Gene expression profiling classifies individual tumors by their gene expression patterns and may also describe and predict therapeutic resistance and sensitivity patterns. Profiling in several cancers, such as breast cancer, colon cancer, lymphoma, leukemia, and melanoma , has already identified molecular subclasses of tumors. Identification of tumor subtypes may be predictive for prognosis or response to drug therapy [6, 7, 28–31].
The potential of routine gene expression profiling to predict clinical outcomes for cancer patients has yet to be determined. The Evaluation of Genomic Applications in Practice and Prevention Working Group stated in 2009 that there was “insufficient evidence to make a recommendation for or against the use of tumor gene expression profiles to improve outcomes in defined populations of women with breast cancer” . Clearly, more work needs to be done to translate promising research findings into clinically relevant results.
Comparison of FFPET sample-derived gene expression profiles with associated clinical outcome data could provide an invaluable resource for the investigation of mechanisms of response and resistance to cancer therapeutics. This will, however, require a suitably reliable method for extracting RNA from these samples and an FFPET-optimized workflow for use in gene expression profiling studies. Two recent studies performed using archived FFPET core biopsies from the NOAH trial (neoadjuvant trastuzumab with chemotherapy in patients with locally advanced/inflammatory breast cancer)  have demonstrated that biomarkers predictive of treatment response can be identified from FFPET processed with an optimized workflow such as that described here [34, 35]. These findings speak to a vast untapped potential for biomarker discovery due to the abundance of archived tissues available for gene expression profiling.
FFPET samples could also be used prospectively to identify biomarkers in clinical trials. Biomarkers discovered via microarray gene expression profiling could then be applied clinically using more widely available technologies, such as immunohistochemistry or quantitative rtPCR. Recently, researchers conducting a multicenter randomized phase III trial in colon cancer have suggested that translational studies based on FFPET samples should be included in prospective clinical trials .
Methods and protocols for RNA isolation from formalin-fixed tissues have been available for over 20 years  but results have been variable. Research continues to develop optimized protocols for formalin fixation  and RNA extraction methods .
The work reported here shows that there is no loss of data if RNA amplification and in silico probe design protocols are optimized towards FFPET samples (which were >12 months old). Specificity of gene detection was 99% using the Affymetrix method, 97% using the NuGEN method, and 100% using the Sigma/Rubicon Genomics method. The gene detection sensitivity results obtained from FFPET samples varied depending on which of the three different amplification kits was used. The combination of the NuGEN RNA purification/amplification method with the Affymetrix Human Genome U133 Plus 2.0 Array resulted in 80% sensitivity with 97% specificity. This agrees with Linton et al.  who also reported 80% sensitivity, assuming FF sensitivity to be 100%, using the same NuGEN amplification and labeling system when comparing sarcoma FFPET with FF samples. Williams et al.  have also demonstrated the superiority of the NuGEN method over the Affymetrix method. It should be noted that newly developed FFPET amplification systems (such as Affymetrix Sensation Plus®) may increase the sensitivity and specificity of signal detection from FFPET. Furthermore, detailed feasibility studies with FFPET samples that are >2 years old should be carried out to provide further evidence that archival tissue is suitable for gene expression profiling.
It is also important to show that the genes identified in this new workflow are an accurate reflection of the biology of the sample. The Ingenuity Pathways Analysis shows that the same ER signaling pathway is identified regardless of whether the sample was FF or FFPET, and that therefore the FFPET data obtained are a true reflection of the biology of the tissue sample.
The correlation between gene expression profiling results obtained from FF and FFPET samples is greatly improved using the new array design and data processing methods reported here. A comprehensive gene expression profile is necessary to both the discovery of novel biomarkers and the development of new disease subclassifications. Our optimized sample processing workflow and data analysis method enables the detection of genes that cannot be detected using a nonoptimized approach. The gene detection rate reported here for FFPET-derived samples is similar to that obtained using FF samples, demonstrating the importance of optimal RNA amplification and raw data processing with refined probe set alignments, which could facilitate the discovery of biomarkers from FFPET samples. In addition, other commercial microarrays are now available which may overcome the current limitations associated with the Affymetrix GeneChip.
The ‘percentage of genes present’ call is a useful tool to evaluate the overall quality of gene expression data. However, if the percentage present call is to be used as a quality control measure, then it may be advisable to set a precise threshold for discarding poor-quality samples so that potentially useful data – such as single gene profiles obtained from individual samples – are not unnecessarily lost. Consequently, the homogenous distribution of percentage present calls in the sample population may be a more useful quality control measure than percentage present call alone. Using this method, a sample with a markedly low percentage present call could be identified as an outlier even in a genome-wide analysis.
The different techniques and kits used in gene expression profiling render direct comparisons with previously reported data problematic. Although the redesigned probe set used here is as close to the original Affymetrix design as possible, comparisons with previously reported data sets using the Affymetrix array may also be problematic for several reasons. For example, differences in FFPET techniques between laboratories result in variable degradation of sample RNA. In addition, the age of the sample can impact results, although Tanney and Kennedy have reported that it is possible to extract good-quality transcriptional data from samples as old as 17 years using the Roche High Pure FFPE extraction kit and NuGEN WT-Ovation FFPE RNA amplification system . The quality of gene expression data can also vary according to the RNA extraction and amplification methods used.
Handling and fixation protocols vary between centers and evolve over time, so the genetic variation identified between tumors could be an artifact of tissue processing rather than a reflection of tumor biology. To overcome this, it may be important to pool samples from several centers to ensure that they represent the population under consideration as a whole . This would require careful study design to avoid adding confounding factors.
It is also important to ensure that the sample size selected is sufficient for the study to be adequately powered to detect differences in gene expression. Most tumor studies are aimed at finding genes with at least twofold higher or lower expression than found in normal tissues. Power analysis for this study (data not shown) demonstrated that a protocol with 10 samples per category was 40% powered to detect a twofold change in expression, but could reliably detect a threefold change.