As researchers have been running microarray experiments with smaller and smaller amounts of starting material, protocols for preparing and amplifying sufficient RNA have flourished.
Earlier this month, for example, NuGen Technologies launched a new RNA amplification and labeling kit, called Ovation Biotin System, which its says requires only five nanograms of RNA as starting material. Affymetrix, Ambion, Arcturus, and others also offer RNA amplification tools.
But amplification protocols can also be a source of error, and “data generated on the same platform using different protocols may not be directly comparable,” according to researchers at the Paterson Institute for Cancer Research in Manchester, UK, who recently conducted a study. The scientists compared results from Affymetrix GeneChip experiments using two different Affymetrix sample protocols. “Clearly, comparison between replicates generated with different protocols is dangerous,” the authors wrote in the study.
But there is a silver lining to this sample prep variation: Entire experiments using different protocols can be compared, the researchers found, by correcting systematic errors during the data analysis step.
The study, “Amplification protocols introduce systematic but reproducible errors into gene expression studies,” was published in the March issue of BioTechniques.
The researchers, led by Crispin Miller, who heads the bioinformatics group at the Paterson Institute, studied gene expression differences in two cell lines, using HG-U133A chips, and compared results obtained from two Affymetrix RNA preparation protocols that use different amounts of starting material.
The standard protocol requires 10 µg of total RNA. It involves converting messenger RNA into double-stranded DNA, which is then transcribed into biotinylated RNA.
The small samples protocol only requires 10 to 100 ng of total RNA to start with, according to the comp any, and adds a second round of DNA and RNA synthesis.
When the researchers conducted an entire experiment using a single protocol, they saw similar results between experiments, no matter which protocol they used, because they were able to eliminate systematic errors by calculating the fold change.
However, when they directly compared arrays using different protocols, they found many discrepancies between the datasets. “When one mixes and matches arrays processed using different protocols, we see significant disparities in the apparent fold-changes,” Miller told BioArray News.
“The take home message, therefore, is that we need to be careful if we are planning to use more than one protocol during the course of a study, evaluate them in order to find out how well they perform, and avoid mixing and matching protocols unless we can be sure that this is not going to be an issue,” according to Miller.
The authors proposed in their study that researchers use a reference pool of RNA to generate an expression pattern for each sample protocol, and use these reference RNAs to normalize. “Alternatively, we must accept that only calculations that side step systematic errors (such as fold change) are appropriate for analyzing microarray data,” they wrote.
Systematic errors, they suggested, are something scientists simply have to learn to live with: “Given that the systematic errors introduced by different RNA preparation protocols are probably unavoidable and likely to be beyond the control of the biologist, data analysis techniques should be chosen with these effects in mind,” they wrote.