaining attention, ‘deciphering’ emotional expressions), or identify ‘phenotypes’ of activation that cut across DSM diagnostic categories (44) and provide more straightforward treatment targets.

Consistency in identified brain states across samples and methods

Others have identified similar activation states in fMRI data outside of HBN. Using adult task-based fMRI data from the Human Connectome Project (HCP), a prior study (45) derived four unique states: low cognition (which resembles our State 1), fixation (which resembles our State 3), and two states, high cognition and cue/ transition which resemble our State 2. The study found good fit for these four states in HBN data, as well as in another cross-sectional dataset of youth 8-21 with various clinical and cognitive profiles ((46)). States also fit well in longitudinal fMRI datasets ((47)), including when enriched for risk for substance use and spanning into early adulthood ((48)). Overall, this work supports the idea that brain states are replicable across datasets and different developmental periods (5-29 years). The authors (45) also identified associations between variability in state occupation and executive function, suggesting other metrics relating to brain activations states can be derived and linked to clinical and cognitive outcomes in future work. Both we and Ye and colleagues found good fit for our activation states in samples that include diverse phenotypes and clinical profiles. However, as research utilizing brain activation states progresses, it will be important to balance group-level characterization and inter-individual variability in brain activation patterns and how they may relate to socio-emotional experiences or clinically relevant symptomatology. It will also be important to validate any interpretation of a given state (e.g., “concentration” or “low cognition”) with objective measures.

Limitations and future directions

The HMM-derived activation states used in the present analyses were derived from a larger sample of pediatric data, representing a group average of participants’ brain activation. Even if a child with elevated ATs is in the same state as a child with fewer ATs concurrently, we cannot know what specific aspects of the scene (e.g., visual saliency, emotional content) influenced each child’s patterns of network co-activation. While frame-level analyses can identify specific scenes in which youth with more ATs were more or less likely to be in a given activation state (see Figures S5–S7), we still cannot know that all higher-AT youth were attending to the same aspects of a given scene to elicit that variation in activation states. Future work utilizing in-scanner eye-tracking can improve our knowledge of what emotional or social aspects of a given scene relate to variation in activation state occupation, and if this may change with development. Additionally, while these analyses investigated a wide age range, cross-sectional data limits our ability to falsify hypotheses related to potential developmental cascades associated with the AT x age interactions we identified. Future longitudinal data collection and analyses are required to confirm these theories.

Similarly, our analyses can only speak to how youth perceive a socially and emotionally engaged movie viewed in an MRI scanner. Animated movies have many advantages compared to static face stimuli, but are still removed from the demands of processing in-vivo social stimuli embedded in the rich sensory and emotional context of the real world. Animated movies often utilize dramatic facial expressions and other social cues to clearly convey social meaning to viewers. Additionally, other content such as music and sound effects (e.g., high-pitched strings to build tension) or lighting and coloring choices (e.g., a scene darkens to convey a character’s negative emotion) further broadcast the emotions that filmmakers aim to evoke for the viewer. Movie viewing in-scanner allows for viewing emotional and social content in the context of a movie and its plot, and as such represents a drastic improvement in many ways compared to static stimuli. However, it still differs meaningfully from real-world, first-person experiences in meaningful ways. Future work should endeavor to assess socio-emotional processing and its associations with ATs and other pediatric psychopathology in more naturalistic contexts.

While current analyses focused on dimensional ATs, supplemental analyses using categorical diagnoses in a subsample of participants (n = 357) did not find similar effects of ASD on activation states (Figures S8–S13). Investigation of categorical ASD diagnosis in this sample are also limited by the fact that HBN recruitment excluded non-verbal children and children with IQs <66 (who are disproportionately more likely to have an ASD diagnosis (20)), and children with low-functioning ASD. Future work can aim to replicate the activation states identified by (18) in a sample specifically recruited for meeting ASD diagnostic criteria. Additionally, while the current analyses were not pre-registered, pre-registration of future analysis can strengthen reproducibility and support transparency more broadly in neuroimaging and developmental neuroscience research.

This study adds to our understanding of the brain activation underlying socio-emotional processing in youth with elevated ATs, within the context of development and common cooccurring symptomatology. ATs influence total time spent in all three brain activation states assessed, with age-dependent effects related to states associated with visual and somatomotor networks in younger participants (State 1) and DMN and VAN in older participants (State 3). Analyses also highlight how accounting for co-occurring symptoms such as ADHD and anxiety provide important context for the effects of ATs, such as decreased time in a state associated with the CON and attention and arousal (State 2). Collectively, findings related to all three states illustrate the possible cascade-like impacts of differences in brain function and connectivity on social processing and functioning for individuals with elevated ATs across childhood and adolescence. These findings highlight the utility of brain activation states to identify and personalize intervention targets, as well as to monitor treatment progress in youth with impairments in social functioning.