Identification and Functional Characterization of pVHL-Dependent Cell Surface Proteins in Renal Cell Carcinoma^,

Abstract

The identification of cell surface accessible biomarkers enabling diagnosis, disease monitoring, and treatment of renal cell carcinoma (RCC) is as challenging as the biology and progression of RCC is unpredictable. A hallmark of most RCC is the loss-of-function of the von Hippel-Lindau (pVHL) protein by mutation of its gene (VHL). Using the cell surface capturing (CSC) technology, we screened and identified cell surface N-glycoproteins in pVHL-negative and positive 786-O cells. One hundred six cell surface N-glycoproteins were identified. Stable isotope labeling with amino acids in cell culture-based quantification of the CSC screen revealed 23 N-glycoproteins whose abundance seemed to change in a pVHL-dependent manner. Targeted validation experiments using transcriptional profiling of primary RCC samples revealed that nine glycoproteins, including CD10 and AXL, could be directly linked to pVHL-mediated transcriptional regulation. Subsequent human tumor tissue analysis of these cell surface candidate markers showed a correlation between epithelial AXL expression and aggressive tumor phenotype, indicating that pVHL-dependent regulation of glycoproteins may influence the biologic behavior of RCC. Functional characterization of the metalloprotease CD10 in cell invasion assays demonstrated a diminished penetrating behavior of pVHL-negative 786-O cells on treatment with the CD10-specific inhibitor thiorphan. Our proteomic surfaceome screening approach in combination with transcriptional profiling and functional validation suggests pVHL-dependent cell surface glycoproteins as potential diagnostic markers for therapeutic targeting and RCC patient monitoring.

Introduction

The prognosis of advanced renal cell carcinoma (RCC) is poor. Most RCC cases are insusceptible to chemotherapies or radiotherapies, which seems to be partly caused by a hypoxia-mediated resistance to radiotherapy and alkylating agents [1–4]. Since 2006, novel anti-angiogenic targeted therapies have been available for patients with metastatic RCC. These therapies rely on a targeted blockade of angiogenic signaling caused by an aberrant overexpression of proangiogenic cell surface proteins, for example, vascular endothelial growth factor and platelet-derived growth factor as well as their corresponding receptors [5–8]. Despite these novel drugs, identification of additional cellular treatment targets, screening markers for early stages, recurrent tumors after nephrectomy, and predictive markers for treatment response is of utmost importance to further improve prognosis of RCC patients.

The discovery of cell surface glycoprotein biomarkers for RCC would be clinically useful because these glycoproteins are exposed to the cellular microenvironment and are, therefore, easily accessible by affinity-based probes such as drugs and antibodies for disease monitoring. Furthermore, cell surface glycoproteins can be shed into the bloodstream and could provide potential targets for recently developed remote-sensing serum-screening strategies [9]. However, the identification of the cellular surfaceome is challenging owing to its lower abundance compared with intracellular subproteomes, its hydrophobicity due to transmembrane domains, and the high degree of posttranslational modifications on these proteins. One solution to overcome this abundance problem is to focus on the cotranslational modification of cell surface proteins with glycostructures. The recently developed mass spectrometry (MS)-based cell surface capturing (CSC) technology is through the glycosylation of cell surface proteins for specific enrichment of this subproteome and subsequent identification [10]. Glycoproteins are also of particular clinical interest given that approximately 80% of all current protein drug targets are glycoproteins [11].

Strategies for the identification of RCC glycoprotein biomarkers can be combined with available knowledge about cancer-causing mutations. The von Hippel-Lindau (VHL) gene is the most commonly inactivated gene in the most frequent subtype clear cell RCC (ccRCC). More than 70% of ccRCCs are characterized by somatic mutations or hypermethylation of VHL [12–16], whereas in the rare papillary and chromophobe RCC subtypes, VHL is hardly affected. VHL encodes two proteins of 30 and 19 kDa, the latter being a result of an alternative, in-frame translation initiation codon [17]. Both proteins seem to differ in part in their subcellular localization, which may imply potential functional differences [18,19]. The most described function of pVHL is its role as a substrate recognition component of an E3 ubiquitin protein ligase complex known to target the α-subunits of the hypoxia-inducible factor (HIF) for ubiquitin-mediated proteolysis in a prolyl-4-hydroxylation-dependent manner under normoxic conditions [20–23]. HIF is a sequence-specific heterodimeric transcription factor composed of an α subunit (HIF-1α, HIF-2α, or HIF-3α) and an invariable β subunit. It promotes adaption and survival under low oxygen by up-regulation of genes that are involved in neovascularization, pH regulation, glucose uptake and glycolysis, apoptosis, and cell cycle [24]. In ccRCC, the HIF transcriptional program is uncoupled from changes in oxygen availability and co-opted by the cancer cell to promote tumor growth. Interestingly, vascular endothelial growth factor and platelet-derived growth factor, the two major angiogenic factors successfully targeted by the current above-mentioned therapies, are HIF target genes.

Here we used the CSC technology to screen and identify N-glycosylated cell surface proteins in pVHL-negative and -positive cell lines. The pVHL-dependent glycoproteins detected within the initial screen were subsequently verified and validated in vitro and in situ at the transcript and at the protein level by using a combination of low-density arrays (LDAs) and tissue microarrays (TMAs), respectively. Our pVHL-guided proteomic surfaceome screening approach in combination with transcriptomic validation and functional characterization revealed pVHL-regulated glycoproteins as candidates for the clinical use in patients with RCC.

Experimental Procedures

Cell Culture and Metabolic Labeling

The ccRCC-derived cell line 786-O, the human embryonic kidney cell line HEK-293, and the human fibrosarcoma cell line HT-1080 were supplied by American Type Culture Collection (Manassas, VA). Stable transfectants of 786-O reexpressing pVHL₃₀ were generated as mentioned before [19]. RCC4 cells and their corresponding transfectants reexpressing wild-type pVHL₃₀ were created as previously described [19,25]. Human renal proximal tubular epithelial cell line HK-2 was provided by R. Wuthrich (Clinic for Nephrology, Department of Internal Medicine, University Hospital Zurich, Zurich, Switzerland). RCC4, 786-O, HEK-293, and HK-2 cells were cultured as reported previously [26]. Hypoxic experiments were performed for 16 hours at 1% oxygen using a hypoxic workstation (Invivo₂ 400; Ruskinn Technology, Leeds, United Kingdom). Dimethyloxalyl glycine (DMOG) (Cayman Chemical, Ann Arbor, MI) treatment was performed for 24 hours at 1 mM. Stable isotope labeling with amino acids in cell culture (SILAC) was used for the in vivo incorporation of isotopically labeled amino acids into all cellular proteins. For SILAC labeling, 786-O and their stable transfectants reexpressing pVHL₃₀ were grown in l-lysine and l-arginine-deficient Dulbecco modified Eagle medium (Sigma) containing 10%vol/vol dialyzed fetal calf serum (FCS), 2 mM glutamine, and either normal or heavy lysine/arginine. Cells were grown for five to six generations. Then, 5 x 10⁷ 786-O and 786-OpVHL₃₀ were used for CSC.

