Autism spectrum disorder (ASD) is a neurodevelopmental disorder affecting approximately 1 in 44 children in the United States. Characterized by social communication deficits and repetitive behaviors, ASD presents significant clinical heterogeneity. While some toddlers with ASD experience improvement in language and social skills over time, others do not, and this variability is not easily predicted using early clinical assessments alone. This lack of predictive power highlights the need for biomarkers that can identify toddlers at risk for poor outcomes and inform early intervention strategies. Previous research has suggested brain overgrowth in young children with ASD, particularly in frontal and temporal regions. However, findings regarding non-cortical regions have been inconsistent, possibly due to variations in study methodologies and participant characteristics. Many studies focused on global measures or single regions, limiting the understanding of complex regional brain alterations in ASD. A recent study by the authors found atypical anterior-posterior and dorsal-ventral genomic cortical patterning in ASD toddlers with cortical enlargement and poor language outcomes, indicating potential regional differences in brain development. Furthermore, previous research has shown correlations between brain size alterations (particularly in frontal and temporal regions) and language/social deficits and facial recognition impairments in ASD, although many of these studies utilized cross-sectional designs and older samples. This study aimed to address these gaps by comprehensively examining regional brain differences in a large sample of toddlers with ASD, compared to typically developing (TD) toddlers, at the age of initial diagnosis and subsequently analyzing the relationship between these structural alterations and future language abilities. Specifically, the researchers sought to determine whether combining structural MRI (sMRI) data with clinical and demographic measures improved the accuracy of predicting language outcomes compared to using clinical data alone. Additionally, the study investigated the correlations between variations in brain structure within specific regions (temporal, fusiform, and frontal) and the severity of ASD symptoms and cognitive abilities in the autistic toddlers.

Converging evidence from neuroanatomical studies suggests brain or cortical overgrowth in young children with ASD, particularly in frontal and temporal regions. However, non-cortical brain regions show inconsistent patterns of alteration. The inconsistency may be due to differences in cohort characteristics, MRI scanners, preprocessing pipelines, and analytical methodologies. Most studies focused on global measures or single regions (amygdala, cerebellum, corpus callosum) and single morphometries (volume, surface area, cortical thickness). One recent study by the authors found atypical anterior-posterior and dorsal-ventral genomic cortical patterning in ASD toddlers with cortical enlargement and poor language outcome. A review of ASD genetic, postmortem, and animal and cell models suggested ASD involves progressive disruption of many different prenatal stages in cortical regional growth. While overall ASD cortex may be enlarged, genetic dysregulation could lead to diverse deviances in volume, surface area, and thickness across cortical regions. Brain size alterations have been widely reported to underlie language, social deficits, and facial recognition impairment in ASD. For example, volumes in frontal and temporal regions were related to repetitive behavior and social and communication deficits. Studies have found increased temporal cortical thickness in ASD with lower intellectual ability and more severe symptoms. Temporal cortex dysfunction in response to social language has been observed in ASD toddlers. Beyond identifying early-age anatomic growth abnormalities, a key challenge is discovering anatomic predictors of heterogeneous developmental outcomes in ASD. It's unclear whether structural alterations at first clinical detection can predict prognosis trajectories.

This study involved two samples: a main sample and a replication sample. **Main Sample:** 275 toddlers (166 with ASD, 109 TD) were recruited through community referrals or a population-based screening. Toddlers underwent a series of clinical and behavioral assessments at each visit, including the Autism Diagnostic Observation Schedule (ADOS), the Mullen Scales of Early Learning, and the Vineland Adaptive Behavior Scales. Structural MRI (sMRI) scans were collected on a 1.5 T General Electric MRI scanner during natural sleep. Scans were parcellated using FreeSurfer 5.3 based on the Desikan-Killiany atlas to obtain global and regional brain morphometric measures (total brain volume, surface area, cortical thickness, etc.). **Replication Sample:** 75 toddlers (38 ASD, 37 TD) from a previous study were used for replication. sMRI scans were acquired using a 1.5 T Siemens Symphony system and preprocessed with FreeSurfer 5.3 using the same pipeline as the main sample. **Analysis:** Linear mixed-effects models (LMEMs) were used to examine ASD vs. TD differences in global and regional brain size, adjusting for age, sex, and global brain measurements. A false discovery rate (FDR) correction was applied to account for multiple comparisons. To predict language outcome, a support vector machine (SVM) with ridge regularization was used. Three models were tested: clinical/demographic only, sMRI only, and a combined model. Linear regression was used to analyze the correlations between brain measures and behavioral assessments (ADOS and Mullen subscales).

Compared to TD toddlers, autistic toddlers in the main sample showed: larger/thicker temporal and fusiform regions; smaller/thinner inferior frontal lobe and midline structures; larger callosal subregion volume; and smaller cerebellum. Most differences were replicated in the independent cohort. These brain alterations improved the accuracy of predicting language outcome at 6-month follow-up beyond intake clinical and demographic variables. Specifically, combining sMRI and clinical/demographic data achieved 79% accuracy (compared to 70% for clinical/demographic data alone and 60% for sMRI data alone) in predicting language outcome at 6-month follow-up, as measured by the Area Under the Curve (AUC = 0.79). Larger (more aberrant) gray matter volume (GMV) in the left hemisphere fusiform, and left and right hemisphere middle temporal regions correlated with higher ADOS total scores (i.e., more severe symptoms) and lower Mullen Early Learning Composite scores (i.e., poorer cognitive performance). Larger GMV in the right cerebellum was associated with higher Vineland Adaptive Behavior Composite and Daily Living Skills scores.

This study provides robust evidence of regional brain structural alterations in toddlers with ASD that are related to future language outcomes. The findings confirm previous research indicating atypical brain growth in ASD, but extend this by demonstrating the differential nature of these alterations—some regions are larger, others smaller, relative to global brain size. The observed alterations were primarily in brain regions involved in language, social, and face processing, supporting the link between brain structure and core ASD symptoms. The improved accuracy of language outcome prediction when combining sMRI and clinical data highlights the potential clinical utility of sMRI for early identification of children at risk of poor language outcomes. The brain-behavior correlations further support the idea that the identified brain regions play a crucial role in the manifestation of ASD symptoms and cognitive abilities. This research underscores the complex interplay between genetic, developmental, and environmental factors in ASD neurodevelopment.

This study demonstrates that regional brain structural differences in toddlers with ASD, particularly in areas related to language, social cognition, and face processing, are associated with future language abilities and symptom severity. The combination of sMRI and clinical data significantly improves prediction of language outcome at six months. Future research could focus on longitudinal studies with larger and more diverse samples, exploring the underlying biological mechanisms driving these structural differences and developing more sophisticated predictive models to further personalize early intervention.

The study's limitations include potential sample size mismatches and age differences between main and replication samples. While age effects were adjusted for, closer age matching in future studies would strengthen the findings. The use of the original ADOS version might have introduced some minor limitations. While the study used a robust methodology for controlling for multiple comparisons, the possibility of overfitting in the predictive models and the need for further replication in larger samples are notable considerations. The relatively small sample size of female participants also limits the generalizability of findings to female ASD individuals.