20106
A Metabolic Profile of Autism Spectrum Disorder from Autism Phenome Project Patient Plasma

Saturday, May 16, 2015: 11:30 AM-1:30 PM
Imperial Ballroom (Grand America Hotel)
R. Burrier1, D. G. Amaral2, A. M. Smith1, P. R. West1, D. D. Li2, B. Fontaine1, E. Donley1 and S. J. Rogers3, (1)Stemina Biomarker Discovery, Madison, WI, (2)MIND Institute and Department of Psychiatry and Behavioral Sciences, University of California Davis Medical Center, Sacramento, CA, (3)University of California at Davis, Sacramento, CA
Background:  Diagnosis of autism spectrum disorder (ASD) at an early age is important for initiating effective intervention. The current average age of diagnosis in the United States is 4.5 years. Increasing evidence indicates that ASD is a complex disorder that has many causes. Identification of one or more metabolic signatures of ASD from blood samples will offer earlier screening and diagnosis to improve therapy and outcome as well as a richer biological interpretation of the disorder. 

Objectives:  Stemina conducted a metabolomic profiling study of blood from patients enrolled in the Autism Phenome Project (APP) to evaluate the metabolic signature in children with ASD as compared to typically developing (TD) children. The goal was to determine the most predictive combination of metabolic biomarkers capable of being translated into a broadly available diagnostic test for ASD. 

Methods:  Plasma was obtained from 180 children (ages 2 to 4) with ASD and 93 age-matched TD children enrolled in the APP.  Samples were analyzed using 4 orthogonal LC/HRMS-based methods that measure a broad range of metabolites. The patient samples were split into a training set (127 ASD, 68 TD) of samples for discovery profiling and a validation test set (42 ASD, 21TD) for evaluation of the diagnostic signatures discovered in the training set. Univariate, multivariate and machine learning methods were applied to the training set to identify the most predictive set of metabolic features that are capable of classifying patient plasma samples as being from ASD or TD children.  The molecular signatures were evaluated in the validation test set to determine their classification performance. 

Results:  Univariate analysis identified 292 differential metabolic features (p value < 0.05) between ASD and TD children.  Computational models were created using these features that classified the ASD and TD samples in the validation test with a maximum accuracy of 81% and AUC of 0.82 utilizing a minimum of 75 features.  Further evaluation of the metabolic features that were altered in children with ASD identified metabolites derived from multiple biochemical processes including lysophospholipids, androgens, and amino acids. A group of metabolic features correlated with the uremic toxin 3-Carboxy-4-methyl-5-propyl-2-furanpropionic (CMPF) were identified that exhibited a large differential abundance (> 10 fold, p val < 1e-6) in a subset of ~20% of ASD patients as compared to other ASD and TD individuals.  We are conducting additional studies to determine if this set of features could describe a metabolic subtype of autism.

Conclusions:  The  metabolomic analysis of plasma identified signatures able to discriminate individuals with ASD from TD individuals. These results form the basis for additional work to 1) developing a diagnostic test to detect ASD in children to improve patient outcomes, 2) gain new knowledge of biochemical mechanisms involved in ASD 3) identify biomolecular targets for new modes of therapy, and 4) identify biomarkers that can be used for personalized treatment and classification of potential responders versus non-responders through analysis of a patient’s biochemistry.