Custom Search

Classification of Renal Cell Carcinoma Based on Expression of VEGF and VEGF Receptors in Both Tumor Cells and Endothelial Cells

Abstract and Introduction

Abstract

Recent development of antiangiogenic therapy for renal cell carcinoma (RCC) has significantly improved the treatment of these often refractory tumors. However, not all patients respond to therapy and assays for predicting outcome are needed. As a first step, we analyzed a retrospective cohort of tumors and assessed the ability of VEGF and VEGF receptors (VEGF-R1, -R2 and -R3) to classify tumors. We analyzed tissue microarrays containing 330 RCCs using a novel method of automated quantitative analysis of VEGF and VEGF-R expression by fluorescent immunohistochemistry. Expression of markers was separately quantified within three tissue components: tumor cells, endothelial cells and adjacent normal epithelium. VEGF and VEGF receptors were tightly coexpressed both within tumors and within adjacent normal cells (all P-values <0.001).>P<0.0001).>

Introduction

Despite the recent success in treating advanced renal cell carcinoma (RCC) with antiangiogenic therapies, surprisingly little is known about expression of targets of these drugs in renal tumors and microvessels. There is some evidence to suggest that antiangiogenic agents target the microvasculature,[1] but they may also target autocrine growth factor pathways within the tumor cells themselves. Two multitarget tyrosine kinase inhibitors have been approved by the Federal Drug Administration for use in unresectable RCC: sorafenib (Bayer Pharmaceuticals, Leverkusen, Germany and Onyx Pharmaceuticals, Emeryville, CA, USA) and sunitinib (Pfizer Inc., New York, NY, USA). Additional inhibitors of the VEGF pathway are in clinical trials. Sorafenib inhibits members of the RAF pathway, as well as VEGF-R2 (KDR/Flk-1), VEGF-R3 (Flt-4) and PDGFR-β, and sunitinib inhibits VEGF-R2, PDGFR-β, Kit and Flt-3. Temsirolimus (Wyeth Pharmaceuticals, Madison, NJ, USA), an mTOR inhibitor, was also approved for advanced RCC and additional VEGF-R pathway-targeting agents are in clinical development.

The advent of these antiangiogenic agents into the clinic for RCC was preceded by advances in our understanding of tumor vasculature. Mutations in the von Hippel-Lindau (VHL) tumor suppressor gene have been found in approximately 75% of clear cell RCCs,[2] resulting in induction of hypoxia-regulated genes in tumor cells, including VEGF.[3-5] Additional steps are required for vessel formation, including loss of integrity of the extracellular matrix.[6] Thus, inhibition of tumor angiogenesis is likely to require agents that target both the vessels and the malignant cells, as is evident by the lack of efficacy of monotherapy that directly targets VEGF.[1]

Coupled with the clinical development of VEGF-R-targeting therapies, there is an urgent need to develop biomarkers that predict response to these agents, as clearly only a subset of patients derive benefit from the drugs. Increased clinical benefit has been shown in patients who harbor VHL mutations, although other patients also derive benefit.[2] Additional approaches have been attempted to identify predictors of response, including assessment of target expression in tumor cells and tumor vasculature, interstitial fluid pressure, tumor oxygen tension, blood circulating endothelial cells, serum protein levels (such as VEGF) and imaging strategies that measure blood flow, as summarized by Jain et al.[1]

Immunohistochemistry (IHC)-based analysis of tumors is a practical approach to take for biomarker identification and validation, as IHC can be performed on small amounts of paraffin-embedded tissue. The typical first step in biomarker development is characterization of the biomarker in the disease population. Several studies have assessed expression of VEGF and VEGF receptors in RCC tumor cells; however few have separately analyzed the endothelial cell component, and none has used quantitative IHC to compare VEGF/VEGF-R within these different tumoral elements. Mertz et al[7] have used automated analysis to assess the microvessel density of RCC and found associations with aggressive disease, but did not look at the VEGF/VEGF-R pathway. Tsuchiya et al[8] assessed expression of VEGF, VEGF-R1 (Flt-1) and VEGF-R2 in RCC tumor, adjacent normal renal tissue and endothelial cells by standard IHC and RT-PCR. However, their analyses only included 23 cases, and no association was made with clinical/pathological variables. Other smaller studies assessing VEGF/VEGF-Rs and microvessel density have been conducted.[9-11] Jacobsen et al[12] assessed a relatively large cohort tissue microarray (TMA) for VEGF expression, but no assessment of VEGF-R expression and microvessel density was made. Our purpose was to assess expression of VEGF, VEGF-R1, VEGF-R2 and VEGF-R3 in three tissue components: RCC cells, endothelial cells and adjacent normal renal tissue on a large patient cohort with associated clinical/pathological data. We also assessed vessel area (VA) in the tumors. To obtain more accurate, objective measures of expression, we used our new method of automated quantitative analysis (AQUA). This method has been validated, and can be more accurate than pathologist-based scoring of brown stain.[13,14] As with some targeted therapies, it is possible that response to VEGF- or VEGF-R-targeting drugs might be associated with expression levels of targets in tumors or stroma, and quantitative assays need to be developed to predict response. Other markers that have both prognostic and predictive value have significantly impacted our ability to appropriately select therapeutic regimens for other cancers, and similar assays might be beneficial in selection of RCC patients for antiangiogenic therapies.

