Treatment-independent miRNA signature in blood of wilms tumor patients

Background Blood-born miRNA signatures have recently been reported for various tumor diseases. Here, we compared the miRNA signature in Wilms tumor patients prior and after preoperative chemotherapy according to SIOP protocol 2001. Results We did not find a significant difference between miRNA signature of both groups. However both, Wilms tumor patients prior and after chemotherapy showed a miRNA signature different from healthy controls. The signature of Wilms tumor patients prior to chemotherapy showed an accuracy of 97.5% and of patients after chemotherapy an accuracy of 97.0%, each as compared to healthy controls. Conclusion Our results provide evidence for a blood-born Wilms tumor miRNA signature largely independent of four weeks preoperative chemotherapy treatment.


Background
Wilms tumor is a common childhood tumor and affects very young children [1]. Most of cases are detected at the median age of 36 months. In Europe, children older than six months are treated with preoperative chemotherapy according to the SIOP protocol (Societé Internationale d'Oncologie Pédiatrique) resulting in a smaller tumor volume when children undergo surgery. The primary aims of the SIOP study are risk-stratified therapies, dimished toxicity and an improved patient outcome. Since children in early stages of Wilms tumor development do not show characteristic symptoms, and low grade tumors have a better prognosis than high grade tumors [2], an early detection marker is useful to increase the overall survival rates. Most recently, we provided strong evidence that blood-born microRNA (miRNA) signatures are highly indicative for various human diseases [3]. MiRNAs are small non-coding RNAs between 17 and 24 nt. They are involved in physiological and pathological processes *Correspondence: jana.schmitt@uks.eu † Equal contributors 1 Department of Human Genetics, Saarland University, 66421 Homburg/Saar, Germany Full list of author information is available at the end of the article through the regulation of gene expression and may indicate tumor development at very early stages [4,5]. Specific miRNA signatures have already been described for clear cell renal cell carcinoma (ccRCC) [6]. They allow to differentiate metastatic and non-metastatic ccRCC and indicate progression-free survival and overall survival. MiRNA profiles of different human kidney cancer subtypes have been established by Petillo et al. [7]. They were able to discriminate chromophobe RCC from oncocytomas, ccRCC from papillary RCC and prognostic subgroups of ccRCC. While these renal miRNA patterns have been established on tumor tissues, in our study we analyzed the miRNA expression pattern in blood cells of Wilms tumor patients. Specifically we compared untreated children with Wilms tumor and children treated according to SIOP protocol. We asked if chemotherapy affects the miRNA expression patterns. In addition, we asked if these expression patterns are different between Wilms tumor patients and healthy controls.

Samples
In total, we collected more than hundred blood samples from Wilms tumor patients. To arrvive at homogenous http://www.  Table S2.

RNA isolation and microarray screening
To isolate the RNA including the miRNA from whole blood samples, we used the miRNeasy Kit (Qiagen GmbH, Hilden) as previously described [8]. We analyzed the total RNA by using the Geniom RT Analyzer (febit biomed GmbH, Heidelberg, Germany) employing the MPEA-assay for miRNA-analysis. This assay allows for detection of microRNAs based on a combination of stringent hybridization and enzymatic primer extension on a microfluidic microarray starting from total RNA material, without the need for enrichment, amplification or labeling of the native RNA samples [9]. The microfluidic biochip (Geniom Biochip Homo sapiens v12, febit biomed GmbH, Heidelberg, Germany) contained 7 replicates of 848 miRNAs [10,11]

Statistical analysis
The background corrected and log transformed microarrays were quantile normalized using the freely available R software [12]. Differentially expressed miRNAs were identified by employing an unpaired two-tailed t-test. The resulting p-values were adjusted for multiple testing by applying the Benjamini-Hochberg procedure (false discovery rate (FDR) adjustment) [13]. In addition, we computed AUC (Area Under the Curve) values for each miRNA. AUC values measure how well the expression values of a single miRNA can separate two groups, e.g. patients versus controls. An AUC value of 0.5 indicates equal distribution of the two groups to compare; this means that a miRNA cannot be used to separate the groups. A miRNA reaching an AUC value of 1 (0) has expression values that are throughout lower (higher) in the diseased group than in the control group corresponding to the most diagnostic information of a biomarker. Besides this single biomarker analysis, we classified the samples using the normalized and log transformed miRNA expression profile with machine learning procedures for calculating Support Vector Machines (SVM) implemented in the R e1071 package. We performed the classification with a linear SVM employing 100 repetitions of standard 10-fold cross validation and calculated mean http://www.biomedcentral.com/1471-2164/13/379 sensitivity, specificity, and accuracy. Clustering has been carried out using complete linkage hierarchical clustering. The Euclidian distance measure was employed to compute the dissimilarity of miRNA rows and sample columns independently of each other. We used the heatmap.2 function of the gplots R package to compute and plot the results of the clustering. For validating the microarray results, we applied quantitative real-time PCR on six selected miRNAs and computed the fold change by using the 2 − Ct method [14].

