Genomic DNA as a cohybridization standard for mammalian microarray measurements

Published online June 9, 2004 Nucleic Acids Research, 2004, Vol. 32, No. 10 e81 DOI: 10.1093/nar/gnh078 Genomic DNA as a cohybridization standard for mammalian microarray measurements Brian A. Williams, Richele M. Gwirtz and Barbara J. Wold* Division of Biology, MC 156-29, California Institute of Technology, Pasadena, CA 91125, USA Received March 7, 2004; Revised and Accepted May 4, 2004 ABSTRACT INTRODUCTION DNA microarrays have quickly become an indispensable tool for transcriptome analysis (1). Mechanical spotting of DNA on glass slides has emerged as a widely used microarray platform because it affords ¯exibility of array design and relative economy. However, this technology also has some signi®cant shortcomings. Because feature geometry and the amount of DNA per feature vary within a gene chip, and also from one chip to another, measurements must be made as internal ratiometric comparisons of one RNA sample with a reference (or `denominator') RNA (2,3). This is done by simultaneous *To whom correspondence should be addressed. Tel: +1 626 395 4916; Fax: +1 626 449 0756; Email: Nucleic Acids Research, Vol. 32 No. 10 ã Oxford University Press 2004; all rights reserved A persistent design problem for ratiometric microarray studies is selecting the `denominator' RNA cohybridization standard. The ideal standard should be readily available, inexpensive, invariant over time and from laboratory to laboratory, and should represent all genes with a uniform signal. RNA references (both commercial `universal' and experimentspeci®c types), fall short of these goals. We show here that mouse genomic DNA is a reliable microarray cohybridization standard which can meet these criteria. Genomic DNA was superior in universality of coverage (>98% of genes from a 16 000 feature mouse 70mer microarray) to the Stratagene Universal Mouse Reference RNA standard. Ratios for genes in very low abundance in the Stratagene standard were more unstable with the Stratagene standard than with genomic DNA. Genes with mid-range, and therefore presumably optimal RNA denominator values, showed comparable reproducibility with both standards. Inferred ratios made between two different experimental RNAs using a genomic DNA standard were found to correlate well with companion, directly measured ratios (Spearman correlation coef®cient = 0.98). The advantage in array feature coverage of genomic DNA will likely increase as newer generation microarrays include genes which are expressed exclusively in minor tissue or developmental domains that are not represented in mixed tissue RNA standards. hybridization of experimental and reference samples, where each RNA population is transcribed into cDNA with a different ¯uorophore (typically Cy3 for one and Cy5 for the other). While this is very effective for direct comparisons of just two RNA samples, the full power of large-scale expression analysis comes from comparisons of multiple (tens to hundreds or even thousands) of different RNA samples. To do this using spotted microarrays, the ratio observed for each feature on the array is compared across all gene chips in a study, each of which has used the same denominator RNA sample (converted to labeled cDNA or cRNA). Although this design has proved very successful, the requirement for internal ratiometric measurement presents a thorny set of problems that come from properties of the reference hybridization standard. For example, instability and error is expected for RNAs not represented in the reference or, alternatively, for RNAs so prevalent in the reference that they saturate their corresponding features (detectors). Moreover, the reference RNA sample composition is not standard from one study to another, usually having been selected based on different criteria for each study. Once a standard is selected, the vagaries of biology make it dif®cult to reproduce precisely from one preparation to another. This means that global comparisons between studies done in the same laboratory over a long time or between different laboratories are compromised. These issues have so far been dealt with using strategies that range from selecting a single tissue standard, such as whole spleen RNA for a study of B cells done by the Alliance for Cell Signaling (AFCS) (http://www.signaling-gateway. org) to making a denominator mixture of RNAs by pooling aliquots from each sample in a given study (4,5), to attempting to make a `general mixture' of RNA from diverse cell lines (e.g. the Stratagene Universal Reference RNA standards) (4,6). Genomic DNA should, in principle, be a more general, invariant and inexpensive solution (1,7). Major virtues of the genome as a cohybridization `standard' include complete sequence representation, sequence stability over time and from one preparation to another, uniform prevalence for most genes and very low cost. These features mean that it is also applicable to any array, independent of which subset of genes is arrayed or which strand, in the case of oligonucleotides, is represented. It is also clear that genomic DNA presents problems and challenges of its own. In the large vertebrate genomes that are our principle interest, mRNA coding sequences are highly diluted by non-coding DNA. This is expected to adversely e81 Nucleic Acids Research, 2004, Vol. 32, No. 10 MATERIALS AND METHODS Oligonucleotide arrays 70mer oligonucleotides representing 13 443 expressed sequences from the mouse genome (Operon Array Ready Oligo Set version 1.0) were printed on SurModics 3-D Link glass slides using a robotic printing apparatus assembled according to instructions from the Pat Brown Laboratory website (http:// cmgm.stanford.edu/pbrown/mguide/index.html). The Operon 70mers were resuspended in SurModics print buffer at a concentration of 20 pmol/ml. Samples of xenotypic DNA (408 features) and sequences informatically determined to be absent from the mouse genome (320 features) served as negative controls. An additional 1436 print buffer features served as blanks for carryover control, and a select group of positive control genes was included for quantitative comparisons and statistical analysis, bringing the ®nal array size to 16 192 features (herein referred to as the 16K array). A 32 pin print head out®tted with MicroQuill 2000 print pins (Majer Precision Engineering) was used to array the features in 32 sectors, each 23 3 22 features in dimension. Slides were post-processed according to the manufacturer's protocol. Hybridizations were carried out in 53 SSC, 50% formamide and 100 ng/ml yeast tRNA, at 46°C for 72 h. Coverslips were removed in 43 SSC, 0.1% SDS; the slides were then washed twice in 13 SSC, 0.1% SDS at 67°C for 5 min, then in 0.23 SSC at room temperature for 1 min, and again in 0.13 SSC for 1 min at room temperature, before spin drying at 900 r.p.m. for 3 min in an IEC Centra GP8 centrifuge using a 216 rotor. Hybridized arrays were scanned on an Axon 4000 duallaser scanning instrument (Axon In (...truncated)