CSC-Based Proteomic Surfaceome Screen

Cells were labeled in 10 ml of labeling buffer containing 5 mM biocytin hydrazide for 60 minutes. On labeling, the cells were harvested by scraping. Oxidation of cell surface glycoproteins subsequent cell surface labeling, cell lysis, membrane preparation, and final processing, and isolation of N-glycopeptides was performed as described in Wollscheid et al. [10].

Mass Spectrometry

Peptide samples were analyzed on a hybrid LTQ-FT-ICR instrument (Thermo Electron, Bremen, Germany) interfaced with a nanoelectrospray ion source. Before the LTQ-FT-ICR analysis, peptides were separated on a RP-HPLC column (75 µm x 15 cm) packed inhouse with C18 resin (Magic C18 AQ 3 µm; Michrom BioResources, Auburn, CA) using a linear gradient from 96% solvent A (0.15% formic acid) and 4% solvent B (98% acetonitrile, 2% water, 0.15% formic acid) to 25% solvent B for 60 minutes at a flow rate of 300 nl/min performed by a Tempo Nano 1D+ HPLC system (Applied Biosystems/MDS Sciex, Foster City, CA).

Each MS1 scan was followed by CID scans of the five most abundant precursor ions with dynamic exclusion for 30 seconds. Only MS1 signals exceeding 150 counts were allowed to trigger MS2 scans with wideband activation disabled. Total cycle time was approximately 1 to 1.5 seconds. For MS1, 3 x 10⁶ ions were accumulated in the ICR cell for a maximum time of 500 milliseconds and scanned at a resolution of 100,000 full width at half maximum (at 400 m/z). MS2 spectra were acquired in the linear trap using the normal scan mode, a target setting of 104 ions and accumulation time of 100 milliseconds. In the LTQ-FT analysis, singly charged ions and ions with unassigned charge state were excluded from triggering MS2 events. The normalized collision energy was set to 32%, and one microscan was acquired for each spectrum.

Data Analysis

The raw data acquired by the LTQ-FT instrument (software, Xcalibur 2.0; Thermo Scientific, Waltham, MA) was converted to mzXML using ReAdW (3.5.1) applying default parameters. MS/MS scans were searched against the human International Protein Index (IPI, version 3.15, 58099 protein entries) database using SEQUEST version 27 with the following parameters. For LTQ-FT data, precursor mass tolerance was set to 0.1 Da, a minimum of one tryptic termini was allowed, and variable modifications were allowed on asparagines (0.9840), on lysines (8.014199), and on arginines (10.008269), with a maximum of four modifications per peptide. No static modifications were defined. Data were further processed by the Trans-Proteomic Pipeline TPP v2.9 including PeptideProphet and ProteinProphet to estimate the false discovery rate. A protein probability cutoff of 0.8 was used, which corresponds to a false discovery rate of approximately 1%.

Quantification

For obtaining quantitative ratios between the differentially SILAC-labeled samples, we applied automated statistical analysis on protein (ASAP) ratio v3.0 [27]. The algorithm was run with a fixed scan range for the light and the heavy ions and with no background subtraction for the peak integration. ASAP ratio is an integrated part of the TPP and it normalizes the ratio distribution and computes a protein ratio. The results were curated manually.

RNA Isolation and LDA Applications

RNA from frozen tissue was isolated using the RNAeasy Mini Kit (Qiagen, Hilden, Germany). The quality and quantity of the isolated RNA were analyzed by means of nanodrop and Agilent Bioanalyzer 2100. Reverse transcription of isolated RNA was performed by the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) according to the manufacturer's instructions. Pilot experiments using RNA derived from tumor cell lines and tissue were performed with selected TaqMan gene expression assays to verify the reverse transcription step linearity and to determine the amount of cDNA, which can be loaded in polymerase chain reaction (PCR) without inhibition. Applied Biosystems TaqMan LDAs were configured using the online software tool TaqMan Custom Array (https://produ cts.appliedbiosystems.com). Only assays whose probes span an exon junction were chosen to avoid detection of genomic DNA. Preparation and loading of the sample-specific PCR mix into the fill reservoirs of the TaqMan Microfluid Cards were accomplished according to the manufacturer's instructions. Relative levels of gene expression were determined from the fluorescence data generated during PCR using the Applied Biosystems 7900HT Fast Real-Time PCR System followed by computational analysis by means of SDS and RQ manager software with HPRT1 and 18S rRNA as internal control.

Quantitative Reverse Transcription-PCR

Total RNA from pools of clones of 786-O and 786-OpVHL₃₀ cells, lentivirally infected with shHIF-2α (target constructs 1 and 4) as well as uninfected control cells, was prepared using the RNAeasy Mini Kit (Qiagen) according to the manufacturer's protocol. Reverse transcription was performed by High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). cDNA reverse transcribed from 50 ng of total RNA was used for quantitative reverse transcription (RT)-PCRs containing 10 µl of TaqMan 2x PCRMasterMix (Applied Biosystems) and 1 µl of TaqMan Gene Assay (Applied Biosystems) for CD10 (Hs00153519_m1) or 200 nM primers and 100 nM of the universal probe library probe 55 (Roche, Mannheim, Germany) for HIF-2α in a volume of 20 µl. Detection was carried out with a Fast Real-Time PCR System (7900HT; Applied Biosystems). Data analysis was performed using the SDS and RQ manager software (Applied Biosystems). Peptidyl-prolyl cis-trans isomerase A (PPIA) served as an internal control (Hs99999904_m1).

Tissue Specimen and TMA Construction

We used TMAs comprising 264 ccRCC collected from the University Hospital of Zurich (Zurich, Switzerland) to validate the in vitro data. All tissue samples were histologic reviewed by one pathologist (H.M.) and selected on the basis of hematoxylin and eosin-stained tissue sections [28]. A more comprehensive TMA study was performed to confirm the HIF dependence of CD10 in ccRCC. We used three TMAs comprising 831 ccRCC collected at the University Hospital of Zurich (Zurich, Switzerland), the Kantonsspital St. Gallen (St. Gallen, Switzerland), and the University Hospital of Basel (Basel, Switzerland). The composition of the TMAs and clinicopathologic data were previously described [29]. This study was approved by the local ethics commission (reference number StV 38-2005).

