Protocols were approved by the institutional review board at the University of California, Davis (UCD). Subjects were recruited from the UCD M.I.N.D. Institute. In addition to a DSM IV diagnostic interview and an Autism Diagnostic Observation Schedule (ADOS) consensus diagnosis of ASD (with use of the Autism Diagnostic Interview-Revised, ADI-R, if supplemental information
for diagnosis required), all subjects had to have an initial Aberrant Behavior Checklist Irritability subscale (ABC-I) rating ≥ 18 (mean 25±6.7).
Exclusion criteria included bipolar disorder, schizophrenia, ASD of known genetic cause, nonverbal IQ < 55, seizures, fever, infection, metabolic disturbance or severe illness in the past year; antipsychotic use within 8 weeks of study entry; or inability of parents/care takers to give informed consent, travel to the visits, administer medication, and arrange for completion of rating scales. Other medications and treatments were permitted if started at least two months prior to initial screening and remained constant for the 8 week study duration. Subjects agreed not to start any new pharmacologic, dietary, behavioral or educational treatment during this study. The dosing schedule mirrored that used in the two recent positive trials of risperidone for treating severe behavioral problems in autism8, 9
. Briefly, risperidone was started at 0.5 mg at bedtime for 4 days. If that dosage was tolerated and there were continued behavioral symptoms, the dose was increased to 1 mg at bedtime for an additional 4 days. If tolerated and indicated, 0.5 mg was added in the morning for a daily total of 1.5 mg.
Affymetrix GeneChip® Human Exon 1.0 ST Arrays were used to obtain gene expression values. Collection of peripheral blood samples and processing of arrays was completed according to previously published protocols17
. Raw data (Affymetrix.CEL files) was imported into Partek Genomics Suite 6.4 (Partek Inc., St. Louis, MO, USA). Probe summarization and probe-set normalization were performed using Robust Multi-Chip Average (RMA), which included background correction, quantile normalization, log2
-transformation, and median polish probe set summarization.
All analyses used exon expression levels in blood prior to risperidone treatment (pre-risperidone expression levels) and pre-post risperidone change in ABC-I scores (ABC-I-%CHG), calculated as [(post-risperidone ABC-I score – pre-risperidone ABC-I score)/pre-risperidone ABC-I score]. Because improvement is reflected by a decrease in the ABC-I score, it is important to note that high responders are those demonstrating greater declines in ABC-I scores.
To initially detect the gene expression differences between subjects with the most pronounced response or lack of risperidone response, 17 subjects with the most extreme responses according to ABC-I-%CHG were identified. These subjects were grouped as high responders (ABC-I-%CHG = −95% to −71%; 9 subjects); or low responders (ABC-I-%-CHG = −29% to +6%; 8 subjects). Between-group gene expression profiles were compared for high vs. low responders using analysis of covariance (ANCOVA), controlling for the effects of age, gender and batch (α < 0.001, fold change > |1.5|). Because analyses used pre-drug blood RNA, and because any effect of dosing on outcome measures would not change the nature of correlations between pre-drug RNA and outcome measures, dosing was not included as a covariate. Multivariate analysis (unsupervised hierarchical clustering) was applied to evaluate relationships between high and low responders determined by these probes.
Medication responses, including those to risperidone, result in a wide range of responses. The initial extremes comparison allowed identification of genes whose expression was significantly different between (in this case) high and low responders. We then sought to identify expression differences that might be associated with not only high and low responders, but the range of responses in between these extremes. Correlation analyses (α < 0.001) using the probes identified in the extremes analysis were therefore performed to detect exons whose expression demonstrated a significant, linear relationship with ABC-I (coded as a continuous variable). Although an alternative approach was to do an omnibus correlation analysis, this typically yields large numbers of genes with significant correlations but ultimately low predictive power due to insufficient difference of expression at the extremes.
We considered gene ontology, pathway overrepresentation, and genomic co-regulation using the Database for Annotation, Visualization, and Integrated Discovery (DAVID, http://niaid.abcc.ncifcrf.gov/
), supplemented with manual curation to consider additional functional overlaps.