Comparison of microarray-based mRNA profiling technologies for identification of psychiatric disease and drug signatures

Abstract

The gene expression profiles of human postmortem parietal and prefrontal cortex samples of normal controls and patients with bipolar disease, or human neuroblastoma flat (NBFL) cells treated with the mood-stabilizing drug, valproate, were used to compare the performance of Affymetrix oligonucleotide U133A GeneChips^® and Agilent Human 1 cDNA microarrays. Among those genes represented on both platforms, the oligo array identified 26–53% more differentially expressed genes compared to the cDNA array in the three experiments, when identical fold change and t-test criteria were applied. The increased sensitivity was primarily the result of more robust fold changes measured by the oligonucleotide system. Essentially all gene changes overlapping between the two platforms were co-directional, and ranged from 4 to 19% depending upon the amount of biological variability within and between the comparison groups. Q-PCR validation rates were virtually identical for the two platforms, with 23–24% validation in the prefrontal cortex experiment, and 56% for both platforms in the cell culture experiment. Validated genes included dopa decarboxylase, dopamine beta-hydroxylase, and dihydropyrimidinase-related protein 3, which were decreased in NBFL cells exposed to valproate, and spinocerebellar ataxia 7, which was increased in bipolar disease. The modest overlap but similar validation rates show that each microarray system identifies a unique set of differentially expressed genes, and thus the greatest information is obtained from the use of both platforms.

Introduction

The first oligonucleotide gene expression arrays (Lockhart et al., 1996) emerged shortly after the advent of cDNA microarray printing (Schena et al., 1995). The increasing availability of both types of microarrays and their coverage of a large proportion of the genome has enticed many researchers to enter the field. However, the high cost and specialized equipment required for microarrays, the vast amounts of data they generate, and the very different principles and methods of their use, require that researchers know which technology is best suited for their needs. Despite a large and increasing literature of microarray results, few direct comparisons between oligonucleotide and cDNA arrays have been published. The conclusions of these studies, most of which analyzed cell lines, are conflicting, and none address the unique technical difficulties associated with CNS tissue (Burns et al., 2003, Kothapalli et al., 2002, Kuo et al., 2002, Nishizuka et al., 2003, Rhodes et al., 2002, Tan et al., 2003, Yuen et al., 2002).

The human brain remains one of the more challenging tissues in which to measure changes in gene expression with high-density microarrays (Bahn et al., 2001, Yue et al., 2001). Investigators have profiled postmortem human brain tissue with cDNA (Bezchlibnyk et al., 2001, Kuromitsu et al., 2001, Middleton et al., 2002, Mirnics et al., 2000, Palfreyman et al., 2002, Vawter et al., 2001, Yue et al., 2001) or oligonucleotide microarrays (Hakak et al., 2001), but not by both. The quality of the data obtained from human postmortem brain RNA can be compromised by poor RNA quality, low RNA abundance, and inter-patient variability (Bahn et al., 2001, Bunney et al., 2003, Williams et al., 2002, Yue et al., 2001). Brain pH, postmortem interval (PMI), age of the patient at death, and heterogeneous anatomical distributions of mRNA are other factors that affect RNA quality and the ability to obtain reproducible gene expression changes (Nisenbaum, 2002). Such variables can be mitigated by using cultured human neuronal cell lines.

The present study analyzed gene expression profiles in the postmortem human brain and in a human neuroblastoma cell line treated with a psychoactive drug. Gene expression was measured in all tissues using two widely-employed microarray technologies, the Affymetrix oligonucleotide-based GeneChip^® and the Agilent cDNA-based microarray. The human U133A GeneChip^® (Affymetrix, Sunnyvale, CA) contains 22,283 sequences that cover approximately 14,000 distinct UniGene clusters. It utilizes 11 oligonucleotide probe pairs per gene to detect transcript levels in the sample. Each probe pair consists of a “perfect match” 25 nucleotide sense-strand oligo, and a “mismatch” sense-strand oligo that is identical except for a centrally located single nucleotide substitution to control for non-specific hybridization. The mRNA from each biological sample serves as a template to generate biotinylated complementary RNA which is then hybridized to the array. Hybridization intensities for each gene are subsequently compared to those from other samples hybridized to independent arrays.

The Human 1 cDNA microarray (Agilent Technologies, Palo Alto, CA) contains 12,811 clones from more than 7000 UniGene clusters. Each clone is represented by a PCR-amplified, double-stranded cDNA product immobilized on the slide. The mRNA obtained from two biological samples are separately converted to cDNA labeled with distinct fluorescent dyes, usually cyanines 3 and 5, mixed together, and hybridized to a single array. Hybridization intensities from the two dyes are measured and compared for each gene within the array to identify gene expression differences between the two samples. Utilization of a common reference sample for each array allows objective comparisons between samples on separate arrays (Garber et al., 2001, Perou et al., 2000, Sterrenburg et al., 2002, Yang et al., 2002).

To assess the performance of each microarray platform from a technical and biological standpoint, we evaluated three different sets of samples in parallel on both platforms. In the first experiment, the technical reproducibility of each system was evaluated with five or six replicates of two parietal cortex samples, one from an unaffected control and one from a bipolar disorder patient. Next, prefrontal cortex samples from eight control and eight bipolar disorder cases were compared. Finally, the effect of a common mood-stabilizing drug, valproic acid, was examined on the human neuroblastoma flat (NBFL) cell line using replicate cell culture experiments. This cell line was used because it contains neuron-like cells that can be propagated easily in culture, unlike primary neuronal cultures or neuronal stem cell lines. An independent mRNA detection method, the quantitative real-time polymerase chain reaction assay (Q-PCR), was used to determine the confirmation rate for each platform in these experiments.