Salivary miRNA profiles identify children with autism spectrum disorder, correlate with adaptive behavior, and implicate ASD candidate genes involved in neurodevelopment

Abstract

Background: Autism spectrum disorder (ASD) is a common neurodevelopmental disorder that lacks adequate screening tools, often delaying diagnosis and therapeutic interventions. Despite a substantial genetic component, no single gene variant accounts for >1 % of ASD incidence. Epigenetic mechanisms that include microRNAs (miRNAs) may contribute to the ASD phenotype by altering networks of neurodevelopmental genes. The extracellular availability of miRNAs allows for painless, noninvasive collection from biofluids. In this study, we investigated the potential for saliva-based miRNAs to serve as diagnostic screening tools and evaluated their potential functional importance.

Methods: Salivary miRNA was purified from 24 ASD subjects and 21 age- and gender-matched control subjects. The ASD group included individuals with mild ASD (DSM-5 criteria and Autism Diagnostic Observation Schedule) and no history of neurologic disorder, pre-term birth, or known chromosomal abnormality. All subjects completed a thorough neurodevelopmental assessment with the Vineland Adaptive Behavior Scales at the time of saliva collection. A total of 246 miRNAs were detected and quantified in at least half the samples by RNA-Seq and used to perform between-group comparisons with non-parametric testing, multivariate logistic regression and classification analyses, as well as Monte-Carlo Cross-Validation (MCCV). The top miRNAs were examined for correlations with measures of adaptive behavior. Functional enrichment analysis of the highest confidence mRNA targets of the top differentially expressed miRNAs was performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID), as well as the Simons Foundation Autism Database (AutDB) of ASD candidate genes.

Results: Fourteen miRNAs were differentially expressed in ASD subjects compared to controls (p <0.05; FDR <0.15) and showed more than 95 % accuracy at distinguishing subject groups in the best-fit logistic regression model. MCCV revealed an average ROC-AUC value of 0.92 across 100 simulations, further supporting the robustness of the findings. Most of the 14 miRNAs showed significant correlations with Vineland neurodevelopmental scores. Functional enrichment analysis detected significant over-representation of target gene clusters related to transcriptional activation, neuronal development, and AutDB genes.

Conclusion: Measurement of salivary miRNA in this pilot study of subjects with mild ASD demonstrated differential expression of 14 miRNAs that are expressed in the developing brain, impact mRNAs related to brain development, and correlate with neurodevelopmental measures of adaptive behavior. These miRNAs have high specificity and cross-validated utility as a potential screening tool for ASD.

Keywords: Biomarker; Next generation sequencing; RNA-Seq; Saliva; miRNA.

Figures

Fig. 1

Differential expression and diagnostic utility of miRNAs in saliva of ASD children. a Hierarchical cluster analysis of the top 14 miRNAs. These miRNAs were differentially expressed in ASD children compared with Controls. Color indicates average Z-score of normalized abundance for each gene. A Euclidian distance metric was used with average cluster linkages for this figure. b Partial Least Squares Discriminant Analysis (PLS-DA) of the top 14 miRNAs showed the general separation of subjects into two clusters, using only three eigenvector components (x, y, and z axes labeled Component 1, Component 2, and Component 3) that collectively accounted for 55 % of the variance of the data set. c ROC-AUC analysis of the training data set indicated a very high level of performance in the logistic regression classification test (100 % sensitivity, 90 % specificity, with an AUC of 0.97)

Fig. 2

Monte-Carlo Cross-Validation analysis of the top 14 miRNAs. a The robustness of the 14 miRNA biomarkers was evaluated in stepwise fashion by determining their ability to correctly classify subjects using 100 iterations of a multivariate PLS-DA with 2, 3, 5, 7, 10, and 14 miRNAs included, and masking of 1/3 of the subjects during the training phase. This revealed an overall ROC-AUC of 0.92 and mis-classification of three ASD and four Control subjects. b Shows the classification of subjects plotted by predicted class probabilities from the MCCV (x axis), with incorrectly classified subjects identified by ID number. The y axis units are arbitrary. c Whisker box plots (showing median and inter-quartile range) of the four most robustly changed miRNAs according to the Mann-Whitney test