Oncogenic Role of miR-15a-3p in 13q Amplicon-Driven Colorectal Adenoma-to-Carcinoma Progression

Progression from colorectal adenoma to carcinoma is strongly associated with an accumulation of genomic alterations, including gain of chromosome 13. This gain affects the whole q arm and is present in 40%–60% of all colorectal cancers (CRCs). Several genes located at this amplicon are known to be overexpressed in carcinomas due to copy number dosage. A subset of these genes, including the mir-17~92 cluster, are functionally involved in CRC development. The present study set out to explore whether apart from mir-17~92, other miRNAs located at the 13q amplicon show a copy number dependent dosage effect that may contribute to 13q-driven colorectal adenoma-to-carcinoma progression. Integration of publically available miRNA expression, target mRNA expression and DNA copy number data from 125 CRCs yielded three miRNAs, miR-15a, -17, and -20a, of which high expression levels were significantly correlated with a 13q gain and which influenced target mRNA expression. These results could be confirmed by qRT-PCR in a series of 100 colon adenomas and carcinomas.Functional analysis of both mature miRNAs encoded by mir-15a, i.e. miR-15a-5p and miR-15a-3p, showed that silencing of miR-15a-3p significantly inhibited viability of CRC cells. Integration of miR-15a expression levels with mRNA expression data of predicted target genes identified mitochondrial uncoupling protein 2 (UCP2) and COP9 signalosome subunit 2 (COPS2) as candidates with significantly decreased expression in CRCs with 13q gain. Upon silencing of miR-15a-3p, mRNA expression of both genes increased in CRC cells, supporting miR-15a-3p mediated regulation of UPC2 and COPS2 expression. In conclusion, significant overexpression of miR-15a-3p due to gain of 13q is functionally relevant in CRC, with UCP2 and COPS2 as candidate target genes. Taken together our findings suggest that miR-15a-3p may contribute to adenoma-to-carcinoma progression.


Introduction
The development of colorectal cancer (CRC) is marked by the accumulation of several recurrent chromosomal alterations, including gains of 8q, 13q, and 20q and losses of 8p, 15q, 17p and 18q [1][2][3], which can lead to altered expression of oncogenes and tumour suppressor genes [4,5]. In fact, for a number of genes identified to be differentially expressed in a copy number dependent manner, functional relevance has been demonstrated. For instance, DIS3, LNX2 and CDK8 at 13q were recently described to play a role in CRC development due to 13q gaindependent overexpression [6][7][8]. The same holds true for AURKA and TPX2 located at 20q, which were found to promote 20q amplicon-driven colorectal adenoma to carcinoma progression [9].
In addition to protein-encoding genes, DNA copy number changes may also affect expression of microRNAs (miRNAs) [10,11]. MiRNAs are a family of small non-coding RNA molecules that play an important role in the regulation of many cellular processes by targeting the 3' UTR of mRNA molecules, thereby leading to gene silencing [12]. Dysregulation of miRNA expression has been shown to play an important role in several human diseases, including cancer [13].
In CRC, altered expression of a number of miRNAs, due to chromosomal alterations or other mechanisms like epigenetic modifications, has been described. A well-documented example is the altered expression of the oncogenic miRNA cluster, mir-17~92 [14]. We previously showed that increased mir-17~92 expression was linked to copy number gain of 13q and increased c-MYC expression during colorectal adenoma to carcinoma progression [15]. Moreover, upregulation of mir-17~92 by c-MYC showed pro-angiogenic activity in colonocytes [16].
Chromosome 13 is gained in 40-60% of CRCs and is strongly associated with colorectal adenoma to carcinoma progression [1,2,5]. This gain mostly encompasses the entire q-arm of chromosome 13 [6,17], raising the question whether next to coding genes and the mir-17~92 cluster, expression of other miRNAs located at this region is also affected by copy number dosage and may contribute to colorectal adenoma to carcinoma progression.
The present study aimed to identify additional candidate oncomiRs located at 13q in CRC. To this end, as an unbiased approach, we analysed all miRNAs mapping at 13q. Following validation of DNA copy number-dependent overexpression of 13q miRNAs by integrative analysis of DNA copy number and expression data-sets, the functional role of selected miRNAs was investigated in CRC cell lines using loss-of-function assays. Candidate target genes were identified by integration of miRNA expression levels with mRNA expression levels.

