Renal cell carcinoma (RCC) accounts for 3% of adult solid tumours and ∼30–40% of patients present with metastatic disease which has a poor prognosis, with a 5-year survival of <10%. Identification of novel therapeutic targets or biomarkers for prognostic, diagnostic or predictive use remains a priority. Cell surface proteins are involved in a number of vital cellular processes which are altered in tumourigenesis and constitute ideal targets for small molecule or antibody based therapies. Soluble shed forms may also act as circulating biomarkers, making this a subcellular compartment of particular interest in proteomic-based biomarker identification studies.

Investigation of the most common form of inherited RCC led to identification of the von Hippel-Lindau (VHL) tumour suppressor gene (1) and it is now clear that loss of VHL function also occurs in a large proportion of sporadic RCCs of the conventional (clear cell) subtype (2). VHL has been implicated in numerous biological processes and has a well established role in regulation of the transcription factor hypoxia-inducible factor (HIF), acting as the substrate recognition component of an E3 ubiquitin ligase complex that targets HIF-α subunits for polyubiquitination and proteasomal degradation in an oxygen-dependent manner. In cells exposed to hypoxia or lacking functional VHL, HIF-α is stabilised, resulting in a number of gene expression changes including the upregulation of vascular endothelial growth factor (VEGF), platelet-derived growth factor (PDGF) and carbonic anhydrase IX (CAIX). It is clear that although the HIF pathway is central to VHL function and tumourigenesis, HIF-independent functions, some of which involve other substrates of VHL ubiquitin ligase, contribute to its role (3,4).

Targeting cell surface proteins that are downstream of VHL is already being exploited, as illustrated by the receptor tyrosine kinase inhibitors sunitinib and sorafenib (5). Similarly CAIX, one of the most consistently upregulated proteins in conventional RCC, has been investigated in many studies including assessment of soluble forms in serum and urine (6). The plasma membrane was therefore chosen as a subcellular fraction to focus on for biomarker identification. During optimisation of a plasma membrane protein enrichment strategy based on cell surface biotinylation and avidin affinity chromatography for a comparative study (7) using the VHL-defective UMRC2- renal cancer cell line as a model system, purified proteins were catalogued using 1D PAGE followed by in-gel tryptic digestion and LC-MS/MS (GeLC-MS/MS). This identified several plasma membrane proteins previously associated with RCC and a number of proteins of interest as potential biomarkers due to their dysregulation in other cancers and/or their known cellular functions. These included β-dystroglycan, the transmembrane subunit of the dystroglycan 1 protein, which was selected for further study and shown to exhibit a VHL-dependent change in glycoform in UMRC2 cells. Using an oligonucelotide array targeting genes involved in glycosylation, changes in several key enzymes were found in UMRC2−/+VHL cells supporting a role for VHL-mediated changes in glycosylation in tumourigenesis. Altered expression of bifunctional UDP-N-acetylglucosamine 2-epimerase/N-acetylmannosamine kinase (GNE) was confirmed in UMRC2−/+VHL cells and also found to occur in conventional RCC.

A method for enrichment of plasma membrane proteins using cell surface biotinylation and avidin affinity chromatography was optimised using the VHL-defective renal cancer cell line UMRC2-. Western blot analysis (Fig. 1) showed significant enrichment of the plasma membrane proteins Na/K-ATPase α1 and GLUT-1 compared to the unbound fraction or a whole cell lysate. The abundant cytosolic proteins HSP70 and β-actin were almost undetectable as were proteins specific to endoplasmic reticulum (GRP94 and calnexin) and the nucleus (lamin A/C). Low levels of golgin-84 were found whilst the mitochondrial protein NADH-ubiquinol oxidoreductase 39 kDa was present at more significant levels, indicating that some contamination may be present.

Using a GeLC-MS/MS approach, a total of 3,991 peptides were identified corresponding to 814 unique proteins with at least two significant peptides (the complete date set is available at www.proteomics.leeds.ac.uk), of which 183 (22%) were known plasma membrane proteins (a selection of these are shown in Table II); this represents a significant enrichment compared with 5% of the 956 proteins identified in a parallel analysis of a whole cell lysate. The identified proteins included several of interest in the context of VHL/RCC including integrin α3, transferrin receptor 1, epidermal growth factor receptor (EGFR) and CAIX.

