MicroRNA profiling of clear cell renal cell cancer identifies a robust signature to define renal malignancy

MicroRNAs are short single-stranded RNAs that are associated with gene regulation at the transcriptional and translational level. Changes in their expression were found in a variety of human cancers. Only few data are available on microRNAs in clear cell renal cell carcinoma (ccRCC). We performed genome-wide expression profiling of microRNAs using microarray analysis and quantification of specific microRNAs by TaqMan real-time RT-PCR. Matched malignant and non-malignant tissue samples from two independent sets of 12 and 72 ccRCC were profiled. The microarray-based experiments identified 13 over-expressed and 20 down-regulated microRNAs in malignant samples. Expression in ccRCC tissue samples compared with matched non-malignant samples measured by RT-PCR was increased on average by 2.7- to 23-fold for the hsa-miR-16, −452*, −224, −155 and −210, but decreased by 4.8- to 138-fold for hsa-miR-200b, −363, −429, −200c, −514 and −141. No significant associations between these differentially expressed microRNAs and the clinico-pathological factors tumour stage, tumour grade and survival rate were found. Nevertheless, malignant and non-malignant tissue could clearly be differentiated by their microRNA profile. A combination of miR-141 and miR-155 resulted in a 97% overall correct classification of samples. The presented differential microRNA pattern provides a solid basis for further validation, including functional studies.

stage and grade, prognosis of disease and progression under therapy or survival of patients was demonstrated for various cancer types [11,12]. These studies constitute the promising potential of microRNAs as tumour classifiers and as prognostic and predictive biomarkers [5,13,14]. Recently, we summarized the current state of knowledge about microRNAs in urological tumours [15]. However, although renal cell carcinoma (RCC) is 1 of the 10 most frequent tumours [16], the published data on microRNA in human RCC are surprisingly sparse [17,18]. There were a few microarray-based studies including expression data from kidney only in comparison with other organs but not to RCC [10,[19][20][21]. During the revision of this paper, a report on microRNAs on RCC was published [22]. On the other hand, there are reports on the role of microRNAs in nontumoural kidney pathology. For example, 20 differently expressed microRNAs were recently identified in acute rejection after renal transplantation [23].
This lack of data prompted us to (i ) determine and compare the expression profiles of microRNA in RCC in comparison with normal renal tissue, (ii ) explore possible associations between microRNA expression patterns and clinico-pathological data, (iii ) test the potential clinical usefulness of microRNAs as diagnostic and prognostic markers and (iv ) provide a reliable differential microRNA profile as a solid basis for further functional studies. This study focussed on clear cell RCC (ccRCC), the most common (75% of cases) renal cancer sub-type with the most serious prognosis [24].

Patients and tissue samples
The study was successively performed on two independent sample sets. Set one included 12

RNA extraction
Matched malignant and non-malignant specimens from the same kidney were collected in RNAlater Stabilization Reagent (Qiagen, Hilden, Germany). Total RNA including microRNAs was isolated using the miRNeasy Mini Kit (Qiagen). RNA yield and quality were determined with a NanoDrop 1000 Spectrophotometer (NanoDrop Technologies, Montchanin, DE) and a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA).

MicroRNA microarray expression data
Analyses were done initially with 12 matched malignant and nonmalignant sample pairs of set 1 ( Table 1). The stepwise statistical analyses of microarray intensity data are described and shown in the Supplemental Figure S1 as well as in Figs. 1 and 2. At first all single 24 intensity profiles were analysed, then the 12 intensity ratios of malignant to non-malignant samples and the presence-absence of sample signal intensities. As a result of these calculations and by application of a Venn diagram, 134 signal features were detected as differently expressed microRNAs and were used for further analysis.
This group of 134 candidate reporters was used for PCA on intensity profile levels ( Fig. 1) to compress the multi-dimensional data to lower dimensions. The PCA-plot visualizes that all tissue samples were classified into two distinct groups. Only two samples, a non-malignant and a malignant sample, showed an exception (Fig. 1).
Furthermore, a hierarchical 2D-cluster analysis was applied without any statistical cuts on all 24 intensity profiles and 134 reporters derived from the primary data-analysis steps. The Z-score was cropped to Ϫ2 to ϩ2 in the heat map (Fig. 2). MicroRNA reporters are displayed as dendrogram on top, and intensity profiles are displayed as tree on the left of the heat map. The heat map depicts two clearly separated clusters. One group corresponds to up-regulated (red squares) reporters, and the other cluster matches to down-regulated (green squares) reporters in ccRCC samples, respectively. Note that the 134 reporters comprise only 76 different human microRNAs. Most samples (left side of the heat map) were grouped correctly into two well-separated ccRCC and non-malignant clusters. Only two misclassified samples were detected. Again, one non-malignant sample (named 3NN) was found in the ccRCC cluster, and one sample (named 14NC) was sorted into the non-malignant cluster. Hence, the cluster analysis was in complete concordance with the PCA analysis.
To render the analysis more robust and to reduce the high number of differentially identified results, reporters comprising individual microRNAs were combined into single values, and a cut-off at two-fold absolute change between ccRCC and matched non-malignant ratio experiments was applied. We identified 33 microRNAs that were regulated over two-fold change ( Table 2). Among this result, 13 microRNAs were over-expressed, and 20 microRNAs were down-regulated in malignant ccRCC samples compared with the matched non-malignant samples.

