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

Background Real-time reverse transcription PCR (qRT-PCR) is frequently used for gene expression quantification due to its methodological reproducibility and sensitivity. The gene expression is quantified by normalization to one or more reference genes which are presumed stably expressed throughout a given experiment. The aim of this study was to validate a standardized experimental setup to identifying reference genes for normalization of qRT-PCR in the metastatic and non-metastatic colon cancer. Methods In this study, expression of 16 commonly used reference genes was quantified in tumour tissue and individual-matched normal mucosa in 18 non-metastatic colon cancer patients and 20 colon cancer patients with distant metastases using TaqMan Low Density Array (TLDA). The expression stability was determined and compared by means of geNorm and NormFinder. Results Two pairs of genes, HPRT1/PPIA and IPO8/PPIA, were identified to be suitable to normalize gene expression data in metastatic and non-metastatic colon cancer patients, according to geNorm and NormFinder respectively. Conclusion We propose a standardized approach of finding the most suitable reference gene(s) in every qRT-PCR experiment using TLDA.

Background 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][6][7][8][9][10][11][12][13][14][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][18][19][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][25][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][29][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 laboursaving method for gene quantification compared with routine qRT-PCR [31]. Well-designed TaqMan probes require little optimization, and TLDA allows simultaneously realtime 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 nonmetastatic 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.

Patients and tissue specimens
RNAlater-stored tumour tissue samples and individualmatched normal mucosa were obtained from 38 patients with colonic adenocarcinoma who underwent resection at Akershus University Hospital Trust between 2004 and 2009. The dissected tissue samples were collected in the operating room and stored immediately in approximately five volumes of RNAlater (Ambion Inc., Austin TX, USA) and frozen at -80°C. Eighteen patients with non-metastatic disease, Dukes B (with a minimum of 12 negative lymph nodes) where no metastases occurred during 5 years follow up, and 20 patients originally staged as Duke C who displayed distant metastases during a 5 year follow-up (Duke C) or patients classified as Dukes D were included in the study. There were 22 women and 16 men with a mean age of 69 +/-14 years (range 29-92) at surgery. Three sectioned pieces of the tumour samples were made. The central piece was further processed for RNA isolation, while the two end pieces were fixed in formalin and embedded in paraffin (FFPE). Four μm sections of FFPE samples were stained with Hagens Hematoxylin and examined by a pathologist for determination of percentage tumour cells. To avoid bias from necrosis or minimal tumour representation we included tumour tissue samples with more than 70% tumour cells.

mRNA isolation
Total RNA isolation was performed using the method of Wei and Khan, 2002, [33] modified according to T. Lüders (unpublished work) to also include miRNA for further analyses. Approximately 60 mg frozen tissue was homogenized in TriReagent (Ambion) using Mixer Mill MM301 (Retch) for 2 × 2 min at 30 Hz. After phase-separation with chloroform, the aqueous phase (containing RNA) was mixed with 1.5 volumes 100% ethanol and transferred to an RNeasy Mini spin column (Qiagen). Further processing was performed following the manufacturer's protocol. A DNase treatment was included in the procedure. RNA was eluted in 60 μl RNase-free water and stored at -80°C. The concentration of each RNA sample was obtained from A 260 measurements using the NanoDrop 2000 (Thermo Fischer Scientific Inc.). The RNA integrity number (RIN) was tested by using the Agilent 2100 Bioanalyzer (Agilent Technologies).

cDNA synthesis
Complementary DNAs (cDNAs) were produced from 1 μg RNA of each sample using the High Capacity RNAto-cDNA Master Mix (Applied Biosystems) according to the manufacturer's instructions. The following thermal cycler conditions were used: 5 min at 25°C, 30 min at 42°C and 5 min at 85°C. Three random RNA samples were additionally run in the absence of reverse transcriptase enzyme to assess the degree of contaminating genomic DNA. Real-time PCR with genomic DNA specific assay revealed that RNA was free of genomic DNA (data not shown).

TLDA design and preparation
TaqMan Endogenous Control Assays (Applied Biosystems) are 384-well microfluidic cards containing 16 preoptimized human TaqMan Gene Expression Assays commonly used as endogenous controls and genes that exhibit minimal differential expression across different tissues ( Table 1). The assay was performed in triplicates. 50 μl cDNA (1 μg mRNA) was used as a template. Matched samples from 4 patients where loaded on each card. NTC (no template control) was added in one loading port. PCR amplification was performed using the ABI Prism 7900 HT Real Time PCR System (Perkin-Elmer Applied Biosystems, Foster City, California, USA). Thermal cycling conditions were used as follows: 2 min at 50°C, 10 min at 94.5°C, 30 sec at 97°C, and 1 min at 59.7°C for 40 cycles.