One protein selected for further study was the DAG1 gene product dystroglycan-1, a protein that has been previously implicated in carcinogenesis (17). Dystroglycan-1 comprises two subunits, α and β, which are generated by proteolytic cleavage of the α/β precursor polypeptide. The peptides identified in this study (corresponding to amino acids 702–714, 783–793 and 795–823) were all from β-dystroglycan, a 43-kDa type I transmembrane protein consisting of amino acids 654–895 of the molecule (18) that anchors the extracellular α-subunit to the cell surface.

Comparative analysis of UMRC2−/+VHL whole cell lysates by Western blotting using an antibody raised against amino acids 655–767 (clone 56) towards the N-terminus of the β-dystroglycan molecule that recognises the full length protein (43 kDa) but not the fragment reported to migrate at ∼31 kDa (19,20) showed no difference in expression level but a slightly increased electrophoretic mobility (corresponding to a difference of <5 kDa) in UMRC2+VHL cells (Fig. 2A). This was confirmed using an alternative anti-β-dystroglycan antibody (clone 43DAG/8D5) recognising a C-terminal epitope. This alteration was not seen in two other renal cancer cell lines (RCC4 and 786-0 −/+ VHL) but a similar change was seen in vivo in 12/15 matched normal and RCC tissue samples (Fig. 2B). The form of β-dystroglycan seen in tumour tissue was found to co-migrate with that in UMRC2− cells but the form in normal renal tissue migrated more slowly than that in UMRC2+VHL cells, thus the overall difference was smaller in magnitude (Fig. 2C).

Removal of N-glycans with PNGase F resulted in a shift in molecular weight of both β-dystroglycan isoforms with the difference in size between the UMRC2− and +VHL cells still apparent (Fig. 3A). Using a combination of exoglycosidases together with PNGase F to remove both N- and O-linked glycans eliminated the difference in size seen between UMRC2- and +VHL cells (Fig. 3B) strongly supporting differential glycosylation as being the cause of the difference, with O-linked glycans contributing at least part of the change. For deglycosylation experiments, GLUT-1 was monitored as an internal control and bovine fetuin as an external control (for an example see Fig. 3C). When α2–3,6,8,9-neuraminidase, which specifically removes all non-reducing terminal branched and unbranched sialic acid residues, was omitted from the deglycosylation reaction, the difference in size of β-dystroglycan between UMRC2− and +VHL cells was still apparent (Fig. 4). Conversely, if deglycosylation was carried out using only α2–3,6,8,9-neuraminidase, the difference in size was eliminated (data not shown). Taken together these data indicate that the difference between β-dystroglycan isoforms is due predominantly to a change in the level of sialylation. The presence of protease inhibitors in the extracts used for the experiment, together with the absence of a change in mobility in the mock reactions carried out without enzymes strongly suggest that proteolytic activity did not contribute to the reactions (additional bands were present in both the mock reactions and the deglycosylated samples in the example shown in Fig. 3B, but these were of significantly lower intensity than the major forms and not present in every reaction as illustrated in Fig. 4). The smaller magnitude of change in tissue made resolution of forms of β-dystroglycan more difficult thus confident interpretation of deglycosylation experiments with tissues was not possible.

Examination of the expression profiles of genes involved in glycosylation using a glycosylation focussed microarray showed changes in several molecules (the complete date set is available at www.proteomics.leeds.ac.uk), several of which related to possible alterations in sialylation (Table III). These included GNE, which encodes bifunctional UDP-N-acetylglucosamine 2-epimerase/N-acetylmannosamine kinase, which was upregulated in UMRC2− cells and NPL, which encodes a sialic acid lyase, which was downregulated. As a positive control, changes in other gene classes on the chip which were found in UMRC2 cells included upregulation of TGF-β and downregulation of clusterin in -VHL cells which are both known VHL-associated changes (21,22).

