Objectives: Atypical eating habits frequently characterize people with Autism Spectrum Disorder (ASD) from early infancy. Food selectivity, defined as a narrow variety of food intake and reticence to new food, is the most frequent feeding disorder in ASD. The objective of this study was to investigate the adaptive functioning and the behavioral profile of individuals with ASD with food selectivity (FS) in comparison to an ASD sample without food selectivity (NFS). Methods: We conducted a retrospective study on 286 children (mean age = 46.95 months) with a diagnosis of ASD; 43.3% of the sample had a history of food selectivity (FS), whereas 56.6% had no history of food selectivity (NFS). Results: No differences were found between the FS and NFS groups on cognition, autism symptom levels, and age. The FS group presented lower adaptive skills and greater behavioral problems in comparison to the NFS group. A worse clinical profile characterized children with more than one kind of food selectivity. Conclusions: The early identification and longitudinal evaluation of specific clinical and behavioral patterns in children with ASD associated with food selectivity could contribute to a better understanding of the relationship between autism symptoms and atypical eating habits.
Autism Spectrum Disorder (ASD) arises from complex and not yet completely understood interactions between genetic and environmental factors. Alongside known hallmarks of neurobiological and structural changes in ASD brain, alterations in gut microbiota are frequently observed in ASD and may contribute to its pathophysiology. Identifying reliable biomarkers through multivariate analysis and machine learning offers promising avenues for improving ASD diagnosis and understanding comorbid gastrointestinal symptoms. In this study, a machine learning model was trained to identify ASD and healthy controls based on the theoretical production of metabolites for each gut bacterial species and each individual, combining the data collected from two global databases (GMRepo v2 and Agora2). Random Forest Classification models reach a mean accuracy of 85%, and a subsequent literature analysis of the 5% most significant metabolites showed a 40% correspondence with previously published in vivo studies. Some of the most relevant compounds detected by the theoretical model are amino acid and amino-acidic derivatives, volatile organic compounds, and short-chain fatty acids. Results are coherent with empirical evidence, supporting microbiota's role in ASD pathophysiology by contributing to neurotransmitters' biosynthesis and degradation, intestinal epithelial barrier integrity, immunological modulation. Future work will focus on stratified sampling, empirical validation, and developing personalized metabolic signatures for early diagnosis and precision medicine.