Microarray gene expression profiling of formalin-fixed paraffin-embedded (FFPE) tissues is a new and evolving technique. This report compares transcript detection rates on Affymetrix U133 Plus 2.0 and Human Exon 1.0 ST GeneChips across several RNA extraction and target labeling protocols, using routinely collected archival FFPE samples. All RNA extraction protocols tested (Ambion-Optimum, Ambion-RecoverAll, and Qiagen-RNeasy FFPE) provided extracts suitable for microarray hybridization. Compared with Affymetrix One-Cycle labeled extracts, NuGEN system protocols utilizing oligo(dT) and random hexamer primers, and cDNA target preparations instead of cRNA, achieved percent present rates up to 55% on Plus 2.0 arrays. Based on two paired-sample analyses, at 90% specificity this equalled an average 30 percentage-point increase (from 50% to 80%) in FFPE transcript sensitivity relative to fresh frozen tissues, which we have assumed to have 100% sensitivity and specificity. The high content of Exon arrays, with multiple probe sets per exon, improved FFPE sensitivity to 92% at 96% specificity, corresponding to an absolute increase of ~600 genes over Plus 2.0 arrays. While larger series are needed to confirm high correspondence between fresh-frozen and FFPE expression patterns, these data suggest that both Plus 2.0 and Exon arrays are suitable platforms for FFPE microarray expression analyses.

Introduction

It is now possible to obtain clinically useful gene expression profiles of formalin-fixed and paraffin-embedded (FFPE) tumor tissues using a range of microarray platforms, including custom-made two-color spotted arrays and commercial platforms from Illumina (San Diego, CA, USA) (1,2), Affymetrix (Santa Clara, CA, USA) (3,4), and Agilent Technologies (Santa Clara, CA, USA) (5). FFPE is the worldwide standard tissue preservation method; however, the fixation process—as well as time in storage—compromises RNA extraction potential and quality and its consequent utility for molecular biology applications such as microarray gene expression profiling (6,7).

The development of FFPE microarray protocols, therefore, represents a long-awaited opportunity to access vast FFPE archives with well-annotated clinical data for the molecular study of human cancers, especially rare tumors where a lack of fresh-frozen tissue has prevented microarray studies using standard protocols.

While the success of recently developed FFPE protocols has been met with enthusiasm from the scientific community, several limitations need to be overcome in order for this approach to reach its full potential. To maximally exploit existing FFPE supplies, protocols must be suitable for routinely processed archival FFPE tissues, where widely varying processing and storage conditions can significantly affect RNA degradation and chemical modification rates (8,9) and, in turn, priming for cDNA synthesis and array performance. The custom-developed Arcturus Paradise system (Molecular Devices, Sunnyvale, CA, USA) for FFPE and Agilent 3′-based custom arrays, for example, was useful for interrogating FFPE tissues prepared under optimal conditions [even after several years in storage (5)], but unsuccessful in up to 76% of routinely processed samples (10). One reason may be the use of relatively long 60-mer target probes, as experience with RT-PCR suggests that shorter probe sets are better at detecting extensively degraded RNAs typical of routinely processed FFPE (11). It is therefore likely that shorter microarray probes such as the 25-mer lengths employed in Affymetrix GeneChips will prove to be better for degraded RNAs (7). Furthermore, most custom arrays for FFPE templates have historically featured a preferential 3′-biased probe set design to accommodate degraded RNA and the standard use of oligo(dT) priming methods. However, even with short probe sets, the rationale for using oligo(dT)-only approaches has recently been questioned (3) since (i) this method will not label non-polyA–tailed fragments and (ii) adenine residues have higher rates of formalin modification than other bases (12). Indeed, gene detection from FFPE has been found to be significantly better when random hexamer priming is used (13,14).

In this study, we examined routinely processed archival FFPE and the impact of different RNA extraction and labeling protocols on final microarray data—using Affymetrix HG U133 Plus 2.0, X3P, and Exon 1.0 ST GeneChips—in order to identify optimal methods for interrogating archival fixed tissues.

Materials and methods

Sections of soft tissue sarcoma (STS) (including six leiomyosarcomas, three liposarcomas, three synovial sarcomas, and one unspecified pleomorphic spindle tumor) were cut from 1–8 year-old routine diagnostic FFPE tissue blocks and processed for RNA extraction as previously described (3). The number of tissue sections required depended on the labeling protocol and tissue cellularity; generally 10–12 sections each containing up to 1 cm² tissue were required to yield sufficient RNA for One-Cycle non-amplification reactions, while 2–3 sections were needed for labeling reactions incorporating an amplification step. In two sarcoma cases, paired unfixed (fresh-frozen) tissues were taken from the same biopsy sent for routine FFPE processing, processed as previously described (3) and used as gold standards for measurement of FFPE array performance (sensitivity and specificity of gene detection).

We previously extracted RNA from STS using an in-house modification of the Optimum FFPE RNA Isolation Kit (Ambion Diagnostics, Austin, TX, USA) (3), which is no longer on the market. In this study, we re-extracted RNA from the same samples using the RNeasy FFPE Kit (Qiagen, Hilden, Germany), having previously obtained more consistent extract results with the latter kit than the RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Ambion Diagnostics) (testing was carried out on cervical carcinoma sections; data not shown). RNA extracts were DNase-treated and technical replicates were labeled using Affymetrix protocols (One-Cycle, Two-Cycle, and Whole Transcript for Exon arrays) and NuGEN WT-Ovation systems (Pico, FFPE v2) for Plus 2.0 arrays and FFPE v2 for Exon arrays coupled with NuGEN FL-Ovation cDNA Biotin Module V2 (San Carlos, CA, USA). Samples were in turn hybridized to three Affymetrix microarray platforms for performance comparison: Human Genome U133 Plus 2.0, X3P, and Human Exon 1.0 ST (all from Affymetrix). Manufacturer's instructions were followed throughout and full details of extraction, labeling and hybridization protocols, and raw array data (.cel files) are available at http://bioinformatics.picr.man.ac.uk/vice.