Urinary Markers of Oxidative Stress in Children with Autism Spectrum Disorder (ASD)

Background: Autism spectrum disorder (ASD) is a developmental disorder characterized by deficit in social interaction, restricted interest and repetitive behavior. Oxidative stress in response to environmental exposure plays a role in virtually every human disease and represents a significant avenue of research into the etiology ASD.

Objectives: The aim of this study was to explore the diagnostic utility of four potential biomarkers in urine.

Methods: One hundred-thirty-nine (139) patients with ASD (89% male, average age = 10.0 years, age range = 2.1 to 18.1 years) and 21 healthy children included in a control group (52% male, average age 8.2, age range = 2.5 to 13.7 years) were recruited for this study. Urinary 8-OH-dG, 8-isoprostane, dityrosine and hexanoil-lisine were determined by the ELISA method. Urinary creatinine was determined by the kinetic Jaffee reaction and all biochemical measurements were normalized in relation to creatinine. Non-parametric tests and support vector machines (SVM) with a three different kernels (linear, radial, polynomial) were employed to explore and optimize the multivariate prediction capability of these biochemical measurements for predicting an ASD diagnosis. The SVM models were computed using bootstrapping, with additional SMOTE sampling procedures (Chawla et al., 2002) utilized because of unbalanced data.

Results: Using non-parametric test we found that children with ASD had an increased concentration of 8-isoprostane compared to the control group (unadjusted p = .046). When all four biochemical measurements were combined using SVM’s with a radial kernel we could predict ASD diagnosis with a balanced accuracy of 87.9% and thereby account for an estimated 31.6% of variance (p < .001). Looking at the standardized variable importance an ASD diagnosis was best explained in terms of 8-isoprostane and 8-OH-dG.

Conclusions: Our results indicate that the examined urinary biomarkers combined together may differentiate children with ASD from healthy peers to a significant extent. However, the etiological importance of these findings is difficult to assesses, due to the high-dimensional nature of SVM’s and radial kernel. In addition, our results should be interpreted with care since we did not have a sufficient large enough group of both children with ASD and healthy peers to have independent training and testing samples for our SVM’s. Nonetheless, our results show that machine learning methods may provide significant insight into ASD and other disorder that could be related to oxidative stress.