MicroRNA RT-PCR expression analysis
Based on the microarray results, we examined the expression level of 16 (highlighted in grey) out of the 33 microRNAs with RT-PCR (Table 2). These microRNAs showed at least three-fold changed expression between malignant and the non-malignant samples by microarray analysis. For that purpose, two RNA pools from the same 12 matched RNA samples used for the microarray experiments were prepared mixing equal parts. In addition, we determined miR-16, RNU6B, RNU19 and Z30 as possible reference genes [25]. We excluded microRNAs from further quantifications if Ct values were Ͼ35 in one of the both pooled samples (microRNAs in italic letters in Table 2). Therefore, only 10 microRNAs (microRNAs in bold letters in Table 2) were quantified in the individual samples. The RNU19 as possible reference gene showed a Ct difference of 0.98 between both pools of the malignant and non-malignant samples and was also excluded from the following analyses. The Ct differences for miR-16, RNU6B and Z30 were Ͻ0.5 between the pool samples and were furthermore investigated by PCR.
Subsequently, two sample sets were analysed with RT-PCR: set 1 for validation of the microarray results and set 2 as an independent control (Table 1). In set 1, we used the same 12 sample pairs as used in the microarray analysis and carried out quantitative PCR of 11 microRNAs, RNU6B and Z30. The miR-16 expression  (Fig. 3, left blank columns ϭ set 1). The PCR results were in accordance with the microarray results since the same microRNAs were up-or down-regulated.

Fig. 2 Two-dimensional cluster analysis across intensity profiles (on left) and microRNA reporters (on top). The Matrix Viewer displays hierarchical trees (on top and left) and a heat map (bottom
To improve the accuracy and reliability of the study according to a sample size and power calculation (Supplemental Text S1), we included 72 additional sample pairs as set 2 (

Correlation between microRNAs and clinico-pathological data
The microRNA expression ratios were correlated among each other (Supplemental Table S1). The statistically significant Spearman's correlation coefficients ranged from rS ϭ Ϫ0.186 ROC analyses were performed to characterize the ability of microRNAs to discriminate malignant and non-malignant samples ( Table 3). The high areas under the ROC curves and the overall correct classification rates illustrate the high discriminatory power of the microRNAs. Particularly, the combination of miR-141 and miR-155 resulted in a 97% overall correct classification of samples (Table 3). However, none of the microRNAs was associated with the survival of the patients (P ϭ 0.203-0.864; Cox regression and Kaplan-Meier analysis), whereas the classical risk factors as tumour stage, grade and size were verified in our study group (Supplemental Table S2).

MicroRNA Fold changes † P values
Over-expressed

Putative miRNAs target genes
The putative targets of the dysregulated microRNAs provides a sound basis for further functional studies in ccRCC. For that purpose, various in silico methods for target gene prediction using different algorithms and ranking criteria have been developed [26]. To demonstrate an example, the corresponding targets of the strongest up-and down-regulated microRNAs were predicted using the computational programs TargetScan, PicTar and miRanda (Supplemental Table S3). Genes identified by at least two of the three programs were considered [27]. For the upregulated miR-210 and miR-155, 10 and 163 putative targets were identified, whereas 291 and 411 targets were found for the down-regulated miR-141 and miR-200c, respectively. Out of these potential 875 targets in total, 157 were commonly recorded by all three algorithms.

