FFPE samples from 10 patients with OSCC and 10 patients with benign oral conditions, such as hyperplasia, were available for analysis. The material comprised tissue from the gingiva (tumour, n=5; control, n=2), hard palate (control, n=2), tongue (tumour, n=4; control, n=4) and buccal mucosa (tumour, n=1; control, n=2). The study was approved by the local Ethics Committee at the Umeå University (dnr 01-057 and dnr 08-003M). FFPE samples were cut into 5-μm sections and, depending on the size of the sample, 3–20 sections were pooled giving an approximate total area of 1 cm2. RNA was isolated using the High Pure RNA Paraffin kit (Roche Diagnostics GmbH, Mannheim, Germany) according to the manufacturer’s instructions. RNA from the pharyngeal squamous cell carcinoma cell line FaDu was extracted by the use of TRIzol (Invitrogen, Carlsbad, CA, USA) and used when creating standard curves for all primers in the q-PCR reaction. The purity of the RNA samples was evaluated by the 260/280-and the 260/230-nm absorbance ratio.

Primers were designed using Primer 3 software, and their specificity was evaluated using the primer search in NCBI BLAST. All amplicons were kept under 100 bp in order to optimize the conditions for the amplification of degraded RNA and were also designed to span an intron, to avoid the amplification of genomic DNA (Table I) (16). First-strand cDNA was synthesised using the Cloned AMV First-Strand cDNA Synthesis kit (Invitrogen) using 200 ng of RNA and random primers in a 20-μl reaction mixture. The cDNA reaction (1 μl) was used as template for the amplification of TUBA6, S100A6, ACTB, OAZ1, GAPDH, RPS13, RPL27 and HPRT1, respectively. The q-PCR reaction was carried out in a 20-μl reaction using IQ SYBR Green Supermix (BioRad, Hercules, CA, USA) according to the manufacturer’s recommendations. Reactions were carried out in triplicate. For a selection of genes (RPS13, OAZ1 and RPL27) all samples having a standard deviation (σ) > 0.3 between the three replicates were re-run at a later occasion for the evaluation of the reproducibility of the assay.

Ct values from the PCR reaction were transformed to relative quantities using the relative standard curve method (17). Identical PCR reactions carried out at two different time points generate different relative quantities, and a dimensionless dispersion parameter like the coefficient of variation (CV) is therefore needed to evaluate the reproducibility of the assay. CV is calculated by dividing the σ-value by the mean. Ratios, necessary for calculating the CV-values, were obtained by dividing samples to one arbitrarily chosen sample analysed with the same primer and from the same run. The σ-value and the mean for each ratio was then calculated between the two runs and divided to calculate the CV-values.

For comparing the stabilities of the candidate genes, two publicly available software packages, GeNorm, a Visual Basic application tool for Microsoft Excel, and NormFinder, a Microsoft Excel add-in, were used (18,19) according to the developer’s instructions. In NormFinder, samples were grouped according to the origin of tissue or the presence or absence of malignancy. A Mann-Whitney U test was used to pre-evaluate the stability of the genes.

Eight genes previously described to have housekeeping properties were selected for the investigation of stability in FFPE samples of normal oral tissue and SCCHN. This included a number of commonly used reference genes, e.g., ACTB and GAPDH, but also more recently suggested reference genes such as OAZ1 and TUBA6 (18,20). Genes of different functions and location in the genome were selected in order to reduce the risk of co-regulation between genes, which would influence particularly results from GeNorm calculations. However, the two ribosomal genes RPL27 and RPS13, which are likely to show correlated expression, were included, as both were reported to have stable expression over a large number of tissues and conditions in a recent study covering 13,629 gene expression arrays (20). Primers optimal for the amplification of degraded RNA were designed for each gene (21). Data on selected genes and primers are summarized in Table I.

All samples were analyzed for one gene at a time on a single PCR plate. RNA extracted from the cell line FaDu, which highly expresses all eight selected reference genes, was used in parallel for obtaining standard curves. All primers showed high efficiency (1.9-2.1) except ACTB, which had a slightly lower efficiency of 1.88 (Table I). The PCR reaction first included two replicates of each sample, but this was increased to three as standard deviations >0.3 were noted between some of the replicates. All samples run with two replicates were re-analyzed using three replicates. To investigate the reproducibility of the assay, all samples showing a σ > 0.3 for three of the genes were also re-analyzed a second time, and CV-values were calculated between the runs. The mean CV was 0.05, showing that the use of three replicates was sufficient to provide highly reproducible means. A bar graph over the means and σ-values of ratios between the two runs is shown in Fig. 1.

HPRT1, one of the candidate reference genes generally showed high Ct-values and could not be detected in some of the samples. This gene was therefore excluded from further analysis. The other seven genes were all detected between 25–35 cycles. Data for malignant and non-malignant samples are shown as separate box plots in Fig. 2. The expression ranges span from 6.2 cycles (GAPDH) to 10.7 cycles (ACTB). None of the genes showed a significantly different expression between tumours and controls using the Mann-Whitney U test. Although a significance test could indicate a gene’s unfitness as a reference, it is important to keep in mind that these calculations were based on non-normalized data relying on equal input material, which is difficult to control; hence, the reason there is a need for good reference genes. Due to this circular problem, a number of programs (GenNorm and NormFinder) using alternative strategies to simple significance testing have been developed for the evaluation of reference gene stability (10).

Ct-values were transformed to relative quantities using the standard curve method and submitted to two different software packages, GeNorm and NormFinder, for the identification of the most stable genes. GeNorm calculates the stability of genes by pairwise comparisons, generating M-values, and is based on the assumption that the ratio of two proper reference genes will be approximately similar in all samples. According to GeNorm, RPL27 and RPL13 had the most stable expression (Table IIA). Since GeNorm is very sensitive to co-regulation, this could be an artefact of the two ribosomal genes showing highly concordant expression. Data was therefore re-analysed, sequentially excluding one of the two ribosomal genes. Excluding RPL27 resulted in RPS13 being the most stable gene in combination with TUBA6, and the other way around (Table IIB).

NormFinder, which is a model-based approach that is less sensitive to co-regulation and takes both the overall expression variability as well as the variation between subgroups into consideration, also suggested TUBA6 to be the most stable reference gene, while RPS13 and RPL27 were placed second and fifth, respectively (Table IIIA).

As NormFinder takes grouping into account, it was also of interest to determine whether the same genes were stably expressed between tissues at different sites in the oral cavity. Comparing tissue from the gingival and palate (n=9) to tongue tissue (n=8), or tissue from the gingival, palate and buccla mucosa (n=12) to tongue tissue (n=8), TUBA6, RPL27 and RPS13 were the three most stable genes (Table IIIB). The samples from buccal mucosa were too few to be considered a separate group in this comparison, as there should preferably be a minimum of eight samples in each group.