Integration of expression and copy number data
To investigate 13q miRNAs of which expression was regulated by DNA copy number changes, associations between DNA copy number and miRNA expression levels were studied. To this end we used an integrated approach, based on the use of covariate sets rather than single covariates, to determine subtle consistent associations with copy number alterations of a chromosomal region. The covariate set was defined by considering each miRNA probe individually and all copy number measurements within a 2Mb-window around the genomic start position of the miRNA. The analysis pipeline and definition of the covariate set is graphically presented in Fig 1 (association 1). The method used is implemented in the BioConductor package SIM [18]. This procedure finds miRNAs of which expression is likely to be regulated by DNA copy number changes within this 2MB-window, by assigning one p-value to each miRNA. Multiple testing correction was done using the false-discovery rate (FDR) step-down procedure of Benjamini & Hoghberg [19]. This analysis yielded a list of FDR values for miRNAs, of which expression is likely to be regulated by copy number changes in these samples. MiRNAs with FDR<0.05 were considered significant.

Target mRNA and miRNA expression integration
To identify which miRNAs are more likely to be functional, we studied their association with expression of predicted target mRNAs in the 125 CRC TCGA samples. To this end we used the approach developed by van Iterson et al [20]. In short, for each miRNA, results of 4 target-prediction tools, namely TargetScan, PITA, microCosm and Mirtarbase [21][22][23][24], were combined to yield a list of possible mRNA targets. For further (i.e. second) analysis, candidate mRNA targets that were predicted by three or more prediction tools were selected. This list of target mRNAs defined the covariate set for this second analysis (Fig 1, association 2). Subsequently, a global test was used [25] to test for associations between miRNA and mRNA expression, yielding one p-value per miRNA. Here the same statistical test was used as for the miRNA-copy number analysis, but now using the mRNA target list as covariate set. These p-values were corrected for multiple testing using the Benjamini & Hochberg FDR [19]. FDR values <0.05 were considered significant. For each significant miRNA, we subsequently prioritised the mRNA targets using the p-value derived from an additional global test applied to the miRNA and each individual mRNA target. We selected for further analyses mRNA targets that had an individual p-value of <0.01.
In order to analyse expression patterns of several selected target mRNAs, previously generated expression data were used of a panel of colorectal adenomas (n = 37) and carcinomas (n = 31), available at GEO under accession number GSE8067 [5]. Expression levels in both groups were compared using the Mann-Whitney U-test, considering p-values <0.05 as significant.

Tumour samples
A series of 100 snap-frozen colorectal tumour samples (48 adenomas and 52 carcinomas) prospectively collected at the VU University Medical Center in Amsterdam was used in this study. RNA from these samples was isolated using TRIzol (Invitrogen, Life technologies, Bleiswijk, The Netherlands), following the supplier's instructions. For a subset of these samples (12 adenomas and 15 carcinomas) whole genome DNA copy number status was known (5). For the remaining samples copy number status of chromosome 13q was determined by multiplex ligation-dependent probe amplification (MLPA) as described previously [26].
In cell lines miRNA expression levels of miR-15a-3p, miR-15a-5p and miR-17 were analysed using the miRCURY LNA Universal RT microRNA PCR (Exiqon, Vedbaek, Denmark) following the manufacturer's instructions. PCRs were run on an ABI 7500 Fast Real-time PCR system (Life technologies). Relative miRNA expression levels were determined using the 2(-ΔΔCt)method [27]. Data were normalised using the small nucleolar RNA transcript RNU43 as a reference gene (Exiqon).