Discussion
The limited number of known microRNAs makes their analysis comparatively convenient, and a wealth of studies have attempted to use microRNAs for diagnostic purposes to classify tumours [28,29]. Although few cases of human clear cell renal cell carcinoma (ccRCC) have already been included in comparative microRNA studies, a larger series of renal tumours has not been profiled yet [17,30]. Nakada et al. [22] very recently published the microRNA profiles from 16 ccRCC and 4 chromophobe RCC samples in comparison with those from 6 normal renal tissue samples. We used an array-based screening step analysing 12 matched pairs of histologically confirmed tumours with adjacent normal tissue. To further validate the most significantly dysregulated microRNAs, a complex three-step validation was conducted. The first step was establishing stem-loop primer RT-PCR protocols using pooled sample material for the 16 microRNAs with a twofold change (tumour versus normal) in the array profiles. We found Ct values Ͼ35 for six microRNAs, which were therefore excluded from further analysis. The next step was to measure all 11 (including miR-16) differentially expressed microRNAs in those samples that had been subjected to array analysis. Generally, a good concordance of both techniques was seen. The third validation step was to confirm differential miRNA expression in an independent test cohort. Importantly, the analysis of the test set could fully reproduce the fold changes described in the training set, which clearly illustrates the high reproducibility of this microRNA profile.
These data demonstrate the power of microRNAs to differentiate between malignant and non-malignant tissue samples (Table 3). Several single microRNAs like miR-200c and miR-141 classified more than 90% of all samples correctly and the combination of the two microRNAs miR-141 and miR-155 identified 97%. To clarify, if this microRNA profile is specific to ccRCC, further analysis of other solid tumours is mandatory and underway. Other studies have demonstrated that microRNA profiles hold promise to classify tumours of uncertain histogenetical origin [31]. This is interesting, for conventional mRNA profiling techniques have yielded unsatisfactory results [32].
In a brief global assessment of our findings, we found more microRNAs down-regulated (n ϭ 20) than up-regulated (n ϭ 13) considering a fold change of two, which is in line with previous studies [19,22]. Also, the fold changes for the down-regulated microRNAs were slightly higher than for up-regulated microRNAs, which was particularly valid for miR-514 and miR-141. Interestingly, Gottardo et al. [17] analysed miR-profiles of 20 renal carcinomas and 3 non-matching normal renal tissues using an array platform that represented 245 microRNAs and found only 4 microRNAs significantly up-regulated in RCC (miR-28, miR-185, miR-7-2 and let-7f-2) with rather low mean fold changes (1.21-1.42), but no down-regulation of microRNAs. Also, there was no overlap with our data, which might be attributed to the different array platforms. A significantly better concordance was seen in comparison with the data of Nakada et al. [22], who used the same array and RT-PCR platforms as in our study. Here, 5 out of 15 down-regulated microRNAs were also among our list of down-regulated microRNAs. An even higher overlap was seen in the list of up-regulated microRNAs (five out of six). Also, no correlation was seen of microRNA expression with nuclear tumour grade, which we can confirm with our larger series of cases.
We can also confirm the strong down-regulation of miR-141, which belongs to the miR-200 family. With the exception of miR-200a, all of these are down-regulated in ccRCC and correlated to each other, which hints at a co-regulation. Using the target prediction softwares TargetScan, miRanda and PicTar and considering the very stringent strategy of only common targets for the three algorithms, the genes chromodomain helicase DNA-binding protein 9 (CHD9), embryonic lethal, abnormal vision Drosophila-like 2 (ELAVL2), oligodendrocyte transcription factor 3 (OLIG3) and protocadherin-8 (PCDH8) were identified as putative targets of the miR-141 and miR-200c (Supplemental Table S3). The relationship between oncogenesis and ELAVL2 and PCDH2 was already proven in other tumours but not in ccRCC [33,34]. So far, CHD9 and OLIG3 have not been related to tumourigenesis. Very recently, Gregory et al. [32] [8,35].
Among the list of differentially expressed microRNAs in ccRCC we also identified miR-21, a microRNA that has been implicated in blocking apoptosis in glioblastoma cells [36]. MiR-21 is also overexpressed in breast cancer and promotes cell proliferation [37]. Inhibition of apoptosis is a hallmark of cancer [38], and likewise up-regulation of miR-21 appears to be a microRNA-hallmark of cancer because it has been found in breast, colon, endocrine neoplasms of the pancreas, lung, prostate and gastric cancer [5,39].
