Distinguishing Autism Spectrum Disorders From Other Developmental Delays Using Blood Rnaseq

Background:

There is an unmet need for objective biomarkers to assist clinicians in the early diagnosis of childhood neurodevelopmental disorders. A number of investigators have reported changes in blood gene expression associated with autism spectrum disorders; Voineagu reviews this literature, while Glatt et al describe a microarray blood gene expression classification signature for distinguishing children with autism spectrum disorders from typically developing children.

Objectives:  

The aim of this study was to assess whether blood gene expression measured using next generation RNA sequencing (RNASeq) could provide a biomarker to distinguish children on the autism spectrum from children with other conditions that might present in the same clinical setting.

Methods:  

The CHARGE (CHildhood Autism Risks from Genetics and the Environment) study recruited children between the ages of 2 and 5, some of whom were diagnosed on the autism spectrum, and others with other developmental delays. Subjects were grouped based on thresholds of the ADOS, ADI-R, Vineland and Mullens test into autism spectrum disorder (ASD) and other developmental delay (DD) groups to approximate the clinical use case of a secondary screen for autism in children suspected of neurodevelopmental disorders.

Blood samples were acquired from each subject in RNA-stabilizing PAXgene tubes.  RNA was isolated and processed using the TrueSeq sequencing prep with poly-A selection for mRNA. RNASeq was then performed on an Illumina HiSeq 2000 Sequencer using 1/3 lane per sample. 174 ASD and 96 DD samples passed final QC, for a total of 270 samples.  

Sequence data were processed through the Tuxedo RNASeq pipeline to yield counts per gene, which were normalized by downsampling. The sample was divided into a training set (n= 153) and a holdout set (N=117), each of which was repeatedly randomly subsampled to achieve gender and age balance between the ASD and DD groups. On each iteration, informative features were selected by t-test and a support vector machine classifier was trained on a balanced subsample of the training set and tested on a balanced subsample of the holdout set; AUC’s (area under the ROC curve) were averaged across iterations. 

Results:  

The mean AUC for the holdout set was 65.6 +/- 2.9%. When a 90% sensitivity threshold was selected on the classifier risk score, the mean specificity was 25.3, with 95% CI [13.6, 40.6%].  Gene categories found significant by ranksum test on the t-statistic include RNA processing, cell cycle, immune and inflammation-related GO categories.

Conclusions:  

To our knowledge this represents the first report of a classification signature for ASD vs. DD using blood RNASeq. While the understanding of genetic contributions to autism spectrum disorders has been making impressive progress in recent years, genetic causes are individually rare, and are thus not sensitive in a diagnostic context. A gene expression signature with moderate AUC has potential clinical utility as a sensitive assay for identifying children at risk for ASDs within a population that is already suspected of neurodevelopmental disorders. Planned followup studies include a multicenter clinical study to further refine and validate a blood-based assay.