Abstract
Abstract
Background
Autism spectrum disorder (ASD) is neurobiologically heterogeneous, and group comparisons may mask effects confined to subgroups. We tested whether distinct biotypes exist within ASD using unsupervised clustering of multimodal MRI.
Methods
We analyzed 149 participants with ASD and 129 typically developing (TD) controls from the Autism Brain Imaging Data Exchange II (6 sites; age 5 to 64 years). Site-harmonized features comprised free water (FW), tissue anisotropy and diffusivity, diffusion tensor image analysis along the perivascular space (DTI-ALPS), cerebrospinal fluid (CSF) volumes and exploratory resting-state regional homogeneity. ASD participants were clustered with k-means using global (7 features) and regional (17 features) sets, with robustness tested by permutation and two further algorithms. TD participants were projected post hoc to test specificity. A composite imaging score was related to Social Responsiveness Scale (SRS) subscales, symptom severity and IQ using partial correlations with false discovery rate correction.
Results
Clustering identified an elevated free-water (EFW) biotype comprising approximately 10% of ASD participants, defined by markedly increased FW (d = +4.91 versus remaining ASD participants), reduced tissue anisotropy, reduced DTI-ALPS and enlarged ventricles. It was recovered across feature sets and algorithms (permutation p
Limitations
Each person was scanned once, so we cannot tell whether these brain differences cause social difficulties or how they change with age. The subgroup was small, clinical data were incomplete, the sample was mostly male with a wide age range, and the findings await confirmation in an independent group. Free water can reflect inflammation, swelling or tissue loss, so its biological meaning remains uncertain.
Conclusions
We identify an ASD subgroup defined by extracellular fluid excess and altered white matter that is associated with greater social difficulties. If replicated, this signature may help stratify individuals with ASD in mechanistic studies and clinical trials.