Evidence for Metabolomic Phenotypes Based on Analysis of Plasma from the APP Cohort

Saturday, May 14, 2016: 11:30 AM-1:30 PM
Hall A (Baltimore Convention Center)
A. Smith1, R. Alexandridis1, R. Burrier1, P. West2, L. A. Egnash3, J. King1, D. D. Li4, D. G. Amaral5 and E. Donley1, (1)Stemina Biomarker Discovery, Madison, WI, (2)BioAnalytical Chemistry, Stemina Biomarker Discovery, Madison, WI, (3)Client Services and Operations, Stemina Biomarker Discovery, Madison, WI, (4)University of California at Davis, Sacramento, CA, (5)UC Davis The M.I.N.D. Institute, Sacramento, CA
Background:   Metabolomics has the potential to identify predictive and actionable biomarker profiles from a child’s inherited biochemistry as well as capture the interactions of the gut microbiome with dietary and environmental factors that contribute to ASD. Identification of common metabolic profiles in children with ASD creates an opportunity to develop metabolic based diagnostics that enable early diagnosis and identification of metabolic subtypes that can facilitate intervention and lead to a better understanding of the biochemical changes associated with ASD.

Objectives: 1) To identify predictive metabolic signatures which distinguish ASD from typically developing (TD) children enrolled in the Autism Phenome Project.  2) To discover metabolic subtypes of ASD as defined by differentially abundant metabolic features that can identify a subset of ASD individuals with a high positive predictive value and specificity.

Methods: Plasma was obtained from 180 children with ASD at the time 1 assessment time point and from 93 age-matched TD children.  Samples were analyzed using 4 orthogonal HILIC and C8 LC/HRMS-based methods as well as GC/MS. Data from the patient samples were split into a training set, utilized for identification of biomarkers, and an independent validation test set used for evaluation of the diagnostic signatures. Univariate, multivariate, machine learning and heuristic methods were applied to the training set to identify predictive metabolic features. The predictive molecular signatures were evaluated in the validation test set to determine their classification performance.

Results:   Computational models were created that classified the ASD and TD samples in the validation set with a maximum accuracy of 79% and AUC of 0.80. Differentially abundant features (p value < 0.05) from the models were identified as metabolites derived from multiple biochemical processes which included lysophospholipids, hormone sulfates, and amino acids. Two metabolites, 3-Carboxy-4-methyl-5-propyl-2-furanpropionic (CMPF) and an unknown metabolite related to CMPF exhibited a large differential abundance (> 6 fold, p val < 1e-6) in a subset of subjects with ASD.  These metabolites discriminates 14% of the ASD population in the APP study and may describe a metabolic subtype of ASD. 


Non-Targeted metabolomic profiling of children with ASD revealed predictive metabolic signatures able to discriminate individuals with ASD from TD individuals and suggests the presence of metabolic subtypes. Applying this paradigm to identify metabolic signatures associated with ASD and elucidating their biochemical implications may be useful in directing therapy on a personalized basis. Work is currently under way to compare these metabolic phenotypes with behavioral and neuroimaging data acquired in the APP. These results form the basis for additional work with the goals of 1) developing diagnostic tests to detect ASD in children to improve their outcomes through personalized treatment, 2) gaining new knowledge of biochemical mechanisms involved in ASD and 3) identifying biomolecular targets for new modes of therapy.