MicroRNA expression profiling to identify and validate reference genes for relative quantification in colorectal cancer

Mi(cro)RNAs are short RNA molecules that bind (generally) to 3' UTR sequences of target messenger RNAs (mRNAs), thereby modulating their expression patterns. This modulated gene expression is manifest either as translational repression [1], or mRNA degradation whereby the RNA interference pathway is initiated to remove targeted sequences [2]. MiRNAs play major roles in governing diverse biological processes such as differentiation, proliferation, and apoptosis [3, 4]. Individual miRNAs have been ascribed oncogenic and tumour suppressor functions [5], and aberrant miRNA expression has been implicated in many malignancies, including colorectal cancer (CRC) [6, 7]. Previous study demonstrated that miRNA profiles may be more accurate in disease classification than mRNA profiles [8]. Moreover, miRNAs have been associated with CRC pathogenesis [9, 10], microsatellite stability status [11, 12], therapeutic outcome and prognosis [12‐15].

High-throughput technology such as microarrays enables simultaneous quantification of hundreds of miRNAs in a single RNA sample. Meaningful interpretation of such large datasets has been made possible by recent advances in bioinformatics. It is critical that the findings of microarray screening methodologies are validated to produce scientifically robust results, using the most sensitive and reproducible method of gene expression quantitation, reverse transcription quantitative PCR (RT-qPCR) [16]. In order to achieve accurate, reproducible and biologically relevant miRNA RT-qPCR data, non-biological sample-to-sample variation that could be introduced by protocol-dependent inconsistencies has to be corrected for by using reference genes. Use of unreliable reference genes for normalisation may lead to inaccurate quantitation of miRNAs of interest [17, 18]. Previous studies have demonstrated that a single universal reference gene for all tissue types is unlikely to exist [19‐23], and the use of a single reference gene for normalisation leads to large errors and is therefore inappropriate [22, 24].

Despite increasing miRNA expression studies in CRC, no previous report detailed a robust identification and validation strategy for suitable reference genes for normalization. The aim of this study was to identify the most stable reference genes using a high-throughput approach, in ten pairs of stage II colorectal tumour and normal tissues. Following TaqMan array card analysis and the established approach of finding miRNAs whose expression pattern is similar to the global mean expression [25], miR-26a, miR-345, miR-425 and miR-454 were identified as the most stably expressed miRNAs. The stability of these miRNAs was further assessed by RT-qPCR in 74 colorectal tissues with an expanded panel of candidate reference miRNAs (let-7a, miR-16) and two small nucleolar RNAs (snoRNAs, RNU48 and Z30). Well established oncogenic miRNAs in CRC: miR-21 [7, 13, 26] and miR-31 [7], and tumour suppressor miRNAs: miR-143 [6, 27, 28] and miR-145 [6, 7, 12, 27] were used as target miRNAs to determine the effect of reference gene choice on relative quantitation.

The discovery of miRNAs as crucial regulators of gene expression has resulted in the rapid expansion of understanding of gene regulation in normal development and disease. Previously, it was demonstrated that miRNA expression profiles may be more accurate in disease classification than mRNA expression profiles [8]. However, accurate and reliable interpretation of RT-qPCR results depends heavily on the use of suitable reference genes for normalisation to eliminate or minimise non-biological variation between test samples. While reference genes for mRNA RT-qPCR studies have been well-established, few miRNA RT-qPCR studies have detailed the validation of reference genes for normalisation to date. Rigorous normalisation of miRNA data may be more critical than that of other RNA functional classes [18]. Indeed, their capability to regulate multiple gene targets within the same pathway may amplify their biological effects [33], hence small changes in miRNA expression may be biologically and clinically significant.