Expression of GNE, the gene of interest which showed the greatest magnitude of change at the mRNA level between UMRC2−/+VHL cells, was investigated further. Western blot analysis showed that GNE protein levels changed in UMRC2−/+VHL cells in a manner that paralleled the changes seen at the mRNA level (Fig. 5A). Furthermore, in matched normal kidney cortex and conventional RCC tissues, GNE was found to be upregulated in 6/8 tumours (Fig. 5B). In normal kidney tissue a dominant lower molecular weight band was also seen; unlike the higher molecular weight band, this was not recognised by an alternative antibody to GNE raised against a different epitope and it remains to be determined whether this represents a smaller form of GNE or a cross reacting protein species.

The tumour suppressor gene VHL plays a central role in development of conventional RCC and the characterisation of VHL-regulated proteins and pathways offers promise in identifying new biomarkers and therapeutic targets. In RCC both global mRNA expression profiling of VHL cell line pairs (23–25) and complementary proteomic approaches (7,8,22,26) have succeeded in identifying changes that are relevant in tumourigenesis.

Previous analysis of the membrane proteome of renal cancer cells fractionated from a post-nuclear supernatant using a 60% sucrose cushion followed by a 15–60% sucrose gradient identified high expression of CD70 and showed that this protein could act as a target in antibody-targeted cytotoxic therapy (27). A further study used cell surface capturing (CSC) technology, where glycans are labelled with biocytin hydrazide and following digestion, labelled N-glycopeptides purified by avidin affinity chromatography, combined with stable isotope labelling with amino acids in cell culture (SILAC) to compare the cell surface of −/+VHL cells (26). In the present study, cell surface biotinylation and avidin affinity chromatography was used to enrich plasma membrane proteins as part of method development for a quantitative comparative proteomic study. The presence of NADH ubiquinol oxidoreductase 39 kDa in the enriched fraction suggested some contamination from other organelles, but this protein is a subunit of a respiratory chain complex demonstrated to localise to the plasma membrane (28). Similarly, a number of proteomic studies have reported intracellular proteins on the cell surface, such as cytoplasmic and ER lumenal chaperones (29–31). The enrichment strategy described here was used in combination with SILAC in a study which identified upregulation of the adhesion molecules CD166 and CD147 in VHL defective cells and some RCC tissues (7).

Whilst examining the enriched fraction to assess the extent of profiling of proteins of relevance in RCC, β-dystroglycan was selected for further analysis due to the known involvement of dystroglycan in cancer. Dystroglycan is a transmembrane glycoprotein encoded by the DAG1 gene that is processed into two subunits − the transmembrane β domain and the extracellular α domain. Loss of α-dystroglycan expression and correlations with prognosis have been reported in a number of tumour types (32–36). In RCC, loss of α-dystroglycan correlated with high grade disease and was an independent predictor of shorter disease-free and overall survival (37). Combined loss of α-dystroglycan and p27^kip1 defined a group of patients with particularly poor outcome (38). Many studies analysing α-dystroglycan used antibodies recognising glycosylation-dependent epitopes and changes in glycosylation have been suggested to account for loss of α-dystroglycan staining (39,40). Changes in expression of both LARGE and β3-N-acetylglucosaminyltransferase-1 restored glycosylation of α-dystroglycan and altered tumour cell behaviour (41,42).

Changes in expression of β-dystroglycan in cancer are less consistent with some studies finding no change in expression. Loss of β-dystroglycan was found in some cancers including prostate, breast, colon and oesophageal (32,43–45) and relationships with progression were reported for breast and colon cancers (32). In oral SCC loss of β-dystroglycan was reported in poorly differentiated tumours (19) whilst in a separate study the presence of the 31 kDa β-dystroglycan fragment correlated with lymph node metastasis and tumour differentiation (46).

Characterisation of β-dystroglycan showed that its form changed in UMRC2 cells in a VHL-dependent manner. No evidence was found to suggest that this was due to changes in phosphorylation or alternative splicing (using dephosphorylation with lambda protein phosphatase and by RT-PCR respectively, data not shown). However, deglycosylation experiments indicated that this change was due, at least in part, to differential sialylation. A similar but smaller change was also seen in the majority of RCC samples compared with matched normal kidney cortex. Previous studies in RCC did not report this change in glycosylation in β-dystroglycan (38) which may be due to the gel systems used as resolution of the forms, which differ by <5 kDa, is difficult, especially in tissues.

