The results of this study suggest exon arrays in conjunction with cell lines and brain samples may be a valuable tool to identify gene expression associated with genotype in the context of psychiatric disorders. Despite a relatively small sample of cell lines grown under two different conditions, this study identified eight differentially expressed transcripts between cases and controls when compared under two glucose conditions, 328 transcripts with significant Glucose deprivation × Probeset effects on expression, and a total of 122 transcripts with significant Diagnosis × Probeset effects on expression demonstrating probeset-specific effects of diagnosis and glucose deprivation on expression. These transcripts passed a 25% step-down Benjamini Hochberg false discovery and the complete dataset is available by request.
There were no significant main effects of diagnosis. This might be due to the fact analyses were performed using summarized transcript data and gene transcript expression was estimated using the mean expression of all probesets, which for some transcripts could be 50 - 100 probesets. On the other hand, the Glucose deprivation × Probeset and Diagnosis × Probeset interaction effects were based on probeset data summarized at the exon level which was 4 - 8 probesets. The finding that expression differences within closely matched schizophrenia and unaffected relatives are restricted within certain exons of a transcript is consistent with the alternatively spliced transcripts being an over-represented significant category between schizophrenia and controls. As the exon array does not contain junction probes, this conclusion is speculative, as the exon array would measure both spliced and unspliced transcripts, and does not actually differentiate spliced and unspliced transcripts except in situations where splicing is an all or none occurrence. Over 70% of human multi-exon genes are alternatively spliced . Also, limited power may have played a role in our lack of diagnosis or glucose deprivation effects. The effect sizes detected for 100 most significant diagnosis effects ranged from 0.05 to 1.19. A power analysis done in G*Power showed 80% power to detect an effect size of 0.55 .
There is a growing body of evidence of glucose metabolism abnormalities in schizophrenia. In a study by Ryan et al (2003), more than 15% of subjects with schizophrenia had impaired fasting glucose tolerance . Subjects with schizophrenia also had higher fasting plasma levels of glucose, insulin and cortisol and were more insulin resistant than control subjects . There is also evidence for linkage between enzymes that control glycolysis and schizophrenia, including PFKFB2, a rate limiting enzyme in glycolysis, and hexokinase, a regulatory enzyme in the metabolic pathway . Insulin has been shown to increase the expression of genes that are decreased in the hippocampus in subjects with schizophrenia in neuroblastoma cells . Moreover, glucose administration improves verbal declarative memory in clozapine-treated subjects with schizophrenia , although the mechanism of action remains unknown. Previous research has implicated metabolic and mitochondrial genes [3, 7, 8, 39, 40], including ornithine aminotransferase (OAT)  and metabolic enzyme cytosolic malate dehydrogenase (MDH1) [3, 41] in schizophrenia, which have not been shown to be influenced by cigarette smoking or antipsychotic treatment. For these reasons, we examined the effects of glucose deprivation on gene expression in subjects with schizophrenia compared to controls.
Of the significantly altered transcripts identified in this study, only GLS showed significant Diagnosis × Probeset, significant Glucose deprivation × Probeset, and significant Diagnosis × Glucose effects. This gene and four other Diagnosis × Probeset effect genes were selected for further study in the DLPFC and only GLS was significant in DLPFC. Glutaminase, which converts glutamine to glutamate, was decreased in expression in glucose-deprived LCLs and decreased in expression in the DLPFC in subjects with schizophrenia compared to controls. This is in contrast to previous findings showing increases in glutaminase in the thalamus  and in DLPFC  of subjects with schizophrenia. In addition to GLS expression being regulated by cellular environment, the effects of diagnosis and glucose on expression varied by exon and may be isoform-specific. This gene may be worth further association study by SNP analysis or resequencing. In addition to glutaminase, there was a significant Diagnosis × Glucose deprivation × Probeset effect on PPARGC1B expression. PPARGC1B is located at 5q33, which has been previously linked to schizophrenia  and PPARGC1B participates in mitochondrial biogenesis and plays a role in fat oxidation and nonoxidative glucose metabolism , which is interesting in light of evidence of glucose metabolism abnormalities and mitochondrial dysfunction in schizophrenia. None of the other Diagnosis × Glucose deprivation × Probeset effect genes that passed a step-down FDR cutoff of 25% is known to play a role in glucose metabolism or to be sensitive to metabolic stressors. Thus, induction of gene expression changes in vitro by glucose reduction could be compared to increased glucose concentrations which may alter mitochondrial and metabolic function.
IRF5 expression was down-regulated in cohort 1. We examined the relationship between Interferon Regulatory Factor 5 (IRF5) expression and genotype and found expression of IRF5 to variation at SNP rs10954213 in all cohorts. However, IRF5 gene expression alteration in LCLs or brain was not observed in schizophrenia in a larger sample of subjects by RT-PCR and no SNP association with schizophrenia was observed.
