M. Burmeister; E. Sliwerska; F. Meng; T. Speed; E. Jones; W. Bunney; H. Akil; S. Watson
43rd American College of Neuropsychopharmacology (ACNP) Annual Meeting. 2004.


Gene expression microarray analysis in psychiatric research typically compares the expression of tens of thousands of genes in postmorten brain tissue between affecteds and matched controls. In contrast, genetic studies identify SNPs associated with these disorders in genomic DNA. Often, SNPs found associated with psychiatric disorders are postulated to affecte gene expression. We determined both gene expression and SNP genotype in >60 brain samples from subjects affected with depression, bipolar disorder or schizophrenia and controls. We genotyped >40 SNPs in a variety of candidate genes previously postulated to be involved in psychiatric disorders. One common SNP, the Val(108/158)Met coding variant in the catechol-O-methyltransferase (COMT) gene, was found highly significantly associated with expression level, as determined by RMA analyses on Affymetrix arrays (Irizarry et al. 2003). This association was reproducible across several brain regions and across hybridization performed in several laboratories. The Met allele has lower enzymatic activity, and showed an apparent increase in expression by about 20-30% per allele, with heterozygotes, as expected, intermediate in expression. It thus appeared that expression compensated for activity differences. However, a second probe set for COMT on the same array did not show this asso-ciation. To investigate this discrepancy, we investigated the expression difference at the probe level of the array. We found that a single oligonucleotide on the Affymetrix array showed a twofold difference in expression, whereas other oligos for COMT showed only non-specific differences in expression. This one oligonucleotide had the polymorphic Met allele in the center of its sequence, thus effectively creating a mismatch probe for the Val allele. We conclude that the apparent expression difference was an artifact due to the presence of the SNP on the oligo. This result was unexpected since typically hybridization of more than 10 oligonucleotides is averaged before analysis, and RMA downweighs results from atypically hybridizating oligos. However, while the larger twofold difference was reduced to only a 20% difference, it remained significant and showed the misleading expression difference. The artifact observed is likely to be quite common. A screen of Affymetrix oligos of the 133AB Plus Chip against dbSNP identified more than 30,000 SNPs that are present with one allele on that chip. It remains to be seen how often these cause a significant effect. Because of linkage disequilibrium between SNPs such artifacts can also be observed when the genotyped SNP is not present on the cDNA. Presence of common SNPs need to be taken into account when interpreting oligonucleotide microarray expression data.