Cell viability
Cell viability was measured using the 3-(4,5-dimethylthiazolyl-2)-2,5-diphenyl tetrazolium bromide assay (MTT; MP Biomedicals, Eindhoven, The Netherlands). Twenty-four hours post-transfection with the antagomirs, cells were seeded in 96-wells culture plates for the MTT assay. Simultaneously, RNA was extracted from a subset of the cells to determine the miRNA expression levels. MTT measurements were performed at t = 0 and subsequently after 3 and 6 days. The relative viability was determined by subtracting the measurement at day 0 from subsequent time points. All measurements were performed in triplicate (technical replicates) and at least 2 independent experiments were performed (biological replicates). Relative viability of cells was compared using Student's t-test, with p-values <0.05 considered significant.

Results
MiR-15a shows copy number-associated differential expression in CRC In the present study, publically available TCGA data from 125 CRC samples were used to analyse both miRNA expression levels and DNA copy number levels of the loci involved on chromosome 13 [17]. According to miRBase release 21 (www.mirbase.org), there are 40 miRNA located at chromosome 13 [29]. For 14 out of these 40 miRNAs on 13q, expression data were available, and 6 of these were significantly overexpressed in CRCs due to copy number gain of 13q, i.e. miR-15a, -17, -19a, -20a, 19b-1 and -92a-1 (Table 1; FDR<0.05). These miRNAs all belong to two well-known clusters of miRNAs on chromosome 13, i.e. mir-15a/16 and mir-17~92.
Next to miRNA expression data, mRNA expression data were also available for the same panel of 125 CRCs. Using these data we examined whether overexpression of these miRNAs on 13q resulted in changes in the expression of predicted target genes [20]. This analysis showed that, for miR-15a, -17 and -20a, overexpression due to copy number also significantly influenced predicted target mRNA expression (Table 1; FDR<0.05).
Next we analysed by qRT-PCR the expression levels of the top candidate miRNAs found in the in silico analyses in a series of 100 colon adenoma and carcinoma samples. Given the fact that mir-15a encodes for two mature miRNA strands, miR-15a-5p and miR-15a-3p, both were studied using strand-specific qRT-PCR. In line with the in silico data, expression levels of miR-15a-3p, miR-17 and miR-20a, were significantly higher in carcinomas compared to adenomas (Fig 2A). However, expression of miR-15a-5p was not significantly increased. Comparison of tumours with 13q gain with tumours without 13q gain showed that expression levels of all four miRNAs were significantly higher in tumours with gain (Fig 2B), corroborating the results from the in silico analysis.