Immunohistochemistry

TMA sections (2.5 µm) were transferred to glass slides followed by immunohistochemical analysis according to Ventana automat protocols (Ventana Medical System, Tucson, AZ). Antibodies were applied for detection of AXL (1:200, AF154; R&D Systems, Planegg-Martinsried, Germany), CAIX M75 (1:200; kindly provided by J. Zavada, Institute of Molecular Genetics, Prague, Czech Republic), CD10 (1:30, NCL-CD10-270; Novocastra Laboratories, Newcastle, United Kingdom), CD13 (1:200, NCL-CD13-304; Novocastra Laboratories Ltd), CD31, CD34, CD99 (1:25, NCL-CD99; Novocastra Laboratories Ltd), CCND1 (prediluted, Ventana, P2D11F11 [29]), E-CADH (1:10, ECH-6; Cell Marque, Rocklin, CA), and SLC2A1 (GLUT-1) (1:1000, AB1341; Chemicon, Temecula, CA). Analysis was performed with a Leitz Aristoplan microscope (Leica Microsystems, Wetzlar, Germany). Pictures of ccRCC specimens were obtained with a digital camera (KY-070; JVC, Friedberg, Germany). The intensity of the staining for AXL, CAIX, CD10, CD13, E-CADH, and GLUT-1 was classified as follows: 0 no, +1 weak, or +2 strong membranous staining. Endothelial AXL was analyzed to measure the microvessel density of the tumors as follows: 1 = low (0–10), 2 = median (>10–50 cells), and 3 = high (>50 cells). Immunostained TMAs were scanned using the NanoZoomer Digital Slide Scanner (Hamamatsu Photonics K.K., Shizuoka, Japan).

Immunoblot Analysis

Cell and tissue lysates were prepared using RIPA buffer (Sigma, Steinheim, Germany) containing complete protease inhibitor cocktail (Roche). Homogenization was performed with a Tissue lyser (Qiagen) for 4 minutes at 20 Hz. Lysates were centrifuged for 10 minutes (2500g, 4°C). Protein concentration of the supernatant was determined by the BCA Assay (Pierce, Rockford, IL) with bovine serum albumin (BSA) as a standard. The following primary antibodies were used for the functional characterization of CD10: mouse monoclonal anti-CA9 M75 (kindly provided by J. Zavada, Institute of Molecular Genetics; 1:1000), mouse monoclonal anti-CD10 (NCL-CD10-270; Novocastra Laboratories; 1:100), rabbit polyclonal anti-Glut-1 (AB1341; Chemicon; 1:4000), rabbit polyclonal HIF-2α (NB100-122; Novus Biologicals, Littleton, CO; 1:1000), mouse monoclonal anti-β-actin (Chemicon; 1:10,000), and rabbit polyclonal α-pVHL_CT [19] (1:500). After washing with Tris-buffered saline Tween 20, the membranes were incubated at room temperature with HRP-conjugated secondary antibody rabbit anti-mouse (Abcam, Cambridge, United Kingdom; 1:3000) or goat anti-rabbit (MAB1501; Chemicon; 1:3000). The protein-antibody complexes were detected by ECL (Amersham, Piscataway, NJ). Biotinylation of proteins was determined by incubating membranes with streptavidin-HRP (1:5,000). The protein-antibody/biotin-streptavidin complexes were detected by ECL (Amersham).

ELISA

CD10 concentration in human serum samples of 21 RCC and 6 healthy controls was measured by ELISA (R&D Systems). This study was approved by the local ethics commission (EK-1017). The patients' sera were collected according to the standardized protocol of the Department of Urology, University Hospital Zurich. Capture antibody was coated overnight at 4°C in 50 mM Tris-HCl, pH 9.4. Plates were subsequently blocked using 1% BSA in phosphate-buffered saline. Samples were diluted 1:5 in diluent buffer, comprising of 1:3 LowCross Buffer (Candor Bioscience, Eggenwatt, Weissensberg), 0.1% goat serum, and 1% BSA in phosphate-buffered saline for 2 hours at room temperature. The biotinylated detection antibody was diluted in diluent buffer and incubated for 2 hours at room temperature followed by 20 minutes of incubation with streptavidin-HRP and substrate detection. The serum diluent was shown in separate experiments to efficiently inhibit false-positive and false-negative signals. Interassay and intra-assay variations were less than 3% and less than 20%, respectively.

Short Hairpin RNA Constructs and Lentiviral Infections

Expression vectors encoding short hairpin RNA (shRNA) sequences targeting human HIF-2α were purchased from Sigma (shHIF-2α#1, clone ID NM_001430.x-517s1c1; and shHIF-2α#4, clone ID NM_001430.x-1694s1c1, respectively). Viral particles were produced using the ViraPower lentiviral expression system according to the manufacturer's instructions (Invitrogen). Briefly, 786-O cells were infected with lentiviral particles of either shHIF-2α#1 or shHIF-2α#4 constructs, and pools of clones were derived by puromycin selection (4 µg/ml). Pools of clones were analyzed for effective HIF-2α knockdown by quantitative RT-PCR and anti-HIF-2α immunoblot analysis.

Reporter Gene Assay

Four different CD10 promoter fragments were obtained by PCR amplification of human genomic DNA. The oligonucleotide primers used in this study are summarized in Table W9. PCR products were sequenced using the AbiPrism 3100 Genetic analyzer (Applied Biosystems). The reporter constructs were generated by insertion of individual CD10 KpnI-NheI fragments from the human CD10 promoter into the KpnI-NheI sites of a promoterless pGL4.10 firefly luciferase reporter gene plasmid (Promega, Madison, WI). Site-directed mutation of HREI in the CD10 promoter was performed on pGL4.10 containing the whole CD10 promoter using the QuickChange II Site-Directed Mutagenesis Kit (Stratagene, Cedar Creek, TX). For transcriptional transactivation experiments, HEK-293 and HK-2 cells were transfected with 1 µg of reporter plasmids along with 100 ng of pGL4.74 renilla luciferase reporter plasmid (Promega) in 24-well plates using FuGene 6 reagent (Roche). pGl3P2P(607 bp) vectors are 5′-truncated versions of the previously published pGl3b(1454/3172)P2P plasmids containing the wild-type or mutated hypoxia response element (HRE) of the human PHD2 promoter [30]. These vectors served as positive or negative controls for HIFα activity. The empty pGL4.10 vector served as MOCK control. Luciferase activity was measured after cell lysis using the dual-luciferase reporter assay system (Promega). Luciferase activities were determined by dividing the relative luciferase units from hypoxic/DMOG-treated and normoxic/untreated cells. Renilla luciferase control reporters were used for normalization of transfection efficiency.

