Total RNA is ideal for gene expression profiling from clinical specimens, small amounts can be used because the methodology preserves the relative abundance of the different mRNAs in the original sample [4
]. In evaluating a new protocol or comparing existing protocols, measures such as the yield of cRNA or the fraction of probe sets detected can be useful, but the key measure is the extent to which differences (or in this analysis similarities) in gene expression can be detected. Several studies have compared linear amplification protocols, similar to those used here, with PCR-based techniques [5
], or one-cycle labeling kits to those that use two or three cycles for sample of 100 ng or less total RNA [7
]. Ma et al
. (2006) [8
] compared IVT labeling reactions from purified double-stranded cDNA using the GE Healthcare CodeLink Expression System and the 3'-amplification reagents from Affymetrix, showing similar results to those obtained here. However, a literature search revealed no publications comparing one-cycle labeling kits directly. To examine the differences introduced by different labeling kits, we compared 3 one-cycle labeling kits commonly used with the Affymetrix GeneChip platform. Each labeling method showed a high degree of intra-method correlations in replicate experiments, but lower inter-method correlation. Exploring the fold changes between Donor A and Donor B across kits (Figure ) and the MA plots (Figure ) showed the One-Cycle and BioArray kits, gave the best correlation. The poorest concordance was evidenced between the two Eberwine-based protocols, One-Cycle and Superscript.
The use of the oligo(dT) primer in the first strand synthesis in all the kits is to selectively amplify polyadenylated transcripts. Therefore the ribosomal RNA probesets (to 18S and 28S rRNA) act as a negative control. As expected for these probesets, the One-Cycle kit gave a signal just above background, however, for both the BioArray and Superscript kits a strong positive signal was detected for the 18S rRNA probesets (Figure ). This is of concern, as any signal detected comes from either cross-hybridization or non-specific priming. We believe these results can be explained by the rapid cooling of the primer annealing reaction in the One-Cycle methodology which reduces some of the non-specific binding that may arise with the slower cooling used in the Superscript or BioArray protocols.
β-actin and GAPDH internal control probesets are commonly used to assess RNA sample and assay quality [3
]. The other probesets classified as internal controls are rarely referred to in the literature, probably because the two rRNA probesets (discussed above), are known to have very high coefficients of variation [11
] and the ISGF3 probesets cover 2639 nucleotides of the mRNA, a particularly long transcript for Affymetrix GeneChips.
Other possible causes for the differences in results between the labeling kits are the use of different reverse transcriptase enzymes, T7oligo dT primers (which could differ in promoter sequence and length of poly-T tail) and concentration of nucleotides. BioArray uses a very different strategy to generate the second strand in the cDNA synthesis – a homopolymeric tail to the cDNA, followed by an extension reaction. The other kits use a nick-translation reaction. The IVT reaction for amplification and labeling of the target is similar in most aspects, except the BioArray method uses two nucleotide labels (biotin-CTP and biotin-UTP), the Superscript kit a single biotin-UTP labeling and the One-Cycle, a ψ-UTP biotinylated nucleotide (Table ). Samples labeled by the BioArray method had higher, unnormalized fluorescence intensity values than the other methods (Table ), possibly due to incorporation of the two labeled nucleotides. The higher number of present calls on the One-Cycle kit array indicates a higher labeling efficiency, but this did not appear to be because of greater label incorporation, as signal intensities were no higher in the One-Cycle kit arrays compared to the other two.
Comparison of labeling methodologies for kits compared in this study.
The methodological differences elucidated above, will each make a contribution towards the differences in cRNA profiles that were noted when an aliquot of each was run on the Bioanalyzer (Figure ). These differences were not a result of loading different quantities of cRNA on the Bioanalyzer chips, as similar results were obtained when equivalent amounts of cRNA were run. The differences included length of product, yield and reproducibility of the methodology. A cRNA profile with a greater proportion of fragments above 500 nt is usually considered as a good preparation, therefore despite the differences, each of the kits provides target cRNA that passes the suggested parameters established by Affymetrix and our results support this. Once the biotinylated target cRNA has been synthesized, it is then fragmented by metal-induced hydrolysis. Considering that a set amount of cRNA goes into each reaction (20 μg) and the same protocol was used for all the fragmentations, we were surprised to note the differences in the fragmentation profiles for each kit. The BioArray and Superscript protocols give far smaller yields when compared to the One-Cycle kit cRNA. The 3 kits each use spin-cartridges with differing reagents to clean-up the cRNA, and the product is eluted in RNAse-free water in each case. The BioArray and Superscript kits require that the target be concentrated in a vacuum concentrator for fragmentation. We cannot be sure what the carry-over (i.e. ion concentration) is from these steps, and it is possible that this could affect the fragmentation reaction, possibly causing complete hydrolysis and loss of product.
Probably the largest draw back that we experienced with any of the kits was the number of times that the BioArray kit failed to produce sufficient product for fragmentation (50% of reactions). This is probably due to losses incurred from the two additional purification steps (after the reverse transcription and again after the homopolymeric tailing) in this methodology. This is not an acceptable situation when processing limited clinical samples. Enzo Life Sciences has recently released a new labeling kit based on the Eberwine protocol, which they currently recommend for use with Affymetrix GeneChips. Other considerations we used in evaluating the kits were ease of use, measured by the number of steps involved in the protocol and cost of each labeling reaction (Table ).