Autism Spectrum Disorder is characterized by reduced spatial autocorrelation of the excitation/inhibition ratio

Despite two decades of intensive investigation, the neurobiological underpinnings of autism spectrum disorder (ASD) defining symptoms remain elusive in large part because ASD represents a highly heterogeneous set of phenotypes, with wide-ranging genetic, neural and behavioural profiles that likely arise from multiple, partially overlapping mechanisms. Among various proposed mechanisms, the “excitation/inhibition (E/I) imbalance hypothesis” has been extensively researched, with substantial although sometimes conflicting evidence. Animal studies show that localised alterations in cortical E/I balance can disrupt network dynamics and reduce long-range connectivity, potentially explaining the weak central coherence that may be observed in ASD. However, neither the E/I imbalance hypothesis nor the weak central coherence hypothesis has provided a fully coherent explanation of ASD’s neurobiology or proved useful in identifying reliable biomarkers.

Here, we present a novel integrated hypothesis that unifies these theories, proposing that it is not the global average E/I ratio but rather its spatial heterogeneity across brain regions that may be altered in ASD and that disrupts the integration of brain network activity. To test this hypothesis, in a series of pre-registered analyses, we assessed EEG-derived E/I spatial autocorrelation in children with ASD (n = 248) and in neurotypical controls (n = 105). Our findings indicate significantly lower spatial autocorrelation in children with ASD, suggesting increased E/I variability across regions ( estimate: -0.51, 95% CI: -0.69 – -0.33), both in wakefulness (total wakefulness EEGs n = 234; estimate -0.36, 95% CI: -0.63 – -0.081) and sleep EEG (total sleep EEGs n = 304; estimate -0.49, 95% CI: -0.75 – -0.23).

Importantly, we validated these results in a large external dataset (ASD n = 230, NT n = 167; estimate : -0.22, 95% CI -0.33 – -0.02) compiled from the National Institute of Mental Health Data Archive (NDA), reinforcing the robustness of our findings.

Moreover, when exploring diagnostic accuracy, EEG-based E/I spatial autocorrelation along with sex differentiated ASD from neurotypical status with fair performance (ROC AUC: 0.725, sensitivity: 0.89, specificity: 0.67), highlighting its potential for optimization as a cost-effective screening tool. Overall, these results illuminate a possible neurobiological signature of ASD and suggest that EEG measures of E/I spatial heterogeneity may serve as a viable biomarker for the condition.