Statistical evaluation of transcriptomic data generated using the Affymetrix one-cycle, two-cycle and IVT-Express RNA labelling protocols with the Arabidopsis ATH1 microarray

Abstract

Background: Microarrays are a powerful tool used for the determination of global RNA expression. There is an increasing requirement to focus on profiling gene expression in tissues where it is difficult to obtain large quantities of material, for example individual tissues within organs such as the root, or individual isolated cells. From such samples, it is difficult to produce the amount of RNA required for labelling and hybridisation in microarray experiments, thus a process of amplification is usually adopted. Despite the increasing use of two-cycle amplification for transcriptomic analyses on the Affymetrix ATH1 array, there has been no report investigating any potential bias in gene representation that may occur as a result.

Results: Here we compare transcriptomic data generated using Affymetrix one-cycle (standard labelling protocol), two-cycle (small-sample protocol) and IVT-Express protocols with the Affymetrix ATH1 array using Arabidopsis root samples. Results obtained with each protocol are broadly similar. However, we show that there are 35 probe sets (of a total of 22810) that are misrepresented in the two-cycle data sets. Of these, 33 probe sets were classed as mis-amplified when comparisons of two independent publicly available data sets were undertaken.

Conclusions: Given the unreliable nature of the highlighted probes, we caution against using data associated with the corresponding genes in analyses involving transcriptomic data generated with two-cycle amplification protocols. We have shown that the Affymetrix IVT-E labelling protocol produces data with less associated bias than the two-cycle protocol, and as such, would recommend this kit for new experiments that involve small samples.

Figure 1

Principal component analysis of the transcriptomic samples. Principal component analysis of the eighteen samples (three replicates each of MS one-cycle (MS-1cyc), MS two-cycle (MS-2cyc), MS IVT-E (MS-IVTE), EZ one-cycle (EZ-1cyc), EZ two-cycle (EZ-2cyc) and EZ IVT-E (EZ-IVTE)). Samples clustered closely together have a high level of similarity in expression levels; samples spread far apart have more divergent expression profiles. One-cycle and IVT-E samples cluster nearby each other for each tissue, whilst the two-cycle data sets are divergent.

Figure 2

Mean bias statistics for the three protocols. Mean bias statistics for the three protocols across the MS and EZ zones and three replicates. More negative numbers indicate increased bias.

Figure 3

A Venn diagram of the differentially expressed genes using each protocol. A proportional Venn diagram showing that lists of differentially regulated loci between MS and EZ tissues created using one- (left circle), two-cycle (right circle) and IVT-E (bottom circle) data sets show a large degree of overlap. Numbers in the segments refer to the number of corresponding loci.