Increase Signal and Detect More Genes on Affymetrix® Arrays
Performance Comparison: MessageAmp II-Biotin Enhanced vs. Leading Competitor
The two kits were compared in triplicate using a common RNA source at 4 and 14 hr IVT reaction times, with 1000 ng of starting total RNA. There was an expectation that the overall signal could be considerably different between kits, so the arrays were not normalized or scaled, because the standard assumptions for global normalization would be violated in this type of study. To reduce technical variation, all arrays were run on the same day by the same operator. In addition, to demonstrate an improvement in performance on Affymetrix microarrays, a combination of ANOVA analysis, hierarchical clustering, and graphics was used.
Analysis methods . Because the microarray is the end product of the amplification process, the first step was microarray signal quantification. Background subtraction, expression summary, and log transformation of gene signals were carried out using the Robust Multichip Average (RMA; developed by Irizarry et al. ) minus the quantile normalized function for each Affymetrix Human Focus Array. Two-way ANOVA analysis was carried out to resolve those genes that were differentially expressed either between kit methods and/or IVT times.
Results. The ANOVA analysis showed a large number of genes that differed significantly between RNA amplification and labeling methods with less of an impact when differences in IVT time were considered. A false discovery rate (FDR) was calculated using a step-up approach with a 0.05 significance value to compensate for multiple testing errors. After the FDR procedure, 4766 out of 8732 genes differed significantly between the MessageAmp II-Biotin Enhanced Kit and Kit A, using a 14 hr IVT. Figure 1 shows the sources of variation within the ANOVA model as measured by average mean square error, but this gave no indication as to which kit is superior, only that the choice of kit used had a major impact on the results.
Figure 1. Source of Variance Shows Choice of Kit for RNA Amplification Matters. A two-way ANOVA analysis was carried out using Partek® Pro 6.1 (Partek Inc., St. Charles, MO) to determine whether genes were differentially expressed either between method and/or IVT times. The average mean square for each variable in the ANOVA model denotes the contribution of each variable to the variance within the model. The error term denotes the variation not accounted for in the model. Most of the variation occurred due to choice of method for RNA amplification and labeling. However, use of a volcano plot and heat map (Figures 2 and 3) were necessary to determine which kit provided superior data. 4766 out of 8732 genes meet the 0.05 FDR Significance criterion.
Using a Volcano plot, which plots the log2 ratio between the MessageAmp II-Biotin Enhanced Kit data and Kit A data vs. the Fisher’s least significant difference pairwise comparisons method p value (Figure 2) , it can clearly be seen that the difference between kits can be attributed to increase in signal when using the MessageAmp II-Biotin Enhanced Kit. Additionally a heat map (Figure 3) created using hierarchical clustering to cluster both genes and methods shows the superior performance of MessageAmp II-Biotin Enhanced Kit due to higher signal. When each kit’s performance was measured on a gene by gene basis, the average signal was increased for 91.3% of all genes when using the MessageAmp II-Biotin Enhanced Kit. These results suggest that the aRNA generated by the MessageAmp II-Biotin Enhanced Kit will lead to arrays with higher signal intensity, allowing you to analyze more data points per input sample than a leading competitor’s kit. The histogram in Figure 4 provides another way to visualize the increase in detectable genes at the low resolution end of the spectrum.
Figure 2. Volcano Plot of MessageAmp™ II-Biotin Enhanced Kit vs Kit A. Plot of the log2 ratio between the MessageAmp™ II-Biotin Enhanced Kit data and Kit A data vs. the Fisher’s least significant difference pairwise comparisons method p value. (i.e., T test is performed just on differentially expressed data points, as identified by prior ANOVA analysis ). This plot shows higher signal intensity for data generated from the RNA amplified with the MessageAmp II-Biotin Enhanced Kit vs. Kit A.
Figure 3. Heat Map of Expression Values for MessageAmp II-Biotin Enhanced Kit and Kit A. Heat map comparing the four experimental conditions: MessageAmp™ II-Biotin Enhanced Kit vs. Kit A using a 4 and 14 hr IVT reaction. (Three arrays per condition are plotted via hierarchical clustering using Euclidean distance and average linkage.) Signal intensity is represented by color in a log2 scale from 1–14, where red is the highest and green is the lowest. There is very little difference between 4 and 14/16 hr IVT as evidenced in the heat map.
Figure 4. Histogram of Signal Distribution of All Array Features. An alternative presentation of the heat map data in Figure 3, distributions of signal intensities in histogram format illustrate that the Ambion MessageAmp™-II Biotin Enhanced Kit provides greater signal intensity (Median for Kit A 5.776 vs. Ambion 6.374). The dark shaded areas denote the samples at the lower end of the signal distribution (log transformed) and their increase between kits. Signal was increased for 91.3% of all genes.
Arrays contain a huge number of data points. Most of the interesting changes in expression are of low resolution, and these are the ones scientists are most interested in investigating. The data presented here shows that the MessageAmp II-Biotin Enhanced Kit provides increased signal and increased Present calls, as well as more data points at the low resolution end of the spectrum.
MessageAmp II-Biotin Enhanced Kit
Charlie Johnson, Robert Setterquist, Sharmili Moturi, Shikha Agarwal, Penn Whitley • Ambion, Inc.