Evaluation of suitable reference genes for normalization of real-time reverse transcription PCR analysis in colon cancer

qRT-PCR is one of the most sensitive methods for mRNA detection and quantification. The method has also become the preferred method for validating results obtained by other techniques, such as microarray [1]. There are differences among different qRT-PCR assays due to biological and technical variations [2, 3]. In order to identify truly gene specific variations it is important to use a suitable normalization method. One of the most commonly used approaches involves relative quantification of target genes against one or more reference genes which are thought to be stably expressed in the examined tissue [4]. There have been a number of reports that demonstrate that the expression levels of putative reference genes vary extensively in different tissues and diseases and thus are unsuitable for normalization purposes [5‐15]. Consequently, each research group has to validate multiple reference genes in their own experimental setup and normalize qRT-PCR data against a few reference genes tested from independent pathways using at least one algorithm. It appears that improvements in methods of identifying reference genes are more important than the identification of the particular reference genes themselves [16].

It has been argued for use of multiple genes in the normalization of qRT-PCR analysis and several algorithms have been developed [17‐20]. Vandesompele et al., 2002, used the geometric mean of the most stable genes to improve the accuracy of the analysis in a method called geNorm [19]. This method relies on the principle that the expression ratio of two ideal reference genes is identical in all samples regardless of the experimental conditions. For every reference gene geNorm determine the pairwise variation with all other reference genes. The average pairwise variation of a particular gene is defined as the internal control stability measure; M. Genes with the lowest M values are the most stable ones. Another algorithm in which the expressional stability of genes is evaluated is NormFinder [17]. NormFinder estimates the intra-group and the inter-group expression variation. Both of these sources of variation are combined into a stability value. This method can account for heterogeneity of the tested tissue samples. Genes with the lowest stability value have the most stable expression.

Colorectal cancer is among the most frequent malignant diseases worldwide, and is one of the leading causes of cancer-related deaths [21]. The majority of colorectal tumours develop along a well-defined adenoma-carcinoma sequence in which oncogenes are activated and tumour suppressor genes lose their function [22]. Despite a high 5-year survival rate in early colorectal cancer, only 10% of the patients with distant metastases survive after five years [23]. Thus, it is important to elucidate the biology that contributes to this progression, especially those processes that facilitates the switch to invasive and metastatic disease. Biological changes are a result of partly differential gene expression, which can be confirmed by qRT-PCR. It is necessary to validate reference genes in the particular experimental system in order to trust the differential gene expressions which are detected. Previous studies have tried to find universally stable reference genes across several types of cancers, including colon cancer [24‐26]. Recent reports, however, claim that stably expressed genes in one tumour type may not predict stable expression in another tumour type [12, 27]. Moreover, results in one tumour type, like colorectal cancer, show stably expressed genes in one experimental in which are different from the stably expressed genes in another experimental setup [28‐30]. Hence, reference genes should be validated and selected in every experiment in any tissue type. Recently, it has been suggested that the focus should be on introducing and validating novel approach for reference gene identification and standardizing experimental setup rather than giving general suggestions for different tissues [16]. Applying TaqMan Low Density Array (TLDA) to examining reference genes is a step towards a more standardized experimental setup. TLDA was evaluated in colorectal cancer by Lü et al., 2008, as a roughly robust and labour-saving method for gene quantification compared with routine qRT-PCR [31]. Well-designed TaqMan probes require little optimization, and TLDA allows simultaneously real-time detection of many gene products in several samples offering higher through put than established single array method [31, 32]. Hence, in the present study we used TLDA to find potential reference genes for data normalization in qRT-PCR experiments in metastatic and non-metastatic colon cancer patients. The gene expression of 16 commonly used reference genes in tumour tissue and individual-matched normal mucosa of metastatic and non-metastatic colon cancer patients were analyzed and the expression stability was determined and compared using geNorm and NormFinder.

qRT-PCR has been a breakthrough for the quantification of gene expression in many biological systems. In this study we assume that no single gene stays unaffected under malignant development in colon cancer and therefore identify genes with least variation. We identified two pairs of genes, HPRT1/PPIA and IPO8/PPIA, which may be suitable to normalize gene expression data in studies conducted in metastatic and non-metastatic colon cancer patients. In addition, we found that B2M, ACTB and 18S were unstable in the tissue examined. We propose a standardized approach of finding the most suitable reference gene(s) in every qRT-PCR experiment using TLDA.

Complex signalling pathways have been related to the metastatic progression of colon cancer, hence pathway based gene expression assays, often revealed by qRT-PCR, are significant in cancer biology. Publications dealing with colon cancer have reported gene expression studies in metastatic diseases [34, 35]. However, the stability of the reference gene expression in metastatic and non-metastatic primary tumours remains crucial. Ramaswamy et al., 2003, described a gene expression signature that distinguished primary and metastatic adenocarcinomas, indicating that the metastatic potential of human tumours is encoded in the bulk of the primary tumour [36], fully in accordance with modern clonal stem cell theories [37]. Hence, one may presume that the metastatic capacity of the primary tumour will influence commonly chosen reference genes.

