Metabolomics as a tool for discovery of biomarkers of autism spectrum disorder in the blood plasma of children

Abstract

Background: The diagnosis of autism spectrum disorder (ASD) at the earliest age possible is important for initiating optimally effective intervention. In the United States the average age of diagnosis is 4 years. Identifying metabolic biomarker signatures of ASD from blood samples offers an opportunity for development of diagnostic tests for detection of ASD at an early age.

Objectives: To discover metabolic features present in plasma samples that can discriminate children with ASD from typically developing (TD) children. The ultimate goal is to identify and develop blood-based ASD biomarkers that can be validated in larger clinical trials and deployed to guide individualized therapy and treatment.

Methods: Blood plasma was obtained from children aged 4 to 6, 52 with ASD and 30 age-matched TD children. Samples were analyzed using 5 mass spectrometry-based methods designed to orthogonally measure a broad range of metabolites. Univariate, multivariate and machine learning methods were used to develop models to rank the importance of features that could distinguish ASD from TD.

Results: A set of 179 statistically significant features resulting from univariate analysis were used for multivariate modeling. Subsets of these features properly classified the ASD and TD samples in the 61-sample training set with average accuracies of 84% and 86%, and with a maximum accuracy of 81% in an independent 21-sample validation set.

Conclusions: This analysis of blood plasma metabolites resulted in the discovery of biomarkers that may be valuable in the diagnosis of young children with ASD. The results will form the basis for additional discovery and validation research for 1) determining biomarkers to develop diagnostic tests to detect ASD earlier and improve patient outcomes, 2) gaining new insight into the biochemical mechanisms of various subtypes of ASD 3) identifying biomolecular targets for new modes of therapy, and 4) providing the basis for individualized treatment recommendations.

Conflict of interest statement

Competing Interests: All authors except Blythe Corbett have, within the past five years, received salary from, and/or hold stocks or stock options in, Stemina Biomarker Discovery, which as a company may gain or lose financially from the publication of this manuscript. PRW, LAE, AMS, MER, JAP, BRF, KRC, GGC, ELRD and REB are employees of Stemina, whose company funded this study and holds, or is currently, applying for patents relating to the content of the manuscript as follows: BIOMARKERS OF AUTISM SPECTRUM DISORDER; PCT App No. PCT/US2014/045397 and METABOLIC BIOMARKERS OF AUTISM; PCT Application No. PCT/US2011/034654. There are no further patents, products in development, or marketed products to declare. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials.

Figures

Figure 1. Classification modeling process.

A three-layer nested cross-validation approach was applied using both PLS-DA and SVM modeling methods to determine significant features capable of classifying children with ASD from TD children. The 179 features of the training set were analyzed using a leave-one-group-out cross-validation loop as described. The results from this cross-validation process were used to estimate model performance and create a robust feature VIP score index to rank the ASD vs TD classification importance of each of the 179 features. These feature ranks were used to evaluate the performance of the molecular signature using an independent validation set.

Figure 2. Feature Importance Rankings.

The top 179 features were compared for rank between SVM and PLS modeling methods. The lowest rank scores represent the most important features.

Figure 3. Performance of the SVM and PLS models.

Average AUC and accuracy of the (a) SVM and (b) PLS models containing different numbers of features. The bar graphs show the number of optimal models which were derived from recursive feature elimination process that was included in the resampling process for the indicated number of features.

Figure 4. ROC curve performance of the classification models from the training and validation sets.

The average of 100 iterations of the classifier for the best performing feature sets following recursive feature elimination comparing ASD vs. TD samples (Black and Grey Lines). The blue (PLS) and red (SVM) lines are ROC curves of the best performing validation feature subsets. Vertical bars represent the standard error of the mean.