Davoren et al. reported the first systematic assessments of candidate reference genes for miRNA RT-qPCR analysis in breast cancer [17]. To our knowledge, such assessment and validation of reference genes for CRC studies has not been reported. The two most commonly used normalisers U6 and 5S RNAs were shown to be the two least stable RNA species [18]. The use of rRNAs as reference genes has been debated as they can be expressed at much greater levels than target RNAs resulting in difficulty quantitating a lowly expressed target RNA [20, 22]. Furthermore, rRNAs have been shown to be involved in apoptosis [34] and cancer [35]. Lastly, it has been argued before that it's best to normalise genes with reference genes belonging to the same RNA class [22]. Let-7a was used as a normaliser in CRC miRNA RT-qPCR studies [7, 10]. However, its tumour-suppressor role in CRC has been reported [27]. In a previous study, miR-191 and miR-25 were identified as the most stable pair of normalisers across 13 distinct human tissue types including 5 pairs of colon tumour and adjacent normal tissues. However, when analysis was performed on an extended cohort of lung cancer and normal tissues, miR-17-5p and miR-24 were the best normalisers [18]. This demonstrates the importance of validating suitable reference genes in a tissue-specific context. Suitable reference genes for colorectal tissue-specific studies needs to be further assessed as previous reports have demonstrated that a single universal reference gene for all tissue types is unlikely to exist [19‐23].

This is the first report detailing identification and validation of suitable reference genes for normalisation of miRNA RT-qPCR in human colorectal tissues. We profiled the expression of 380 miRNAs (including U6 rRNA) on 20 colorectal tissues. A robust method using the mean expression value was used to identify the most stably expressed miRNAs: let-7a, miR-26a, miR-345, miR-425 and miR-454. Mean normalisation was previously shown to outperform other methods of normalisation in terms of better reduction of technical variation and more accurate appreciation of biological changes [25]. Validation by RT-qPCR was subsequently carried out in a larger cohort of 74 tissues with assessment of three more candidate reference genes (miR-16, RNU48 and Z30) [17]. Our initial validation step confirmed no difference in reference gene quantities between tumour and normal tissues, allowing subsequent use of NormFinder and geNorm as these models assume that reference genes are not differentially expressed between experimental groups. Equivalent expression of reference genes between tumour and normal tissues was then confirmed using a fold change cut-off of 3 [23]. Both NormFinder and geNorm identified miR-16 and miR-345 as the most stable normalisers. The five most stably expressed miRNAs in the TaqMan array card dataset of stage II tumours remained stably expressed when a larger cohort of variable disease stages was evaluated. This suggests that true reference genes are non-functional in the disease process, and should remain stably expressed throughout all stages, grades and subtypes.

As evident from our results, inappropriate use of reference genes can significantly alter the results of target miRNAs quantitation. With the use of the best combination of reference genes (miR-16 and miR-345), significant dysregulation of all four target miRNAs (miR-21, miR-31, miR-143 and miR-145) was detected. These target miRNAs have repeatedly been shown to be dysregulated in CRC in previous studies. However, despite a relatively large sample size, when inappropriate reference genes were used for normalisation, a true biological difference in expression between tumour and normal was not detected. Even though miR-345 and miR-454 detected significant difference between tumour and normal tissues when used alone as a reference gene, geNorm analysis identified them as only the third and the fifth most stably expressed genes. The p values of the differential expression of the four target miRNAs between tumour and normal tissues were slightly lower when using the miR-16/miR-345 combination in most instances, which could prove significant in a small scale study. Furthermore, previous studies have reported that the use of more than one reference genes increases the accuracy of quantitation compared to the use of a single reference gene [22, 32].

The results of our study have important implications for CRC translational research. The clinical and pathologically diverse nature of the tissues used in this study should be of interest to a broad spectrum of the CRC research community. While it may not be feasible due to cost and sample availability, the stability of the top six most stably expressed miRNAs in colorectal tissues (let-7a, miR-16, miR-26a, miR-345, miR-425 and miR454) should be assessed to determine the most appropriate normalisers within each study as patient and tumour characteristics may vary between different study cohorts. Furthermore, with evidence to suggest that miRNA expression in formalin-fixed paraffin-embedded (FFPE) tissue samples remains relatively stable and consistent with that in fresh-frozen samples [36], and that reference miRNA stabilities are extremely consistent between the two tissue sources procured and processed independently of one another [18], the reference genes identified in this study may be useful for miRNA RT-qPCR study in FFPE colorectal tissues. This study also demonstrated that the use of the mean expression value is a useful means of identifying stable reference genes in high-throughput miRNA profiling studies, and the findings were confirmed to be robust after external validation.