Materials and Methods

Tissue Microarray Construction

TMAs were constructed as described.[14] RCC cores from 334 patients, 294 with matching adjacent normal renal tissue, each measuring 0.6 mm in diameter, were spaced 0.8 mm apart on slides. Tumors were represented by two cores from different areas of tumor and adjacent normal kidney by one core. Specimens and clinical information were collected with the approval of a Yale University institutional review board. Histological subtypes included clear cell (71%), papillary (14%), chromophobe (2%), mixed histology (4%), oncocytomas (6%) and sarcomatoid tumors (3%). Oncocytomas were excluded from survival analyses. Among them, 8% had stage II and stage IV disease, 56% stage I and 28% stage IV; 12 were Fuhrman nuclear grade I, 52% grade II, 27% grade III and 9% grade IV. See Supplementary Figure 1 for survival based on tumor type, stage and grade. Specimens were resected between 1987 and 1999, with follow-up of 2-240 months (median, 89.7). Age at diagnosis was 25-87 years (median, 63). Treatment history was not available for the cohort.

Immunohistochemistry

For analysis of markers in tumor cells and normal epithelium, TMAs were stained with a cocktail of anti-cytokeratin and streptavidin (which binds endogenous biotin) and visualized with a green fluorophore (Alexa 488), in conjunction with an antibody to the target marker (VEGF or VEGF-R1, -R2 or -R3). Target markers were visualized by incubating with an appropriate horseradish peroxidase-conjugated secondary antibody and Cyanine-5 tyramide. Cyanine-5 was used because its far-red emission spectrum is outside of the range of tissue autofluorescence. For analysis of markers in microvessels, TMAs were stained with CD34 in conjunction with an antibody to the target marker. Details of methods and antibodies appear in the Supplementary Materials.

Automated Image Acquisition and Analysis

Images were acquired and analyzed using algorithms that have extensively been described.[13] Briefly, monochromatic, high-resolution (1280 × 1024 pixel) images were obtained of each histospot. Tumor was distinguished from stroma by the cytokeratin/streptavidin signal, and endothelial cells were distinguished using CD34. The target signal (VEGF, VEGF-R1, -R2 and -R3) from the tumor cells, adjacent normal epithelium or endothelial cells was scored on a scale of 0-255, and expressed as the average signal intensity within the assayed component (AQUA score). Histospots containing <3%>

Statistical Analysis

The StatView (SAS Institute Inc., Cary, NC, USA) and R[15] software packages were used. AQUA scores for replicate tumor cores were averaged. Unsupervised hierarchical clustering was performed using TreeView and Cluster software.[16] For clustering, AQUA scores were converted into z-scores and analysis was limited to cases with scores for >80% of markers.[17] The prognostic significance of parameters was assessed for predictive value using the Cox proportional hazard model (PHM) with RCC-specific survival as an end point. Kaplan-Meier survival curves were generated for patient subsets defined by the cluster analyses, with significance evaluated using the Mantel-Cox log-rank test and multivariate Cox PHMs. Correlations between markers were assessed using the Spearman's ρ-test. Variables were univariately and bivariately entered into Cox PHMs to assess the significance at α=0.05. To evaluate whether individual variables and combinations of variables are able to predict if a patient died of disease or was alive at 10 years, leave-one-out cross-validation (LOOCV) was employed with logistic regression models. This form of cross-validation iteratively splits the N observations into a training set of size N-1 and a test set of size 1. At each iteration the coefficients for the logistic model are estimated based on N-1 observations in the training set, and used to predict the one observation in the test set. Misclassification was assessed based on whether the observation is incorrectly or correctly predicted. The model's prediction performance, or error, was calculated as the average misclassification over the N iterations. The lower the average misclassification, the better the prediction.[18] Confidence intervals (CIs) for prediction error estimates were constructed by nonparametric bootstrap resampling.[19] See Supplementary Information for details.