As mentioned above, a truncated ∼31-kDa fragment of β-dystroglycan lacking the extracellular domain has been identified in cell lines and tissues (20,47) with processing by matrix metalloproteinases being implicated in its formation (46–48); an MMP-9 cleavage site has recently been defined (49). Tyrosine phosphorylated forms have also been described (50,51). In a study examining the role of dystroglycan in prostate cancer cell lines, β-dystroglycan was shown to exhibit reversible cell density dependent changes in form, with lower molecular weight forms of 38–43 kDa due to mis-glycosylation and bands at 31 and 26 kDa resulting from proteolysis being seen in supra-confluent cells (52). The altered glycosylation in our study seems to be distinct from this effect. Treatment with PNGase F resulted in a similar increase in gel mobility of β-dystroglycan in UMRC2− and +VHL cells thus N-glycosylation was present irrespective of VHL status. The cells used in our study were all harvested at the same growth state (that is, approaching confluence), but it is possible that VHL alters the point at which a density-dependent change in form is induced.

Possible mechanisms underlying the VHL-dependent changes in sialylation were investigated using a glycoarray. Overall, the glycoarray results for the three −/+VHL cell line pairs analysed did not show obvious patterns, which may reflect the complexity of glycosylation, the potential for differences between differing genetic backgrounds and the limitations of cell line models. Altered expression of glycogenes involved in sialylation that were seen in UMRC2 cells, like the change in form of β-dystroglycan, were restricted to this cell line, with the exception of NPL, expression of which was found to be VHL-dependent in all three cell lines. Previous studies have also found differences in VHL-dependent gene expression in different cell lines and this is apparent here not just for genes involved in glycosylation but also known VHL-regulated genes present on the array, with no one cell line pair behaving as an outlier.

In UMRC2−/+VHL cells, lower expression of NEU1 which encodes sialidase-1, which localises to the lysosome and the cell surface and increased expression of the sialyltransferases ST3GAL6 and ST6GAL1, in UMRC2− cells may contribute to altered sialylation. Similarly, upregulation of GNE, which is a key enzyme in sialic acid biosynthesis, together with downregulation of NPL, which is involved in sialic acid turnover, in UMRC2− cells may alter the availability of sialic acid and thereby affect sialylation. Indeed there is mounting evidence in the literature that substrate availability and levels of GNE do impact on sialylation. In an analysis of N-linked sialoglycopeptides in human pancreatic carcinoma (SW1990) cells, increased metabolic flux through the sialic acid pathway by exogenously supplied substrate was found to selectively increase the sialylation of individual glycoproteins (53). In hematopoietic cell lines, GNE was found to be an important regulator of cell surface sialylation (54) and knockdown of GNE in HEK293 cells reduced total cell surface sialic acid content (55). Expression of GNE in UMRC2−/+VHL cells validated the changes found using the microarray, with reduced expression of GNE being seen in VHL transfectants and GNE levels were also upregulated in a significant proportion of tumours compared to normal renal tissue. Changes in tumour tissue at the mRNA level reported in microarray data sets correlate with this result (56,57).

A similar finding of involvement of a tumour suppressor gene in glycosylation has been previously reported, with altered expression of glycosyltransferases and decreased sialylation of N- and O-glycans being seen in Capan-1 pancreatic carcinoma cells following transfection with p16^INK4a (58). In an extension of this study, decreased levels of GNE were found to be an important consequence of p16^INK4a transfection, correlating with loss of membrane bound sialic acid and hyposialylation of α5 and β1 integrin (56). The changes in these integrin subunits were very similar to the change seen in β-dystroglycan.

The results described here raise the question of the functional impact of changes in glycosylation of β-dystroglycan in renal cancer and there is clearly enormous potential for glycosylation-dependent biomarkers (59). Alterations in glycosylation associated with cancer progression are well documented and include changes in sialylation (reviewed in ref. 60). Changes in glycosylation in RCC have been shown using lectin staining (61) and we have previously described VHL-dependent changes in glycosylation of CD166 (7). The extent to which glycosylation of other proteins can also be regulated by VHL remains to be determined. Further work is now required to build on the preliminary findings presented here and establish the extent of changes in glycosylation in RCC that may be attributable to VHL and their functional consequences.