Chromatin Immunoprecipitation

Chromatin immunoprecipitation (ChIP) was performed according the protocol (version 10.0) of mammalian ChiP-on-chip (Agilent Technologies). Briefly, 5 x 10⁷ to 1 x 10⁸ cells grown in normoxia or hypoxia were fixed in 1% formaldehyde, washed, harvested, and lysed. The cell lysates were sonicated to fragment cellular DNA into fragments of 100 to 600 bp in size and incubated with an anti-HIF-1α (mouse monoclonal, NB100-105; Novus Biologicals) antibody/magnetic bead (Dynabeads Potein G; Invitrogen) mixture overnight at 4°C. After washing to deplete unbound DNA, elution of bounded DNA/antibody complexes from the beads was performed in elution buffer at 65°C. Beads were separated from the eluent by centrifugation (16,000g, 1 minute). Cross-linking of DNA and antibody was reversed by incubating the supernatant at 65°C overnight. Finally, cellular protein and RNA were digested by means of proteinase K and RNase A. DNA was extracted by means of phenolchloroform-isoamyl alcohol in Phase Lock Gel tubes (Eppendorf, NY). DNA concentrations were measured using a nanodrop. SYBR Green Real-time PCR was performed with 10 ng of isolated DNA in a 20-µl reaction using a 7900HT PCR cycler (Applied Biosystems) to quantify the differential binding of HIF-1α to the analyzed HREs of CD10. The primers used for the ChIP experiments are listed in Table W9.

Cell Invasion Assay

Cell invasion was assessed using the QCM 24-well fluorometric cell invasion assay (Millipore, Zug, Switzerland) according to the manufacturer's protocol. Cells were starved in serum-free medium for 24 hours. Afterward, 0.5 x 10⁶ cells/ml were seeded into extracellular matrix-coated inserts and cultured for 72 hours. FCS was used as a chemoattractant in the bottom chamber. The CD10 inhibitor thiorphan (Merck, Darmstadt, Germany) was added to the bottom chamber using the following final concentrations: 1 and 10 µM, respectively.

Statistical and Computational Analyses

Contingency table analysis, Pearson χ² tests, Kaplan-Meier curves, and log-rank tests for evaluating correlations between AXL, CD10, CD13, GLUT-1, and E-cadherin expression and clinical parameters were calculated using StatView (SAS, Institute, Cary, NC). One-tailed unpaired t test with Welsch correction to compare CD10 serum levels in tumor patients with healthy controls was performed using GraphPad Prism 4 (GraphPad Software, San Diego, CA). Automatic classification of the LDA expression data was performed using pocus [31] and R [32] package for assessing the uncertainty in hierarchical cluster analysis by resampling. Average linkage clustering was performed with correlation as a distance measure using 1000 bootstrap replications. The random forest algorithm, implemented as R package, was used for unsupervised classification. It was set to assess the importance of predictors.

Results

Identification of pVHL-Dependent Cell Surface N-glycoproteins in ccRCC

The CSC technology was used as an initial screen to identify potentially pVHL-dependent cell surface N-glycoproteins. In brief, cell surface-exposed glycans were oxidized and tagged with biocytin hydrazide. Then, the cells were lysed, proteins were digested, and the tagged peptides were captured with streptavidin. After thorough washing, N-glycosylated peptides were released with peptide N-glycosidase F, cleaned up with C18 tips, and subjected to liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis. In a first step, we combined SILAC- and CSC-based labeling and isolation of N-glycosylated cell surface N-glycoproteins from ccRCC-derived 786-O cells and stable transfectants reexpressing the long form of wild-type pVHL (referred to as 786-OpVHL₃₀). Subsequent shotgun proteomics analysis of enriched N-glycosylated plasma membrane proteins from 786-O and 786-OpVHL₃₀ cells resulted in the identification of 110 proteins (peptide prophet score ≥ 0.8) (Tables W1–W4 and Figure W1). The 106 proteins, which contained a peptide with at least one consensus N-glycosylation motif (NXS/T), were subjected to further analysis (Table W5). In a first in silico functional classification procedure using the Panther (www.pantherdb.org) and IPA (Ingenuity Systems) classification tools, we uncovered significant enrichments of proteins whose functions have been associated with cell adhesion and cell communication (Figure 1A). A second analysis focused on the involvement of the proteins in specific well-described canonical pathways. The identified proteins are mostly involved in signaling cascades mediated by integrins, cadherins, and Wnt (data not shown).

Figure 1.

In silico characterization of the identified cell surface proteins. (A) Most prominent molecular pathways at which the identified cell surface proteins participate. The functional classification of all identified proteins (n = 106) has been performed by using the PANTHER and KEGG databases. (B) Relative protein expression profiling in ccRCC-derived cells using metabolic labeling of proteins by SILAC and LC-MS/MS-based peptide identification. pVHL expressing or pVHL-negative 786-O cells were grown in medium containing heavy- or light-labeled l-Arg/l-Lys isotopes, respectively. Ratios of the peak intensity were computationally calculated by the ASAP ratio algorithm for every peptide. Shown in the figure is the reconstructed MS1 signal of a pair of light/heavy GLUT-1 peptides detected in pVHL-positive and -negative cells in the ASAP ratio view. The ratios of the same peptide in different charge states are averaged and weighted by the corresponding spectrum intensity to obtain the peptide light-heavy ratio and its error. Up-regulation of GLUT-1 in pVHL (light-labeled) 786-O cells compared with their pVHL-reexpressing transfectants (heavy-labeled) functioned as internal control for the activity of the HIF pathway. (C) Immunohistochemical validation of the expression of GLUT-1 and EGFR on formalin-fixed and paraffin-embedded 786-O cells. pVHL- indicates 786-O cells lacking the pVHL protein. pVHL+ indicates 786-O cells reexpressing pVHL. The top row represents cells stained for GLUT-1. The bottom row represents cells stained for EGFR. Figure 3C was modified from Boysen et al. (Pathologe, 2009;30 Suppl 2:188–192). With permission from Springer Science and Business Media.