MiR-15a-3p inhibits viability of CRC cells
Based on these analyses, miR-15a, -17 and -20a were identified as candidate oncogenic miR-NAs involved in CRC. Of these, both miR-17 and -20a have recently been described as oncogenes in CRC and prostate cancer [30][31][32], confirming the validity of the approach. MiR-15a  on the other hand has been described to function as a tumour suppressor in chronic lymphocytic leukaemia (CLL) [33,34] and, more recently, in CRC [35]. To further support our current observations pointing to an oncogenic role of miR-15a in CRC, we set out to determine the functional relevance of miR-15a overexpression in CRC. Both strands of miR-15a, miR-15a-5p and miR-15a-3p, were studied using strand-specific miRNA inhibitors (antagomirs). An antagomir directed against miR-17, silencing of which was shown to decrease viability of CRC cells, was included as a positive control [30]. A non-targeting antagomir was included as a negative control. The antagomirs were introduced into the SW480 CRC cell line, which has a gain of 13q encompassing the miR-15a locus. Efficiency of silencing varied per experiment between 75%-97%, 60-80% and 94%-96% for miR-15a-5p, miR-15a-3p and miR-17, respectively (Fig 3A). Silencing of miR-15a-3p and miR-17 led to a significant decrease in cell viability (Fig 3B). Silencing of miR-15a-5p also decreased cell viability significantly, although results were less consistent throughout replicate experiments (data not shown).
UCP2 and COPS2 as potential targets of miR-15a-3p in CRC By integrating miRNA expression profiles with mRNA expression profiles of predicted targets in 125 CRCs, we identified candidate target genes of miR-15a [20]. The top five are shown in Table 2 (p <0.01). Of these five candidate target genes, four genes (TYRO3, UCP2, COPS2, RASGEF1B) were negatively associated with miR-15a expression, whereas (RNF125) was positively associated.
Since copy number gain of chromosome 13 has been associated with progression from colorectal adenoma-to-carcinoma, and current data show that this gain is associated with overexpression of miR-15a, we next evaluated whether the expression levels of the top five predicted target genes would show a decrease in carcinomas compared to adenomas, in conjunction with 13q gain. To this end, previously obtained array-CGH and mRNA expression microarray data of a panel of 37 colon adenomas and 31 carcinomas were used [5]. In this set, expression of UCP2 was significantly decreased in carcinomas compared to adenomas (Fig 4A). Both UCP2 and COPS2 were significantly decreased in adenomas and carcinomas with a 13q gain compared to those without (Fig 4B). The comparison of adenomas without 13q gain to carcinomas with 13q gain, showed for both UCP2 and COPS2 significantly decreased expression in carcinomas with 13q gain (Fig 4C). In the case of TYRO3, a significant increase in expression was observed in both carcinomas (compared to adenomas), and in samples with 13q gain (compared to tumours without 13q gain). RNF125 and RASGEF1B expression levels were not related to progression or 13q copy number gain.
To further establish a role of miR-15a overexpression in UCP2 and COPS2 downregulation in CRC, mRNA expression levels of both genes were determined in SW480 cells, in which either miR15a-3p, miR15a-5p or control miR-17 were silenced. Compared to the non-targeting control, mRNA expression levels of UCP2 and, to a lesser extent, COPS2 were increased upon silencing of miR-15a-3p (Fig 5). Hardly any effect was seen when using antagomirs against miR-15a-5p or miR-17 (Fig 5). These results indicate that miR-15a-3p indeed targets both UCP2 and COPS2 in SW480 CRC cells.

Discussion
Here we investigated miRNAs that are overexpressed in CRC due to copy number gain of chromosome 13q. Analysis of publicly available TCGA miRNA expression and DNA copy number data of 125 CRC samples [17] revealed that miR-15a, -17 and -20a were significantly upregulated in CRC in association with copy number gain of 13q, and that miR-15a, -17 and -20a significantly influenced expression of target genes. For both miR-20a and -17, an oncogenic role in CRC has been established [30,36]. Interestingly, for miR-15a actually a tumour suppressor role has been claimed [35]. The present results however indicate that miR-15a, in particular miR-15a-3p, functions as an oncogenic miRNA in CRC. Silencing of miR-15a-3p significantly decreased viability of CRC cells.

