Clear Cell Renal Cell Carcinoma is linked to Epithelial-to-Mesenchymal Transition and to Fibrosis

Abstract

Clear cell renal cell carcinoma (ccRCC) represents the most common type of kidney cancer with high mortality in its advanced stages. Our study aim was to explore the correlation between tumor epithelial-to-mesenchymal transition (EMT) and patient survival. Renal biopsies of tumorous and adjacent nontumorous tissue were taken with a 16 g needle from our patients (n = 26) undergoing partial or radical nephrectomy due to ccRCC RNA sequencing libraries were generated using Illumina TruSeq^® Access library preparation protocol and TruSeq Small RNA library preparation kit. Next generation sequencing (NGS) was performed on Illumina HiSeq2500. Comparative analysis of matched sample pairs was done using the Bioconductor Limma/voom R-package. Liquid chromatography-tandem mass spectrometry and immunohistochemistry were applied to measure and visualize protein abundance. We detected an increased generic EMT transcript score in ccRCC Gene expression analysis showed augmented abundance of AXL and MMP14, as well as down-regulated expression of KL (klotho). Moreover, microRNA analyses demonstrated a positive expression correlation of miR-34a and its targets MMP14 and AXL Survival analysis based on a subset of genes from our list EMT-related genes in a publicly available dataset showed that the EMT genes correlated with ccRCC patient survival. Several of these genes also play a known role in fibrosis. Accordingly, recently published classifiers of solid organ fibrosis correctly identified EMT-affected tumor samples and were correlated with patient survival. EMT in ccRCC linked to fibrosis is associated with worse survival and may represent a target for novel therapeutic interventions.

Keywords: Clear cell renal cell carcinoma; epithelial‐to‐mesenchymal transition; fibrosis.

Figure 1

Epithelial‐Mesenchymal Transition (EMT) score of 26 adjacent normal‐tumor pairs from renal clear cell carcinoma patients. P‐value shown is computed by matched‐pairs two‐sided Wilcoxon signed rank test. NO, normal, non‐carcinoma tissues; TU, respective ccRCC samples.

Figure 2

RNA sequencing data analysis. (A) Unsupervised hierarchical cluster analysis with 137 differentially expressed EMT‐related mRNA. The samples segregate into two large clusters, the normal (“NO”) samples, and the ccRCC samples (“TU”). (B) Scatterplot illustration of expression levels of selected genes.

Figure 3

Principal component analysis with differentially expressed EMT genes. (A) Principal component analysis (PCA) with 137 differentially expressed EMT‐mRNA. The sample groups “healthy” (normal, NO) and “ccRCC” (tumor, TU) are separated along the principal component 1 (PC1). Ellipsoids indicate the 95th percentile of data points per group. (B) The biplot demonstrates the samples as dots, and the contribution of each variable toward the explanation of the variance of the data as red arrows. Each arrow can be attributed a value, the loading score. (C) Ten mRNA with the largest absolute loading scores. ITGA5, AXL and MMP14 are among the mRNAs with the largest loading scores.

Figure 4

Bivariate polynomial regression of the mRNA abundance and cancer stage. CAV1,VIM,IGFBP3, and ITGA5 were the mRNAs with strongest regression. The expression level for all mRNAs declined in samples with the highest tumor stage. The dark red area is the confidence limit for the expected fitted mean, the light red area displays the confidence limits for the individual predicted value. The confidence limits reflect variation in the error and variation in the parameter estimates. Cancer stage (see also Table 1) is indicated as numerical values from 1 to 3, with 0 indicating tumor‐unaffected status corresponding to the respective noncancerous normal tissues.

Figure 5

Detection of expression changes of MMP14 in ccRCC. Top left (“RNA‐seq”): MMP14 mRNA is about 1.46‐logfold increased in ccRCC. Matched normal samples from the same patients are connected by a line to illustrate the expression change. Top right (“Proteomics”): Increased abundance of MMP14 protein in ccRCC. Two samples, patients 27 and 29, showed decrease in MMP14 expression in ccRCC. In RNA sequencing data, MMP14 mRNA levels were increased in ccRCC also for these two patients. Matched normal and ccRCC samples from the same patient are connected to illustrate the expression change. Bottom (“IHC”): Representative immunohistochemistry (IHC) results showing the increased detection of MMP14 epitope in an ccRCC sample (“TU”) in comparison to the matched healthy (“NO”) sample from the same patient.

Figure 6

Representative immunohistochemistry (IHC) analyses of protein abundance by proteomics of three genes involved in EMT. Protein level of AXL and VIM were increased in ccRCC, while CDH1 protein was decreased (A). IHC results are supported by proteomics data for VIM and CDH1 (B). AXL protein was not detected in the proteomics dataset.

Figure 7

Correlation of expression levels of miR‐34a and target genes AXL and MMP14. Pearson correlation of expression levels of miR‐34a and the target genes AXL (A) and MMP14 (B), of AXL and MMP14 (C), and all mir‐34a and AXL and MMP14 (D). Red dots indicate normal samples, blue dots ccRCC tumor samples. Ellipsoids indicate the 95th percentile of data points per group.

Figure 8

Survival analysis for patients with ccRCC based on individual genes. Higher mRNA expression levels of MMP14 (A and B), and AXL (C and D) are significantly associated with worse survival. Higher KL mRNA expression levels are significantly associated with improved patient outcome (E and F).

Figure 9

Survival analysis for patients with ccRCC based on two gene panels. Performance of a panel of four collagen genes with co‐expression in the matched pairs data (A and B). Performance of a panel of eleven genes with superior rank hazard ratio, and their expression level in risk groups (C and D). The genes have been selected based on their performance in the survival analysis from a set of 20 genes with highest TU/NO‐fold change listed in Table 2.

Figure 10

Fibrosis classifier. Two MARGS‐based classifiers of fibrosis are diagnostic of ccRCC. Most of the genes of the classifier panels which had been developed to diagnose interstitial fibrosis in renal allografts (19 genes, (Rodder et al. 2009)), and solid organ fibrosis (10 genes, (Rodder et al. 2011)) are differentially expressed in the ccRCC dataset (A). Applying linear discriminant analysis with the 19‐gene panel (B) or with the 10‐gene panel (C) has a 100% correct classification rate, and indicates a role of fibrosis in ccRCC.