Quantification of the pVHL-Dependent Cell Surface N-glycoproteins

SILAC-based labeling of the screened glycoproteins in 786-O and 786-OpVHL₃₀ allowed for the initial relative quantification of 63 cell surface proteins (Table W5). Twenty-three proteins (20.5%) met the criteria of differential expression between pVHL-negative and pVHL₃₀-positive 786-O (ASAP ratio 0.33 > x > 3). Differential expression of the HIF target GLUT-1 served as control for the activity of the HIF pathway and demonstrated the reliability of the measurements (Figure 1B). Immunohistochemical validation on formalin-fixed and paraffin-embedded 786-O and 786-OpVHL30 cells confirmed the quantitative MS/MS data (Figure 1C).

In silico functional classification of the 23 differentially expressed proteins using the Panther database suggested these proteins to participate in cell proliferation (EGFR signaling), cell cycle regulation (p53 signaling), and cell motility (integrin and cadherin signaling) (data not shown).

LDA-Based Validation of RCC-Associated Cell Surface N-glycoproteins in Cell Lines and Primary Tumors

In a second step we combined discovery-driven proteomic screening data with transcriptomic profiling of the preliminary 106 candidates for validation in vitro. The identified cell surface N-glycoproteins were validated in vitro by using a customized LDA platform. HPRT1 mRNA and 18S rRNA were used for normalization of gene expression. Real-time RT-PCR-based amplification of cDNA was detected for 82 candidates (73.2%; Table W7). Of 23 pVHL-dependent proteins, 15 (65%) also showed a pVHL-related transcriptional regulation (Table W7).

To validate the pVHL-dependent transcriptionally regulated candidates from the 786-O cells in human tissue samples, we expanded our targeted LDA analysis to RNA samples from ccRCC (n = 6), pRCC (n = 4), and normal kidney tissue samples (n = 5, pooled). Of the 73 detectable genes, 34 (46.6%) were differentially expressed in ccRCC compared with normal kidney (Table W8). Of 73 detectable genes, 25 (34.3%) were differentially expressed in ccRCC compared with pRCC or normal kidney (Table W8). Thus, we identified significant differences between normal and tumor tissues suggesting the existence of a tumor subtype-specific signature. Random forest [33] and bootstrapping algorithms (pvclust) identified the genes that most accurately distinguished between both RCC subtypes (Figure 2A). The genes THBS1, ENPEP, SLC2A1, BST2, MME (CD10), ABCB1, and SCARB1 were identified, among others, as important discriminators between both renal carcinoma subtypes. To select proteins for subsequent focused tissue-based analysis, we compared the transcriptomic data obtained from cell lines and tissue samples. This resulted in a final list of nine genes that showed concerted deregulation in cell lines and tissues (Table 1).

Figure 2.

Supervised hierarchical clustering on genes analyzed by LDA in the two major subtypes of RCC. (A and B) Molecular subtyping based on gene expression data derived from 10 RCC samples analyzed by LDAs correlated mostly with the histologic features of the tumors. Histologic subtypes are named as follows: pRCC—papillary RCC and ccRCC—clear cell RCC. Genes that differentiate most significantly between the histologic subtypes are listed. (C) TMA-based expression analyses of selected candidates from the initial in vitro and in vivo measurements. Shown are examples of positive and negative tumors for each protein analyzed.

Table 1.

Final List of Candidate Cell Surface Proteins Specific for ccRCC.

Number	Name	IPI	Entrez	Protein 786-O/786-OpVHL30	RNA 786-O/786-OpVHL30	RNA ccRCC/pRCC	RNA ccRCC/Normal
1	AXL	00296992	558	0.27	1.75	1.53	2.44
2	BCAM	00002406	4059	1.21	0.23	1.45	0.23
3	CDH2	00290085	1000	0.88	2.04	0.83	4.88
4	CDH6	00024035	1004	1.27	0.37	0.27	3.40
5	F3	00010338	2152	0.22	11.11	0.58	0.10
6	HEG1	00297263	57493	NA	1.82	2.24	0.57
7	MME	00247063	4311	NA	3.7	9.6	0.8
8	SLC22A4	00171334	6583	1.5	2.2	3.0	13.9
9	SLC2A1	00220194	6513	5.5	2.9	8.5	2.2

Genes deregulated on the protein and mRNA level in 786-O cells and differentially expressed in ccRCC compared with pRCC and normal kidney.

NA indicates not analyzed.

Moreover, to further study the difference between ccRCC and pRCC, we performed hierarchical clustering using multiscale bootstrap resampling (pvclust) of the low-density expression data. This analysis revealed significant tumor subtype-dependent clustering based on differentially expressed cell surface N-glycoproteins (Figure 2B).

TMA-Based Expression Analysis of Selected Human RCC-Specific Protein Candidates

We next performed immunohistochemical expression analysis of three of the nine candidates listed in Table 1 for which specific antibodies were commercially available. We analyzed the expression patterns of AXL, CD10, and GLUT-1 using a TMA that contained 264 ccRCC (Figures 2C and W2). The associations with pathologic parameters are listed in Table 2.

Table 2.

TMA-Based Expression Analysis in ccRCC.