IPA network analysis of transcripts with significant Diagnosis × Glucose deprivation × Probeset effects revealed a network of 18 genes that showed diagnosis-dependent expression alterations following glucose deprivation. D-glucose and hepatocyte nuclear factor 4, alpha (HNF4A) were two nodes in this gene network. HNF4A mutations have been associated with type II diabetes [46–48]. This gene is located at 20q12, a previously identified locus of schizophrenia . This gene is interesting in light of findings of increased prevalence of diabetes in schizophrenia , and further investigation of HNF4A variants in schizophrenia may be warranted. IPA network analysis of transcripts with significant Diagnosis × Probeset effects revealed a network of 18 down-regulated and four up-regulated genes. Interestingly, calmodulin, which has previously been shown to be altered in expression in schizophrenia , was a centrally located complex within this network. If replicated, genes in these networks may represent novel targets for further study in schizophrenia.
The results of this study were compared with a recent study of microarray gene expression in the DLPFC of subjects with schizophrenia and control subjects . Of the 54 genes identified by Maycox et al (2009) , there were 12 genes with diagnosis or Diagnosis × Probeset effects displaying nominal p-values < 0.05 in the current study (see Additional file 1). These results indicate that case and control derived cell lines might be a reliable and easily maintained resource for the study of the expression phenotype that is disease associated.
The human transcriptome was recently shown to have over 90% alternatively spliced variants  and it will be increasingly important to identify those variants at the exon level that differentiate cases and controls. We have shown that differential exon usage is associated with case - control differences, while few transcripts showed a complete differentiation across the entire gene; instead, differential expression was restricted to specific exons within a transcript. To fully pursue individual splice variants requires more validation experiments, but having a profile across the transcript represents a first approximation.
The use of exon arrays allowed for the possibility of detecting exon-specific changes in expression . Previous work has shown a high level of agreement between digital counts by deep sequencing agreement with exon array . In identifying gene expression alterations in LCLs, it may be important to combine theoretically valid experimental manipulations of cell lines from subjects with schizophrenia to induce gene expression alterations relevant to the disorder. An example would be to silence gene expression through siRNA transfection, and measure effects on known interactors.
Expression of schizophrenia candidate genes in lymphoblastoid cell lines
One aspect of the present work was that many of the top candidate genes did not exhibit diagnosis or glucose deprivation effects. Although many schizophrenia candidate genes were expressed above background levels in LCLs, there was no overlap between the candidate genes identified in a recent meta-analysis of gene association studies  and genes with significant Diagnosis × Probeset × Glucose deprivation effects on expression found in this study. This may be due to tissue specificity (brain versus lymphocyte) of gene expression alterations in schizophrenia, or due to the selection of the control group, which consisted of relatives of subjects with schizophrenia and who likely carry many of the same genetic risk factors as their affected family members. Alternatively, reactive oxygen species generation might be increased more by increased glucose utilization , as we did not detect an over representation of mitochondrial or reactive oxygen genes by lowering glucose concentrations in cells. It is to be expected that increased glucose concentration would lead to more alterations in mitochondrial related genes.
A second aim of this study was to identify cis-regulated alterations in gene expression from the mRNA by SNP Browser  as one criterion for further study in the context of an association with the disease. There was a significant association between an IRF5 genotype and expression. A genotype effect of SNP rs10954213 on IRF5 expression was significant in LCLs, and was validated in the prefrontal cortex and anterior cingulate. However, there was no significant Diagnosis × Genotype effect on expression or a significant association of genotype with diagnosis in a larger sample for rs10954213. IRF5 exhibits genetically controlled polymorphic transcript variation . The SNP rs10954213 variant in IRF5 is in the 3'- untranslated region (UTR) previously shown to influence IRF5 gene expression  in LCLs. This SNP creates a functional polyadenylation site and the A genotype correlates with increased expression of a transcript variant containing a shorter 3'-UTR , which was the same association found in LCLs, and brain tissue in the present study.
The 3'-UTR of mRNA has important functions in the stability, localization, and translation of mRNA . MicroRNAs (miRNAs), are small noncoding RNAs that recruit an argonaute protein complex to a complementary target mRNA. Each miRNA-Ago complex interacts with a specific mRNA, typically through pairing of nucleotide bases between the miRNA sequence and complementary sequences in the mRNA's 3-untranslated region (3'UTR), which results in translation repression or degradation of the mRNA. A miRNA binding site for hsa-miR-185 was identified 1 bp downstream of SNP rs10954213 using the web based tool, MicroInspector . It is possible that the effect of SNP rs10954213 variant on IRF5 protein levels may be mediated via this shortened 3'-UTR lacking this miRNA binding site. A further experiment to correlate reduction in protein expression with genotype would address this speculation.
We cannot rule out the possible effects of antipsychotics or cigarette smoking on gene expression. Recent research suggests that cigarette smoke and antipsychotic treatment may lead to gene-silencing effects through DNA methylation changes [58–60] which could lead to long-term changes in cell lines derived from smokers. The overlap between genes included in the IPA analysis and those known to be altered by antipsychotics or nicotine were assessed with Ingenuity and no overlap was observed.