The most recent study of reference genes in colon cancer was reported by Kheirelseid et al., 2010, where 64 colorectal tumours and tumour associated normal specimens were examined using qRT-PCR followed by three different statistical algorithms, geNorm, NormFinder and qBaseplus [30]. Kheirelseid et al., 2010, found that the combination of two reference genes, B2M and PPIA, more accurately normalized qRT-PCR data in colorectal cancer. This is in concordance with our findings, where PPIA was one of the two genes identified as the most stable pair. In contrast, B2M was identified as one of the most variable genes in the tissue examined. This disparity may be explained by the difference in patient material since Kheirelseid et al., 2010, included all stages of colon cancer and even included rectum tumour samples. Furthermore the percentage of tumour cells in the samples was not addressed. In the study of Kheirelseid et al., 2010, all three algorithms confirmed the selection of the B2M and PPIA pairing as the best combination of reference genes. In the present study however, the geNorm algorithm differs from the results obtained by NormFinder. According to geNorm HPRT1 and PPIA were the most suitable genes for normalization, but NormFinder suggested IPO8 and PPIA. This discrepancy confirms previous results reported by Caradec et al., 2010, concluding that the evaluation of suitable reference genes dramatically differs according to the statistical method used [12]. Caradec et al., 2010, investigated reference genes in four cell lines treated with four different oxygen concentrations, and observed large variations in gene expression results depending of statistical method used. Thus Caradec et al., 2010, recommended Ct coefficients of variation (CtCV%) calculation for each reference gene for validation of the statistical methods. It is defined as the ratio of the standard deviation to the mean. Genes with low CtCV% value indicate more stable expression of those genes. In the present study, IPO8 was the most stable gene on the basis of CtCV% (5.12%), followed by GUSB (5.55%) and HPRT1 (6.04%) as the second and third most stable gene. Using NormFinder IPO8 was one of the genes which were identified as the most stable pair of genes, which may indicate that the CtCV% verifies the NormFinder results. Nevertheless, PPIA, which was suggested by both geNorm and NormFinder as one of the stable pair of genes, was ranked as the tenth most stable gene with a CtCV% of 7.34%. This may be explained by the low Ct mean of this particular gene (18.0), resulting in a relatively high CtCV% despite a low standard deviation. Another aspect which strengthens the results achieved by NormFinder compared with geNorm is the argument that geNorm lacks robustness compared with NormFinder [32]. Reports show that exclusion of one sample in a specific tissue collection led geNorm to change a suggested reference gene (18S) from being an unstable gene to one of the top ranking stable genes [32]. NormFinder also enables estimation of the variation between sample subgroups, like tumour and normal tissue, thus this algorithm can account for heterogeneity in the tested samples, which may be important considering the heterogeneity of the samples studied.

The optimal normalization will vary with study design. The most suitable reference gene in one medical condition may be regulated in other tissues or diseases. Blanquicett et al., 2002, found that 18S, S9 and GUS were the least regulated genes among 15 putative reference genes when examining tumour and normal colorectal and liver tissues [28]. Furthermore, Dydensborg et al., 2006, identified B2M as the most appropriate gene for normalizing colon carcinomas comparing to normal mucosa when they investigated seven colon adenocarcinomas containing both epithelial and stromal cells [29]. B2M was in this study identified as the least stable gene using NormFinder, and the third most variable gene using geNorm. In the present study where the tumour tissue samples consisted of more than 70% tumour cells some of the stromal cells are excluded. This might explain the discrepancies in the ranking of B2M since tumour tissue is heterogeneous and the fraction of different cells may influence the gene expression results. Moreover, different patient groups, including age and clinical background, may also give dissimilarities across studies. Experimental variations may also influence the gene expression results, though using triplicates in the qRT-PCR analysis as used in this study will diminish this variation.

Single assays qRT-PCR are time- and labour-intensive, and require relatively large amounts of cDNA and PCR reagents in multivariate gene expression studies. TLDA overcome these drawbacks since this technique allows for simultaneously detection of expression of up to 384 genes and requires less template cDNA and PCR reagents than routine qRT-PCR [1, 31, 38‐40].

In this study we applied TaqMan Low Density Array in order to identify reference genes in metastatic and non-metastatic colon cancer. The genes often used for normalization of gene expression data may be unstable and thus not suited for use, and therefore identifying stable reference genes in the specific experiment is vital for the results. The approach described herein can serve as a template to identify valid reference genes in any disease state. However, the optimal statistical approach to identify the best reference gene(s) remains to be determined. In the present study NormFinder and geNorm identified two different pairs of the most stable genes. The use of CTCV% might be a good validation of the two results. Nevertheless, the expression of target genes should be evaluated and a comparison of the effect of each pair of reference genes should be determined.