Characteristic	AXL Epithelial			AXL Endothelial			CD10			GLUT-1
	Negative, n (%)	Weak, n (%)	Strong, n (%)	Low, n (%)	Moderate, n (%)	Strong, n (%)	Negative, n (%)	Weak, n (%)	Strong, n (%)	Negative, n (%)	Weak, n (%)	Strong, n (%)
Tumor stage
Total	65	130	23	66	53	99	29	65	151	15	60	150
pT1–pT2	34 (52.3)	62 (47.7)	8 (34.8)	21 (31.8)	23 (43.4)	60 (60.6)	15 (51.7)	32 (49.2)	69 (45.7)	7 (46.7)	33 (55)	64 (42.7)
pT3–pT4	31 (47.7)	68 (52.3)	15 (65.2)	45 (68.2)	30 (56.6)	39 (39.4)	14 (48.3)	33 (50.8)	82 (54.3)	8 (53.3)	27 (45)	86 (57.3)
P		NS			.0011			NS			NS
Differentiation grade
Total	65	130	23	64	53	99	29	65	151	15	60	150
I	24 (36.9)	25 (19.2)	1 (4.3)	5 (7.8)	7 (13.2)	38 (38.4)	3 (10.3)	19 (29.2)	34 (22.5)	1 (6.7)	5 (8.3)	45 (30)
II	29 (44.6)	60 (46.2)	7 (30.4)	23 (35.9)	25 (47.2)	47 (47.5)	9 (31)	27 (41.5)	73 (48.3)	6 (40)	41 (68.3)	50 (33.3)
III	12 (18.5)	45 (34.6)	15 (65.2)	36 (56.2)	21 (39.6)	14 (14.1)	17 (58.6)	19 (29.2)	44 (29.1)	8 (53.3)	14 (23.3)	55 (36.7)
P		.0002			<.0001			.0193			<.0001
Sarcomatoid differentiation
Total	65	129	23	66	52	99	29	65	150	15	60	150
Not sarcomatoid	57 (87.7)	91 (70.5)	8 (34.8)	30 (45.5)	38 (73.1)	88 (88.9)	15 (51.7)	45 (69.2)	116 (77.3)	8 (53.3)	49 (81.7)	102 (68)
Sarcomatoid	8 (12.3)	38 (29.5)	15 (65.2)	36 (54.5)	14 (26.9)	11 (11.1)	14 (48.3)	20 (30.8)	34 (22.7)	7 (46.7)	11 (18.3)	48 (32)
P		<.0001			<.0001			.0158			.0453
Tumor necrosis
Total	65	130	23	66	53	99	29	65	151	15	60	150
No necrosis	46 (70.7)	81 (62.3)	8 (34.8)	20 (30.3)	30 (56.6)	85 (85.9)	9 (31)	41 (63.1)	105 (69.5)	12 (80)	36 (60)	89 (59.3)
Necrosis	19 (29.2)	49 (37.7)	15 (65.2)	46 (69.7)	23 (43.4)	14 (14.1)	20 (69)	24 (36.9)	46 (30.5)	3 (20)	24 (40)	61 (40.7)
P		.0093			<.0001			.0004			NS
Lymphocyte infiltration
Total	65	130	23	66	53	99	29	65	151	15	60	150
Low	16 (24.6)	17 (13.1)	3 (13)	5 (7.6)	5 (9.4)	26 (26.3)	6 (20.7)	18 (27.7)	23 (15.2)	1 (6.7)	5 (8.3)	30 (20)
Moderate	36 (55.4)	68 (52.3	10 (43.5)	26 (39.4)	32 (60.4)	56 (56.6)	12 (41.4)	27 (41.5)	83 (55)	9 (60)	33 (55)	73 (48.7)
Strong	13 (20)	45 (34.6)	10 (43.5)	35 (53)	16 (30.2)	17 (17.2)	11 (37.9)	20 (30.8)	45 (29.8)	5 (33.3)	22 (36.7)	47 (31.3)
P		NS			<.0001			NS			NS

Statistic evaluation of the correlation between pVHL-dependent cell surface proteins (AXL, CD10, and GLUT-1) and clinicopathologic parameters.

NS indicates not significant.

AXL protein expression was observed in both tumor cells and endothelial cells. Only 23 (10.6%) of ccRCCs presented a strong membranous expression in tumor cells. Epithelial AXL expression correlated strongly with poor differentiation grade (P = .0002) and sarcomatoid differentiation of tumor cells (P < .0001).

A high density of AXL-positive endothelial cells correlated significantly with high nuclear differentiation grade (P < .0001). An inverse correlation was found between a high density of AXL-positive endothelial cells and sarcomatoid differentiation (P < .0001). A high endothelial density determined by AXL correlated with a better patient outcome (P = .0028, log-rank test).

CD10 positivity was mainly observed in nonnecrotic ccRCC (P = .0004) and strong GLUT-1 expression was found primarily in tumors with lower nuclear differentiation grade (P < .0001). There were no associations with patient survival.

CD10 Is a HIF Regulatory Target in ccRCC

As CD10 was identified as a pVHL-regulated protein, we asked whether CD10 is also regulated by HIF similar to known HIF targets CAIX and GLUT-1. We first investigated CD10 expression in VHL-negative RCC4 and 786-O cell lines as well as in stable transfectants reexpressing pVHL (RCC4-pVHL₃₀, 786-O-pVHL₃₀). Immunoblot analysis demonstrated CD10 up-regulation in pVHL-negative cell lines compared with pVHL reexpressing cells (Figure 3A). HIF-2α protein levels and the expression of the HIF target gene GLUT-1 correlated with the expression of CD10 (Figure 3A).

Figure 3.

CD10 expression is increased due to pVHL loss-of-function, HIFα stabilization, and hypoxia. (A) Immunoblot analysis of CD10, GLUT-1, HIF-1α, HIF-2α, and pVHL expression in 786-O, 786-O-pVHL₁₉, 786-O-pVHL₃₀, RCC4, and RCC4pVHL₃₀. Actin was used as protein loading control. (B) Immunoblot analysis of CD10, HIF-1α, and HIF-2α expression in RCC4, RCC4-pVHL₃₀, and HK-2 under normoxic (N, 20% O₂) or hypoxic (H, 1% O₂) conditions. (C) CD10 expression after silencing of HIF-2α. Western blot analysis of CD10, HIF-2α, and GLUT-1 expression in 786-O with or with no shRNA against HIF-2α. shRNA expression reduces HIF-2α protein levels followed by a decrease of GLUT-1 and CD10. Laminin A/C was used as protein loading control. (D) mRNA expression levels of HIF-2α and CD10 in 786-O-positive or -negative for shRNA against HIF-2α.

CD10 expression increased under hypoxia (1% oxygen) in HK-2 and RCC4-pVHL₃₀ cells but remained unchanged at high levels under normoxic and hypoxic conditions in pVHL-negative RCC4 (Figure 3B). The increase of CD10 expression correlated with an increase of HIF-1α and HIF-2α in RCC4pVHL30 cells. In HK-2 cells, only an increase in HIF-1α was detectable, suggesting that CD10 is regulated by both HIF-1α and HIF-2α in a cell type-specific manner (Figure 3B). To confirm up-regulation of CD10 by HIF-2α in ccRCC, shRNA against HIF-2α was expressed in 786-O using two different short hairpin constructs (shHIF-2α#1, shHIF-2α#4). The expression of each construct resulted in a significant reduction of HIF-2α in 786-O on the RNA and the protein levels, which was accompanied by a decrease of CD10 mRNA and protein in VHL-negative 786-O cells (Figure 3, C and D). This strongly suggests that CD10 expression is regulated by the pVHL-HIF axis.

Next, a computer-assisted search was performed to investigate a 4-kb genomic DNA fragment of the 5′ region flanking the CD10 gene (EMBL: X79438) for the presence of the HRE RCGTG (R corresponds to A or G) [34–37]. Four potential HREs were identified in this region. The binding of HIF-1α at the identified HREs was investigated by using the ChIP approach (Figure 4A). This analysis could not be performed with HIF-2α because of the lack of antibodies suitable for ChIP analysis. The ChIP experiments were performed in HEK-293 cells cultured in normoxia or hypoxia (1% O₂). Hypoxia induced direct binding of HIF-1α to all four identified HREs in the 5′ region of CD10. The strongest accumulation of HIF-1α under hypoxic conditions was found at HREII. No binding of HIF-1α was measured for the negative control gene KCNJ5. Initial experiments using IgG isotype control antibodies revealed no increased binding after hypoxic treatment.