Another microRNA from our profile commonly found up-regulated in human tumours is miR-155. Coded by the gene BIC, which cooperates with c-myc to induce lymphomagenesis in a chicken model of lymphoma, miR-155 was the first microRNA that was ascribed oncogenic properties [40]. Recent studies found miR-155 highly up-regulated in lymphomas [41,42] and various solid tumours [4,7,43,44]. A prognostic value of miR-155 has been reported for non-small cell lung cancer [4]. However, having profiled the six most differentially expressed microRNAs in a cohort of 84 clinically well-characterized RCC cases without demonstrating any prognostic value of any microRNAs attenuates our optimism to find novel prognostic markers using microRNAs.
The microRNA we found strongest up-regulated in ccRCC by RT-PCR is miR-210. This finding is not completely unexpected because the expression of miR-210 is strongly responsive to the hypoxia-inducible factor-1alpha (HIF-1alpha) [45]. Moreover, a majority of hereditary and sporadic clear cell carcinoma of the kidney is caused by an inactivation of the Von Hippel-Lindau (VHL) tumour suppressor gene [46] that leads to an accumulation of HIF-1alpha with subsequent activation of HIF target genes, including miR-210 [45]. The VHL dependency of miR-210 induction has been confirmed in cell line models of ccRCC [47]. Functionally, miR-210 over-expression inhibits apoptosis in cancer cell lines [45]. Recently, Ephrin-A3 has been shown a direct target of miR-210 affecting the migration, survival, apoptosis of cells and possibly modulating the angiogenic response to hypoxia [48]. Because we have no data concerning the VHL mutation status of our tumour cohort, we can only assume that the strong miR-210 upregulation we observed is caused by VHL inactivation that is nearly endogenous to ccRCC. However, we could not show a prognostic value of miR-210 as demonstrated for breast cancer [47]. We also identified specific microRNAs with a highly reproducible dysregulation in ccRCC that have not been described before and which could prove useful to differentiate solid tumours and their metastases.
The prediction of the putative targets of the differentially expressed microRNAs exemplified by the approach with the strongest up-and down-regulated microRNAs (Supplemental Table S3) showed that a single microRNA can affect hundreds of genes. Thus, it is obvious that our profiling study as a first systematic investigation on microRNAs in ccRCC can only be considered as a basic, but essential step towards identifying the putative functional role of microRNAs in this tumour entity. The newly introduced method of microRNA induction or knockdown on a proteome-wide scale could essentially support this aim [49].
Some limitations of this study should be mentioned and discussed. First, one critical point in each study is the number of samples. A sound survival analysis to assess the prognostic value of a marker generally requires a sample size and power calculated tumour cohort. The number of samples examined in this study based on a corresponding calculation (Supplemental Text S1) were appropriate to identify prognostic risk factors in RCC patients as shown by the survival data depending on the typical tumour characteristics like tumour stage and grade. Thus, the critical point of a limited number of samples is rather unwarranted. In addition, the similar expression results obtained in the stepwise validation procedure let us assume that the risk of type II error does not exist in this study. Moreover, the probability of a type I error could be excluded as far as possible because of the high significance level of differential expressions of microRNAs between malignant and non-malignant tissue samples ( Fig. 3; P Ͻ 0.0001). Second, the current study is limited by the strictly conservative data-reduction approach that used only microRNAs differentially expressed for at least two-fold using the microarray analysis and three-fold using the PCR data of the pooled samples from set 1. This approach was deliberately pursued in order to discover very distinctly expressed microRNAs in ccRCC. Third, this study was limited to the ccRCC sub-type; the other histological sub-types should be profiled as well in further studies. Fourth, the issue of appropriate reference genes for the relative quantification of microRNA quantitative RT-PCR data is up to now an open question [50]. To normalize the microRNA RT-PCR data, we initially analysed the candidate housekeeping genes RNU6B, RNU19, Z30 and miR-16 that were also used by others [32,[50][51][52][53]. Only RNU6B was stably expressed in both normal and tumour compartments. RNU6B is a small nuclear RNA with 106 nucleotides in contrast to the microRNAs with about 22 nucleotides. However, it is obvious that its similar expression in both sample groups ensures a more accurate normalization than microRNAs like miR- 16. In conclusion, this comprehensive study of microRNA profiles in human ccRCC provides plenty of new and valuable data and information concerning the differential expression of microRNAs. The microRNAs identified in this study might open up new therapeutic options for ccRCC, which poorly responds to conventional chemotherapy. Targeting microRNAs directly with antisense nucleotides, for example, by using so-called antagomirs that represent chemically engineered anti-sense nucleotides, can silence endogenous microRNA expression [54]. This novel approach appears promising and clearly warrants intense study.