Results
In total, we collected 43 Wilms tumor samples for analysis. To generate miRNA profiles, we screened 848 mature miRNA transcripts. After applying background correction, log transformation, and normalization procedures, we tested if the miRNA profile changes after chemotherapy. To this end, we applied the paired two-tailed t-test for the matched samples of 13 Wilms tumor patients before and after chemotherapy. We considered only miRNAs with more than a two-fold expression difference between the two groups. Using the remanining 214 miRNAs, our analysis did not indicate a significant difference in miRNA profiles of Wilms tumor patients prior to chemotherapy and Wilms tumor patients after chemotherapy. Specifically, we did not find significantly deregulated miRNAs using a significance threshold of 0.05 after FDR adjustment. In addition, we repeated the comparison of Wilms patients before and after therapy using all 43 samples and applying unpaired two-tailed t-test, which confirmed the results of the matched samples (no significant miRNA using a significance threshold of 0.05 and FDR adjustment). The lack of a difference between post-and pretreatment samples prompted us to ask whether Wilms tumors are characterized by a blood-born miRNA signature.  Table S1. An overview of the 20 most significantly deregulated miRNAs can be found in Table 1.
Using hierarchical clustering with the Euclidian distance measure, we analyzed how the 42 Wilms tumor and control samples relate to each other. For this task we used the 100 most variable miRNAs out of the 848 miRNAs. Figure 1 shows the resulting heatmap of the hierarchical clustering. The control samples clearly cluster different from the Wilms tumor samples. To determine if the expression pattern of the 848 miRNAs are able to separate Wilms tumor patients prior to chemotherapy from controls, we used a Support Vector Machine (SVM) with linear kernel. Briefly, we applied 100 repetitions of 10-fold cross-validations and computed the mean accuracy, sensitivity, specificity, and the corresponding 95% confidence intervals. Using this clustering, we were not able to separate Wilms tumor patients prior to chemotherapy and patients after chemotherapy, confirming the fold-expression results as shown above. But we were able to reach high values for the separation between blood samples taken from Wilms tumor patients prior to chemotherapy and samples taken from healthy controls with very low variances. In detail we found 97.5% accuracy (97.4 -97.7%), 99.8% sensitivity (99.6 -99.9%), and 94.7% specificity (94.6-94.8%) for this separation.
Likewise we found comparable results for the separation between samples taken from Wilms tumor patients after chemotherapy and samples taken from controls. In detail we calculated an accuracy of 97.0% (96.7 -97.3%) a sensitivity of 99.1% (98.5 -99.7%) and a specificity of 94.8% (94.7 -95.0%). These results support the abovementioned results that showed no significant difference of the blood-born miRNA signature between patients prior and after chemotherapy. Finally, we determined those miR-NAs that are differently regulated in both comparisions between controls and samples taken prior to preoperative chemotherapy and between controls and samples taken after preoperative chemotherapy. We found a total of 106 miRNAs that were differentially regulated in both comparisons (adjusted t-test, significance level 0.05) (see Figure 2). Interestingly, the overlap consists of 43 miRNAs that are down-regulated and of 63 miRNAs that are upregulated in both comparisons. These miRNAs are likely to contribute most to the treatment independent Wilms tumor signature.
To validate our microarry results, we selected six miR-NAs for performing quantitative reverse transcription PCR (qRT-PCR). In total, we chose three miRNAs that were up-regulated (miR-520d-3p, miR-197, miR-224), and three that were down-regulated (miR-20a, miR-126, miR-144*) in the microarray experiment comparing Wilms tumor samples against normal controls. We performed the qRT-PCR in duplicates with 10 samples per group. The qRT-PCR results for the tested miRNAs were largely concordant with the microarray data (Figure 3), except http://www.biomedcentral.com/1471-2164/13/379 The 20 most significant miRNAs when comparing Wilms patients before therapy to healthy controls. Down-regulated fold-change is indicated by negative values, up-regulated fold-change is indicated by positive values.
for miR-520d-3p, which is down-regulated in qRT-PCR instead of its up-regulation in the microarry experiment.