Figure 4.

HIF-1α binds and activates the CD10 promoter, resulting in increased CD10-dependent tumor cell invasion. (A) Binding of HIF-1α at the CD10 promoter analyzed by ChIP. HREI-IV indicates in silico identified HRE elements in the CD10 promoter; KCNJ5, negative control. Experiments were performed in HEK-293 cells grown in normoxia or hypoxia. Results are shown as ratios of the amplicons detected under hypoxia versus normoxia. (B) Luciferase reporter gene assay. Relative luciferase gene expression differences of a reporter construct containing a 4-kb 5′ region flanking CD10 in HK-2 cells treated with hypoxia or DMOG normalized to untreated cells. A cotransfected renilla luciferase expression vector was used for normalization of transfection efficiency. P2PWT served as positive control for HIF activity. Data are expressed as mean ± SDs. (C) Relative luciferase activity of a reporter construct containing wild-type or mutated HREI of CD10 after hypoxic treatment of HK-2 cells. Values are normalized to HK-2 cultured under normoxia. A cotransfected renilla luciferase expression vector was used for normalization of transfection efficiency. P2PWT served as a positive control for HIF activity. Data are expressed as mean ± SDs. (D) Inhibition of CD10 by thiorphan reduces the invasive potential of 786-O cells. Invasion of 786-O and 786-OpVHL₃₀ cells through a biomatrix was quantified fluorometrically in the absence or presence of the indicated concentrations of thiorphan. Data are expressed as mean ± SDs. RFU indicates relative fluorescence unit.

We then performed luciferase reporter gene assays to confirm whether HIF binding at the identified HREs in the 5′ region of CD10 leads to increased luciferase activity. For this purpose, a DNA fragment containing the 5′ region of CD10 (CD10_{(-5134/-1005)}) was inserted upstream of the luciferase cDNA. After transient transfection, HK-2 cells were treated with or without DMOG. DMOG is a prolyl-4-hydroxylase inhibitor resulting in normoxic HIFα stabilization and increased HIF activity. Compared with untreated cells, the luciferase activity of the CD10_{(-5134/-1005)} reporter gene construct was increased 2.3-fold in HK-2 (Figure 4B). Specificity of DMOG-induced HIF activity was shown by using a reporter construct harboring wild-type HREs of the HIF target gene prolyl-4-hydroxylase domain (PHD) 2 promoter (P2P) [30]. A 3.9-fold induction of the P2P wild-type reporter (P2PWT) in HK-2 indicated the activation of HIF after DMOG treatment.

The effect of HIF stabilization on CD10 activation was also investigated under hypoxic conditions. HK-2 cells transfected with the CD10_{(-5134/-1005)} reporter construct showed a 1.8-fold increase of luciferase activity compared with normoxia. These results confirmed that, in human renal epithelial cells, HIFα stabilization leads to CD10 promoter activation.

We next performed site-directed mutagenesis to confirm the inducibility of the CD10 promoter by HIF through HREI, which showed the strongest induction of luciferase activity (Figure W1). HK-2 cells transfected with a CD10_(5134/-1005) reporter construct containing a mutated HREI (CD10_MUT) were cultured under normoxic or hypoxic conditions (Figure 4C). Mutagenesis of HREI resulted in a two-fold reduction of luciferase activity under hypoxia. There was no induction of CD10_MUT activity under hypoxia compared with normoxia. Hypoxia-induced HIF activity was verified by a six-fold induction of the wild-type P2P reporter construct. This result suggests that HREI is a functional HIF-binding site in the CD10 promoter and that CD10 is a bona fideHIF target gene.

A comprehensive TMA analysis of CD10 expression in ccRCC was performed to confirm the regulatory effect of HIF on CD10 in ccRCC tissue. Three TMAs containing 831 ccRCCs were analyzed for membrane staining of CD10 and the HIF targets CAIX, GLUT-1, and CCND1. CD10 expression correlated significantly with CAIX, GLUT-1 (P < .0001, each), and CCND1 (P = .0005). Because HIF suppresses E-cadherin gene transcription, E-cadherin expression was also analyzed. We found an inverse correlation between E-cadherin-negative and CD10-positive tumors (P < .0001). No correlation was found between CD10, CAIX, or GLUT-1 expression and survival.

CD10 Inhibition Reduces the Invasive Behavior of Renal Carcinoma Cells

The effect of CD10 on the invasive behavior of ccRCC cells was determined using cell invasion assays. For this purpose, 786-O and 786-OpVHL₃₀ were seeded on a reconstituted basement membrane matrix in the presence or absence of the specific CD10 inhibitor thiorphan. The highly invasive fibrosarcoma cell line HT-1080 was used as a positive (with 10% FCS in the lower chamber) or a negative (without FCS in the lower chamber) control. 786-O cells, which express high levels of CD10, showed a 1.6-fold enhanced invasive potential compared with 786-OpVHL₃₀ cells (Figure 4D). A 1.9- and 2.3-fold decrease of cell invasion was measured after addition of the CD10-specific inhibitor thiorphan to 786-O and 786-OpVHL₃₀ cells, respectively. Notably, in the presence of pVHL, 1 µM thiorphan was sufficient to decrease the invasive potential of the tumor cells. In contrast, in VHL-negative 786-O, a 10-fold dose (10 µM) of thiorphan was needed to yield the same effect.

Detection of CD10 in the Serum of RCC Patients

Identification of serum biomarkers for diagnosis, prognosis, or disease progression monitoring of ccRCC represents an ultimate goal in renal cancer research. We analyzed thus the potential of CD10 as ccRCC-associated serum biomarker. CD10 is frequently expressed in tissue, but its potentially elevated abundance in sera from patients with ccRCC is unknown [38–40]. Therefore, sera from ccRCC patients (n = 15), patients with other renal carcinoma subtypes (n = 6), and normal healthy controls (n = 8) were analyzed using a commercial CD10-specific ELISA. The analysis showed that the CD10 serum protein concentration was significantly increased in a subset of ccRCC patients (494.3 ± 125.8 pg/ml) compared with healthy individuals (173.7 ± 43.99 pg/ml, P = .01). Patients diagnosed with other renal carcinoma subtypes without VHL alterations had CD10 levels comparable to healthy controls (117.4 ± 25.32 pg/ml; Figure 5). This result indicates that CD10 may be useful as a potential serum biomarker for the detection of recurrent ccRCC after nephrectomy or preoperative treatment stratification of RCC patients based on the tumor subtype.

