miRNA expression in colorectal tumours and adjacent non-tumor tissues

In order to investigate whether miRNAs are differentially expressed in CRC versus normal colon tissues, we analyzed miRNA expression in 12 matched-pairs of tumoral and non-tumoral tissues. After testing three different approaches to normalize the Ct raw data in CRC cell lines, median-normalization was selected as method for clinical samples since normal distribution was not required. Meanwhile in our study in CRC cell lines no differences were found in let-7a expression between tumoral and normal cell line, recent evidences identify let7-family as differentially expressed in CRC [9] and lung cancer [5]. Moreover, global median normalization could provide results more easily comparable with those already published with microarray technology.

To identify miRNA with significantly differential expression among CRC samples, two multivariate permutation test provided in BRB-ArrayTools were performed: Class Comparison between Groups of Arrays and SAM (Significance Analysis of Microarrays). In both cases we selected paired t-test options and a FDR (False Discovery Rate) less to 10%. Fifty-nine miRNAs were significant when Class Comparison test was applied, 68 miRNA were significant using SAM test and 53 miRNA are common in both test. As expected, fold-change observed in clinical samples is less homogeneously distributed among samples that already obtained in CRC cell lines. It is not surprising regarding that patients samples are composed of mixed populations, whereas cell lines are clearly more uniform.

Interestingly, our results in CRC samples are in agreement with recent data published in CRC using direct miRNA cloning and SAGE (miRAGE). In this context, we detected an over-expression of miR-19a, miR-21, miR-29a, miR-92, miR-148a, miR-200b, and a down-regulation of miR-30c, miR-133a and miR-145 (figure ​(figure2).2). Moreover, change of expression of some of these miRNA has been previously reported in lung and breast cancer, B-cell lymphomas, and glioblastoma. miR-19a and miR-20 are including in the cluster miR-17-92 and it is located at intron 3 of C13orf25. The transfection of C13orf25 in lung cancer cell line enhancing cell growth and the introduction of miR-17-92 into hematopoietic stem cells in Eu-myc transgenic mice accelerates the formation of lymphoid malignancies. Furthermore, miR-21 has been described as an antiapoptotic factor in human glioblastoma cell lines. In contrast, other authors report that miR-21 suppression increase growth in HeLa cells without affecting their apoptosis. The different biologic effects of any particular miRNA in different cells could be dependent of the cell-specific repertoire in target genes. Some of miRNA differentially expressed in CRC samples have been associated with clinical parameters in other cancers. In particular, miR-145 is progressively down-regulated from normal breast to cancer with high proliferation index and miR-21 is progressively up-regulated with high grade tumor stage.

Open in a separate window

Figure 2

miRNA expression data from 12 CRC tumor samples. Each miRNA listed was detected as significantly differentially expressed between tumoral and paired-non-tumoral tissues with SAM and Class Comparison tests. Samples are in columns and miRNAs in rows. The expression values ranged from + 1.5 log₁₀ to - 1.5 log₁₀.

To identify the smallest set of predictive miRNAs differentiating normal versus cancer tissues, we have used support vector machines (SVMs) techniques. We attempted to use the class prediction tool (BRB-Array tools) that creates a multivariate predictor for determining to which of the two classes a given sample belongs. Several multivariate classification methods are available, including the Compound Covariate Predictor, Diagonal Linear Discriminate Analysis, Nearest Neighbor Predictor, Nearest Centroid Predictor, and Support Vector Machine Predictor. The classifier is composed for 18 miRNA, 10 down-regulated and 8 up-regulated, all of them significantly different by Class Comparison and SAM tests.

When we compared expression of these miRNA in CRC cell lines, 5 of 18 miRNA were revealed in the k-means analysis as those of highest fold-changes (in relation to CDC18Co). However, Class Comparison analysis between 15 CRC cell lines and the 12 non-tumoral colon tissues identifies 13 miRNA altered in both systems, CRC patients samples and CRC cell lines (Table ​(Table2).2). These results could indicate that miRNA profile in CRC cell lines can not be used to infer miRNA expression in clinical samples meanwhile cell lines can be used as model to validate and performed functional assays of data obtained in clinical samples. In this sense, hierarchical clustering of expression of these 13 miRNAs in CRC patients samples and CRC cell lines clearly separate samples in two groups: in one branch, the most different of non-tumoral samples were included all the CRC cell lines, and in the other branch tumoral samples of patients (Figure ​(Figure33).

Open in a separate window

Figure 3

Hierachical clustering of CRC cell lines and tumor samples by using expression of 13 miRNAs that have found differentially expressed between neoplastic conditions (CRC cell lines and tumor samples) and non-tumoral colon tissues. Samples are in columns and miRNAs in rows.