Discussion
As shown in several studies, miRNA expression signatures allow differentiation between human cancers [15,16]. There is also evidence that miRNA signatures correlate with metastasis, progression-free survival and overall survival [6]. As for human kidney cancer, specific signatures have been reported for clear cell renal cell carcinoma (ccRCC), papillary RCC and prognostic subgroups of ccRCC [7]. While the majority of studies focused on miR-NAs in tissues, there is increasing evidence that miRNAs isolated from body fluids like serum, plasma or blood cells, can serve as an informative marker for different cancer types like ovarian cancer, prostate cancer, lung cancer, colorectal cancer as reviewed by Cortez et al. [17]. Most recently we reported specific miRNA signatures of peripheral blood cells in a variety of cancer and non-cancer diseases. In detail, we detected consistently deregulated profiles for lung cancer, prostate cancer, melanoma, ovarian cancer, gastric tumors, pancreatic tumors, multiple sclerosis, chronic obstructive pulmonary disease (COPD), sarcoidosis, periodontitis, pancreatitis and acute myocardial infarction [3]. Based on our data and the data of others [18], the overall pattern of miRNAs from blood cells seems to be rather robust. In addition, blood cells allow isolation of rather large amounts of miRNAs that in turn are sufficient for next generation sequencing of miRNAs [19].
In the present study, we were able to differentiate between healthy controls and samples that were taken from Wilms tumor patients prior to chemotherapy with an accuracy of 97.5%. Several of the miRNAs that were significantly deregulated in this comparison have been associated with non-tumor diseases including miR-144* that is overexpressed in patients with active tuberculosis (TB) [20], miR-1246 that seems to play a role in the downregulation of the Down Syndrome-associated DYRK1A [21] and miR-216a that was deregulated in mouse kidney with diabetic nephropathy [22]. In addition some miRNAs that we found differentially expressed in our analysis, have previously been associated with cancer including miR-20b that is decreased in plasma vesicles of non-small cell lung cancer patients [23], miR-197 that is under-expressed in http://www.biomedcentral.com/1471-2164/13/379 oral tumor tissue [24], and miR-144 that has been generally described as tumor specific miRNA by Wang et al. [25].
Notably, some of the miRNAs that we found differentially expressed between Wilms tumor patients and controls have previously been related to kidney cancers. Specifically miR-20a that is a member of the miR-17-92 cluster, has been correlated to alveolar rhabdomyosarcoma as reported by the Children's Oncology Group (COG) [26] and is implicated in pathways linked to Wilms tumor [27]. There is evidence that miR-106b, which is highly expressed during nephrogenesis in nephron progenitors [28], is a predictive marker of early metastasis after nephrectomy in RCC patients [29]. The miR-17-92 http://www.biomedcentral.com/1471-2164/13/379 cluster, also known as oncomir-1, is a miRNA polycistron that contains some of the most potent oncogenic miRNAs. The oncogenic effect of oncomir-1 was not only found in kidney cancer [30,31] but also in a broad spectrum of other malignancies including e.g. colon [32], bladder [33] or gastrointestinal cancer [34]. miR-126 a known angiogenic miRNA allowed differentiation between papillary RCC and ccRCC [35]. miR-126 has also been proposed to separate conventional from papillary renal tumors [36]. Increased levels of miR-224 have been found in ccRCCs [37].
Another interesting result of this study is that we did not find significantly deregulated miRNAs between Wilms tumor patients prior and after chemotherapy. Presently, we conclude that the tumor type seems to mainly determine the miRNA signature that is found in blood cells of the patients. Since the analyzed blood cells comprise B cells, T cells and NK cells, it is legitimate to speculate that the identified miRNA signatures reflect a tumor type specific immune response. This hypothesis awaits of course future experimental confirmation.
We are well aware that healthy adults are not the optimal control for diseased children. The improvement for blood withdrawal of healthy children is, however, a major hurdle for the time being, we do not have such an approval by the local ethic committee. However, we compared our Wilms tumor samples to samples taken from children with other tumor diseases. We have preliminary data that seems to indicate that these both groups can be separated by miRNA signatures (data not shown). These results provide further evidence for a Wilms specific miRNA signature.