Figure 5.

Abundance of CD10 in human RCC samples. CD10 serum levels in ccRCC patients (n = 15), other RCC subtypes (n = 6), and healthy controls (n = 6) measured by ELISA.

Discussion

Reliable and specific cell surface protein markers for the early detection of most cancer types are still lacking. Discovery-driven approaches thus attempt to identify tumor-specific protein markers, which are cell surface accessible for drug targeting, or detectable in body fluids, for example, serum, plasma, or urine, for noninvasive diagnostic analysis. During the last few years, LC-MS/MS-based proteomics became a powerful technology capable of addressing the needs of the clinical community for biomarker discovery on the protein level. Here, CSC-based proteomic technology was used for an initial screen toward the identification of potential cell surface glycoprotein targets in the context of VHL inactivation in ccRCC. The screening data were validated through transcriptomic, immunohistochemical, as well as functional assays.

Using a ccRCC-derived model of pVHL-positive and pVHL-negative cells, we screened for N-glyco-cell surface-exposed proteins. This initial screen revealed ccRCC-expressed cell surface glycoproteins including their specific cotranslational N-glycosylation sites. In contrast to antibody-dependent strategies for cell surface protein identification, we could identify ccRCC glycoproteins through their N-glycosites using the complementary CSC technology for which no commercial antibodies are currently available. These cell surface-exposed glycoproteins and potential diagnostic markers are currently invisible for antibody-based phenotyping strategies. The combination of the CSC technology with relative SILAC-based quantification of the surfaceome in the context of pVHL expression enabled us to detect potential changes in the protein expression pattern within the detected pool of RCC glycoproteins. These relative quantitative changes of glycoproteins in the cell surface location render the cells' capabilities to communicate with the microenvironment, limiting or enhancing signaling capacities. We were also able to further subdivide the cell surface proteome into pVHL-dependent and nonassociated pathways. A clear separation into pVHL-dependent and pVHL-independent pathways may improve the efficiency and decrease unwanted adverse effects of future therapeutic strategies. Our initial proteomic screening results suggested that, besides known pVHL-dependent pathways (e.g., inflammation, MAPK, cadherin signaling), additional signaling pathways such as the endothelin pathway contribute to ccRCC carcinogenesis [41].

Our screen of the surfaceome indicated CD10 and AXL as pVHL-regulated cell surface proteins. The molecular mechanism of the cell surface metalloproteinase CD10 regulation in ccRCC was so far unknown. Previous studies have reported variable expression of CD10 in human sporadic ccRCC [38,39,42]. Here we demonstrate that the activation of CD10 is directly regulated through HIF, which occurs as a consequence of loss of pVHL function in ccRCC. In addition, cell invasion assays implicate CD10 in ccRCC progression potentially linked to increased invasive behavior. Indeed, inhibition of CD10 by the specific inhibitor thiorphan resulted in a reduced invasive behavior of tumor cells. Interestingly, pVHL-expressing tumor cells were more sensitive to lower amounts of thiorphan than pVHL-negative tumor cells. The therapeutic efficiency of CD10 inhibitors may thus depend on the VHL mutation status in ccRCC patients.

Serum assays, enabling early diagnosis of ccRCC patients are currently lacking. Shedding and accumulation of tumor-specific cell surface proteins, such as CD10, in the body fluids exploited for novel ELISA-based strategies would allow earlier diagnosis of primary tumors or metastasis occurring after nephrectomy, resulting in improved prognosis of ccRCC. On the basis of our proteomic results and the hypothesis that proteins such as the cell surface protein CD10 might also be shed in the serum, we established and tested an ELISA to analyze the serum protein levels of CD10. This analysis revealed a significantly higher CD10 serum level in VHL-related ccRCC compared with normal controls and non-VHL-related RCC subtypes, for example, papillary RCC (pRCC). Therefore, CD10 may represent a promising serum marker for ccRCC, which is potentially useful to screen patients after RCC-related nephrectomy for the development of metastases.

The second interesting candidate, which was discovered in our relative quantitative experiments, was the receptor tyrosine kinase AXL. AXL was recently highlighted as a therapeutic target in pancreatic and lung adenocarcinoma [43,44]. Our immunohistochemical analysis demonstrated that AXL is also expressed in ccRCC cells as well as in endothelial cells. There was a significant correlation between high number of AXL-positive endothelial cells and better patient's outcome. This finding correlates with the data of previous studies, which showed an association between highmicrovessel density in ccRCC measured by the endothelial marker CD34 and better patient's outcome [45,46]. In contrast, strong epithelial AXL expression is related to parameters of poor prognosis, for example, high nuclear differentiation grade, presence of necrosis, and sarcomatoid differentiation. This is consistent with the recently reported data confirming that epithelial AXL expression is an indicator of tumor progression in pancreatic cancer, glioblastoma, drug-resistant myeloid leukemia, and non-small cell lung cancer [43,47]. Very recently, AXL mRNA and serum protein levels were shown to correlate with poor patient prognosis in RCC, but the exact cellular protein expression pattern was not studied [47]. This recent identification of AXL the in serum of RCC patients demonstrates that our SILAC CSC strategy was able to identify, in addition to CD10, a second ccRCC-specific member of the N-glyco-cell surface subproteome, which is also shed into bloodstream circulation.

In summary, our proteotranscriptomic screening identified novel pVHL and HIF-dependent components, which may represent the basis for novel molecular targeted therapies and diagnostic strategies in ccRCC. Among a number of interesting candidates, we characterized in-depth CD10 as a HIF target gene and show that its inhibition leads to a less invasive phenotype of ccRCC cells. Furthermore, CD10 and AXL may be promising biomarker candidates for the identification of ccRCC in serum samples. Further analysis of the pVHL-dependent surfaceome may thus help to improve current therapeutic applications toward a more personalized tumor-specific treatment.

Abbreviations

ASAP ratio

automated statistical analysis on protein ratio

ccRCC

clear cell renal cell carcinoma

ChIP

chromatin immunoprecipitation

CSC

cell surface capturing

DMOG

dimethyloxalyl glycine

HIF

hypoxia-inducible factor

HRE

hypoxia response element

LC-MS/MS

liquid chromatography tandem mass spectrometry

LDA

low-density array

MS

mass spectrometry

pRCC

papillary renal cell carcinoma

pVHL

von Hippel-Lindau protein

RCC

renal cell carcinoma

SILAC

stable isotope labeling with amino acids in cell culture

TMA

tissue microarray

VHL

von Hippel-Lindau