TLDA analysis and Statistical analysis
RealTime Statminer 3.0 Software (Integromics, Madrid, Spain) was used for implementation of quality controls in addition to calculation of optimal endogenous controls.
This program uses the comparative Ct method for relative quantification analysis, and the results are expressed as a fold change of expression levels (DDCt values). The mean value of triplicates was applied for all calculations. Medians were used to replace missing values that occurred due to inconsistencies between replicates rather than from low expression. The detectability threshold was set to 36, meaning failing detectors are those with a Ct greater than or equal to 36. To measure the expressional stability of the candidate endogenous control genes, two commonly used programs were employed: geNorm [19] and NormFinder [17], both of which available in RealTime Statminer. Ct coefficients of variations (CtCV%) were calculated for every reference gene across all samples. All data are expressed as means ± SD. Except from the analyses in RealTime Statminer, all other calculations were performed using SPSS (version 14.0; SPSS, Chicago, IL, USA).

Research Ethics
This project was approved by the Regional Committee for Medical Research Ethics, Eastern Norway. The Norwegian Social Science Data Service has approved the collection and analysis of data.

Results
RNA quality control mRNAs of 16 potential reference genes were quantified by qRT-PCR using equal amounts of RNA templates from every tissue samples. The integrity of RNA (RIN)

Results of validation programs
In order to determine the stability of genes and thus find the best endogenous controls, the data were analysed by geNorm and NormFinder. In these analyses, medians were used to replace missing values because they occurred due to inconsistencies between replicates rather than from low expression. The ranking of the gene expression stability values (M) of the tested endogenous control genes using geNorm is illustrated in Figure 1.A. The genes with the highest M, i.e. the least stable genes, gets stepwise excluded until the most stable genes remain. The best two genes are ranked without distinguishing between them. HPRT1 and PPIA were identified as the most stable pair of genes, followed by PGK1 as the third most stable gene. Furthermore, pairwise variation were also calculated using geNorm in order to determine the optimal number of genes required for normalization, Figure 1.B. The analysis showed that HPRT1 and PPIA may be sufficient for calculation of the normalization factor and normalization to genes of interest, since the V2/3 value is in this analysis equal to the cut-off value of 0.15 [19]. However, there is a gradual decrease in the pairwise variability plot and thereby an improvement to the normalization factor by adding additional genes to the calculation. Nevertheless, two or three genes would be satisfactory for normalization according to the cut-off value of 0.15. While geNorm uses a pairwise comparison approach, NormFinder first estimates the intra-group and then the inter-group variability of expression of a control gene [17]. In contrast to the geNorm results, NormFinder ranked RPLP0 as the most stable gene, with TBP and GUSB closely behind as second and third, respectively ( Figure 2). However, using this algorithm the combination of IPO8 and PPIA turned out to have a lower stability score than the most stable single gene. Thus this combination is more suitable for normalizing qPCR. There was considerably closer agreement between the geNorm and Normfinder results on the least stable genes, with the order of 4 out of 5 worst ranking genes being identical; ACTB, 18S, B2M and TFRC. These genes had a stability value more than twice so high (geNorm) and more than 3 times so high (NormFinder) as the best ranking genes. Due to different ranking by geNorm and NormFinder of the most stable genes, cycle threshold coefficient of variation (CtCV%) was calculated for each of them. This calculation was recommended by Caradec et al., 2010, in order to validate the NormFinder and geNorm results [12]. According to the CtCV% calculation, one of the NormFinder pairing genes, IPO8, was ranked as the most stable gene with a CtCV% of 5.12%, which supports the NormFinder result. GUSB (5.5%) and HPRT1 (6.04%) are ranked as the second and third respectively, which do not give identical ranking of results obtain using geNorm and NormFinder. The least stable gene using CtCV% was 18S (14.99%), which was according to geNorm and NormFinder ranked as the second and fifth least stable gene, respectively. The summary of the best ranking genes as determined by each of these calculations is illustrated in Table 4.

Discussion
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  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 Norm-Finder 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][39][40].

Conclusions
In this study we applied TaqMan Low Density Array in order to identify reference genes in metastatic and nonmetastatic 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.