The expression of 5 of 13 miRNA is already described altered in CRC, lung and breast cancer, glioblastoma, B-cell lymphoma and CLL. Among the differentially expressed miRNAs, miR-31, miR-96, miR-133b, miR-135b, miR-145 and miR-183 as the most consistently deregulated in CRC. Two of them, miR-133b and miR-145 were down-regulated and the remaining four, miR-31, miR-96, miR-135b and miR-183, were up-regulated, suggesting that they may potentially act as tumor suppressor genes or oncogenes, respectively.

miR-145 was identified as a specific miRNA down-regulated in colorectal neoplasia and analysis of their pre-miRNA indicate that this reduction is due to posttranscriptional process [32]. Recently, Cummins et al obtained similar results in CRC [9] and down-regulation of miR-145 have also reported in lung [8] and breast cancer [13]. In our study, expression of miR-145 was not detected in any of 15 CRC cell lines tested and down-regulation was detected in all tumor samples. Other important down-regulated miRNA in our study was miR-133b. In our knowledge, this miRNA has not previously identified deregulated in cancer. For both down-regulated miRNAs (miR-145 and miR-133b), it may be expected that potential targets could include oncogenes or genes encoding proteins with potential oncogenic functions. Indeed, among putative targets for miR-145 with potential oncogenic functions, Iorio et al [13] described MYCN, FOS, YES, and FLI, cell cycle promoters such as cyclins D2 and L1; and MAPK transduction proteins such as MAP3K3 and MAPK4K4. Among putative targets of miR-133b, the most notable oncogenic target is KRAS. Interestingly, the proto-oncogen YES1 and the transduction protein MAP3K3 were potential targets of both miR-145 and miR-133b.

For the up-regulated miRNAs, miR-135b, miR-31, miR-96 and miR-183, it may be expected that gene targets belong to the class of tumor suppressor genes. miR-96, miR-182 and miR-183 are located in the same chromosomal region, 7q32.2. miR-182 was not detected as preferentially over-expressed with the most restricted analysis, but their up-regulation was clearly observed in CRC cell lines analysis (table ​(table1).1). CHES1 protein was identified as a potential target of both miR-96 and miR-182. CHES 1 is a member of the forkhead family of transcription factors that repress genes involved in apoptosis. Other members of this family, including FOXF2, FOXK2, FOXO1A, FOXO3A and FOXQ1, were also found as putative targets of miR-182, miR-183 and miR-96.

Finally, our analysis of a small number of CRC samples compared miRNA expression in tumors according to pathological stage (stage II versus stage IV). The up-regulation of miR-31 was significantly higher in stage IV than in CRC samples stage II (p = 0.028) (Figure ​(Figure4).4). The expression levels of miR-31 were higher in the tumor samples and CRC cell lines in comparison to the non-tumoral samples and was related to pathological stage, suggesting that this miRNA could contribute to both, the tumorogenesis and the acquisition of a more aggressive phenotype in CRC. Other members of the forkhead family transcription factors, such as FOXC2 and FOXP3, were identified as putative targets of miR-31. Future studies will determine the correlation between of these miRNAs and their host genes in CRC.

Open in a separate window

Figure 4

Real-time PCR analysis of miR-31 expression between stage II and stage IV tumor samples. Differences was significant after Mann Whitney U test (p = 0.028).

In summary, our results by Real-time PCR identify alterations of miRNA expression in CRC that may deregulate cancer-related genes and would provide potential mechanisms that underly in the carcinogenesis and further acquisition of a more aggressive phenotype in colon cancer.

Normalization and data analysis

Careful normalization is essential for the accurate quantification of mRNA levels and commonly, normalization of the target gene with an endogenous standard, mainly housekeeping genes, is applied. However for miRNA, there is not data about the expression of miRNA as normalization control.

In our study, we have tried different approach for normalization data. First, miRNA expression data was normalized to let7-a miRNA (according to the manufacturer's suggestions). Relative quantification of miRNA expression was calculated with the 2^-ΔΔCt method (Applied Biosystems User Bulletin N°2 (P/N 4303859)). The data were presented as log₁₀ of relative quantity (RQ) of target miRNA, normalized respect to miR-let-7a and relative to a calibrator sample. As calibrator, we are used for colorectal cancer cell lines CDC18Co (human normal colon) and for CRC samples the paired-normal tissues.

The TaqMan Human Endogenous Control Plate (Applied Biosystems) precoated with lyophilized primers and probes for 11 differently commonly human control genes was also used to asses gene expression in two tumours and two normal colon tissues. PCR was set up according to the manufacturer's instructions. Among 11 genes analyzed, the variability in expression of 18s rRNA was shown to be the lowest (data not shown). Therefore, our second normalization approach showed the data as log₁₀ of relative quantity (RQ) of target miRNA, normalized to 18s rRNA and relative to control sample.

Finally, similar to microarray data, raw data Ct was normalized and analyzed using BRB ArrayTools version 3.3.2. (Richard Simon and Amy Peng Lam, National Cancer Institute, Bethesda). After global median normalization, normalized data were presented as log₁₀ of relative quantity (RQ) of target miRNA relative to a control sample. Class Comparison and Significant analysis of microarrays (SAM) was performed to identify differentially expressed miRNAs between tumors and normal samples.

Visualization of results was performed with the different normalized data using average linkage and Euclidean distance as a measurement of similarity using GENESIS Software (Alexander Sturn, Institute for Genomics and Bioinformatics, Graz University of Technology).