An optimized workflow for improved gene expression profiling for formalin-fixed, paraffin-embedded tumor samples

Abstract

Background: Whole genome microarray gene expression profiling is the 'gold standard' for the discovery of prognostic and predictive genetic markers for human cancers. However, suitable research material is lacking as most diagnostic samples are preserved as formalin-fixed, paraffin-embedded tissue (FFPET). We tested a new workflow and data analysis method optimized for use with FFPET samples.

Methods: Sixteen breast tumor samples were split into matched pairs and preserved as FFPET or fresh-frozen (FF). Total RNA was extracted and tested for yield and purity. RNA from FFPET samples was amplified using three different commercially available kits in parallel, and hybridized to Affymetrix GeneChip® Human Genome U133 Plus 2.0 Arrays. The array probe set was optimized in silico to exclude misdesigned and misannotated probes.

Results: FFPET samples processed using the WT-Ovation™ FFPE System V2 (NuGEN) provided 80% specificity and 97% sensitivity compared with FF samples (assuming values of 100%). In addition, in silico probe set redesign improved sequence detection sensitivity and, thus, may rescue potentially significant small-magnitude gene expression changes that could otherwise be diluted by the overall probe set background.

Conclusion: In conclusion, our FFPET-optimized workflow enables the detection of more genes than previous, nonoptimized approaches, opening new possibilities for the discovery, validation, and clinical application of mRNA biomarkers in human diseases.

Figure 1

Plot showing (A) typical poor-quality Affymetrix probe set and (B) the revised probe set used in this study.* OY axis corresponds to probe-dependent signal component, also known as affinity profile. Signals are normalized by RMA method. Black dots = ER–, red dots = ER+. * The set of 11 probes shown in Figure 1A contains seven probes that either do not match the genome and transcriptome (probes 1 to 4) or that are not sufficiently specific for their target sequence and may, therefore, detect multiple genes (probes 6 to 8). Concentrating on an unambiguous part of the probe set (Figure 1B) significantly improves the power of the statistical tests used (eg, a t-statistic of 3.07 comparing expression in estrogen receptor-negative versus expression in estrogen receptor-positive tumors as opposed to a t-statistic of 0.98 when the original Affymetrix probe set is used).

Figure 2

Pre-normalization microarray plots. (A) Fresh-frozen tissue, (B) Formalin-fixed, paraffin-embedded tissue (FFPET) processed using the Affymetrix Two-Cycle kit, (C) FFPET tissue processed using the NuGEN kit, and (D) FFPET tissue processed using the Sigma/Rubicon Genomics kit.

Figure 3

Estrogen receptor status of FFPET samples: sequence detection sensitivity and specificity by different methods. Each point of a particular color in the plot represents the effect size (difference between mean expression in ER+ and ER– samples) found in FF samples and FFPET samples using a particular workflow. Different quadrants of the plot are marked so as to assist in classifying the findings. For example, false negative (FN) findings are those which are present in FF but absent in FFPET. Presence and absence is decided based on the observed effect size with the widely accepted threshold of 2-fold change (up or down). One particular case, referred to as a wrong positive finding (WP), not observed in our experiments, is when a relevant (>1 in absolute value) effect in FF stays relevant but changes its direction in FFPET.

Figure 4

Principal components analysis plots. (A) Fresh-frozen tissue, (B) Formalin-fixed, paraffin-embedded tissue (FFPET) samples processed using the NuGEN RNA amplification kit, (C) FFPET samples processed using the Affymetrix Two-Cycle RNA amplification kit, and (D) FFPET samples processed using the Sigma/Rubicon Genomics RNA amplification kit. Note the extremely similar shape of fresh-frozen and FFPET data clusters associated with the NuGEN workflow. The NuGEN kit also provides the most parsimonious description: in order to capture the effect of estrogen receptor (ER) status on gene expression, both the Two-Cycle and the Sigma/Rubicon Genomics kit would require at least two leading principal components. Circles and triangles represent ER– and ER+ samples correspondingly. Technical replicates are shown in red.