MiR-15a-3p in CRC Progression
MiR-15a is part of the mir-15a/16-1 cluster located at chromosome 13q14, which is encoded by its host gene deleted in leukaemia 2 (DLEU2) [34,37]. To date, members of the mir-15a/16-1 cluster have been described as tumour suppressor miRNAs in several types of cancer [34,35,[38][39][40]. In chronic lymphocytic leukaemia, inactivation of these miRNAs occurs through a chromosomal loss at the locus of the mir-15a/16-1 cluster [34]. In multiple myeloma, expression of mir-15a/16-1 was decreased whereas expression of its target gene VEGF-A was increased. Moreover, in these tumours expression of mir-15a/16-1 was shown to inhibit tumour formation by targeting pro-angiogenic factor VEGF-A [39]. Similarly, in prostate cancer the expression of the miRNA cluster was downregulated, which was associated with upregulation of target genes BCL2, CCND1 and WNT3A. Knockdown of miR-15a and miR-16 in an in vitro model enhanced cancer progression by affecting cell survival, proliferation and invasion [38].
In CRC, miR-16 of this cluster had already been described to have a function as tumour suppressor [41,42]. More recently however, the entire mir-15a/16-1 cluster was investigated and found to target AP4 in CRC, downregulation of which resulted in induction of mesenchymalepithelial transition (MET), inhibition of migration and invasion, and induction of cell cycle arrest [35]. Additionally, expression of DLEU2, the gene encoding the mir-15a/16-1 cluster, was repressed in CRC samples compared to normal colon tissue and inversely correlated to AP4 expression. Altogether, those data suggest that miR-15a expression is decreased in CRC compared to normal tissue, which seems contradictory to the findings in the current study. In fact, the TCGA data presented here, also show that in CRC samples the expression of miR-15a actually increases in association with a DNA copy number gain of chromosome 13q. This gain is a common event in CRC, and in particular, in the progression from colorectal adenoma to carcinoma [2,6,17].
With respect to the role of the mir-15a/16-1 cluster in CRC, it is important to realize that the present study specifically focused on the association with a 13q gain involved in adenoma to carcinoma progression, whereas the above mentioned study of Shi et al. focused on CRC metastasis, representing a later event in the carcinogenic process [35]. More importantly, we specifically identified miR-15a-3p as the driver strand and did not find evidence for the whole cluster to be involved. Recent research also showed that miR-15a-3p plays a tumour suppressor role in several cancer cell lines by targeting BCL-X L . However, no significant effect was seen in CRC cell lines [40]. It is known that both the 5p and 3p strands of a miRNA can be expressed in cells and regulate target gene expression [43,44]. The expression of either strand can also be tissue-specific [45]. We therefore speculate that expression of different mature miRNAs from the mir-15a/16-1 cluster may have different roles in CRC, perhaps depending on the stage of CRC.
Evaluation of miR-15a-3p expression in a series of 48 adenomas and 52 carcinomas, for which 13q copy number status was known, showed that miR-15a-3p was significantly higher expressed in carcinomas compared to adenomas in a DNA copy number-dependent manner. These results corroborate an oncogenic role for increased miR-15a-3p expression on the 13q amplicon.
By combining the results of target prediction algorithms with mRNA expression profile data of CRCs, UCP2 and COPS2 were identified as candidate target genes of miR-15a. In support of this observation, upregulation of UCP2 and, to a lesser extent, of COPS2 was found in CRC cells when miR-15a-3p was silenced, compared to control transfectants. The absence of a similar effect in miR-15a-5p silenced CRC cells is consistent with our functional data showing that both strands seem to act independently. The present data indicate that UCP2, and potentially also COPS2, may function as tumour suppressors in CRC. Interestingly, overexpression of UCP2 in human pancreatic cancer and glioblastoma cell lines was recently described to repress malignancy by controlling mitochondrial function and redirecting energy production away from glycolysis [46]. Also COPS2 was described as a putative tumour suppressor gene in a panel of different cancer types, where loss-of-function showed partial bypass of cellular senescence [47]. Altogether, one could hypothesize that tumour cells would benefit from the inhibition of these genes through miRNA-mediated modulation.
According to the most recent release of the miRNA database (miRBase release 21), there are 40 miRNAs located at chromosome 13q. However, for this study we only had expression data of 14 13q-mapping miRNAs available, as expression levels of the remaining miRNAs were undetectable or not measured. Therefore, it cannot be excluded that additional miRNAs located at 13q may also be involved in CRC.
In summary, miR-15a-3p is gained and overexpressed in CRC, and this overexpression affects cell viability, supporting an oncogenic role of this miRNA in CRC. As a consequence, miR-15a-3p may play a role in the 13q gain-driven colorectal adenoma-to-carcinoma progression.
Supporting Information S1 Table. Overview of all the TCGA cases used in this study. Additional information on the data of 125 cases used from the TCGA. For all cases the sample ID, type of data (platform type), platform used, data level and the corresponding file names are listed. More information on the data used can be found at the TCGA Data Portal (https://tcga-data.nci.nih.gov/tcga/ tcgaDataType.jsp). (DOC)