ABSTRACT: Alcohol Use Disorder (AUD) is a chronic, relapsing syndrome diagnosed by a heterogeneous set of behavioral signs and symptoms. There are no laboratory tests that provide direct objective evidence for diagnosis. Microarray and RNA-Seq technologies enable genome-wide transcriptome profiling at low costs and provide an opportunity to identify biomarkers to facilitate diagnosis, prognosis, and treatment of patients. Brain gene expression patterns can discriminate alcohol-dependent and non-dependent people and predict drugs that reduce drinking in rodents. However, access to brain tissue in living patients is not possible. Blood contains cellular and extracellular RNAs that provide disease-relevant information for some brain diseases. We hypothesized that blood gene expression profiles can be used to diagnose AUD. We profiled brain (prefrontal cortex, amygdala, and hypothalamus) and blood gene expression levels in C57BL/6J mice using RNA-seq one week after chronic intermittent ethanol (CIE) exposure, a mouse model of alcohol dependence. To determine the preservation of gene expression levels between blood and brain, we calculated the Spearman correlation coefficient between blood and brain mean gene expression levels across all subjects and found a high degree of preservation (rho range: [0.50, 0.67]) with hundreds of transcripts in blood correlated with their brain transcript levels. To determine whether the transcriptional response to alcohol dependence was similar in blood and brain, we studied the overlapping differentially expressed genes (DEGs) and gene coexpression networks. Although there was small overlap between blood and brain DEGs, there was considerable overlap of gene networks perturbed after CIE related to cell-cell signaling (e.g., GABA and glutamate receptor signaling, endocannabinoid signaling, synaptogenesis), immune responses (e.g., antigen presentation, communication between innate and adaptive immune systems), and protein processing / mitochondrial functioning (e.g., ubiquitination, unfolded protein responses, oxidative phosphorylation). To determine whether blood gene expression can predict alcohol dependence status, blood gene expression data were used to train classifiers (logistic regression, random forest, and partial least squares discriminant analysis), which were highly accurate at predicting alcohol dependence status (maximum AUC for females: 90.1%; males: 80.5%). These results suggest that gene expression profiles from peripheral blood samples contain a biological signature of alcohol dependence that can discriminate between alcohol-dependent and non-dependent subjects.
