VHL Gene Mutations and Their Effects on Hypoxia Inducible Factor HIFα: Identification of Potential Driver and Passenger Mutations Free

Renal cell carcinoma (RCC) is the most frequent malignant tumor arising from the kidney with approximately 210,000 new cases diagnosed per year worldwide (1). About 80% of the RCC belong to the clear cell RCC (ccRCC) which is commonly characterized by loss of 1 short arm of chromosome 3 and mutation of the von Hippel-Lindau (VHL) tumor suppressor gene on the second short arm of chromosome 3 at position 3p25. The high rate of VHL mutations suggests that the inactivation of the multiadaptor VHL protein (pVHL) plays a critical part in ccRCC initiation (2).

The best investigated function of pVHL is its role as a substrate recognition component of an E3 ubiquitin protein ligase complex (3). Under normoxic conditions, pVHL binds the hydroxylated hypoxia inducible factor (HIF)α subunits which leads to their ubiquitination and degradation (4). Under hypoxic conditions or absence of pVHL, the stabilization of HIFα leads to the formation of a heterodimer with HIF1β which initiates enhanced transcription of HIF target genes. Both HIF1α and HIF2α show common but also distinct transcription patterns. HIF1α preferably drives the expression of genes important for apoptotic and glycolytic pathways, whereas HIF2α activates genes involved in cell proliferation and angiogenesis (5–7). Based on these results, it was suggested that HIF2α is more oncogenic than HIF1α. This finding was supported by the observation that silencing of HIF2α in a human VHL-negative RCC cell line was sufficient to prevent tumor formation in mice and that HIF2α promotes c-myc activity (8–11). However, recent evidence suggests that HIF1α is responsible for genomic instability which may favor the accumulation of additional genetic hits leading to carcinogenesis (12).

Over 800 VHL mutations were identified in both hereditary and sporadic ccRCC (13). More than 50% of these mutations are frameshift and nonsense mutations which are highly likely to cause loss of pVHL function (14, 15). Because of the large number of missense mutations distributed over the 3 exons of VHL, the consequences of such alterations on pVHL's integrity and HIFα stabilization are difficult to predict.

Several studies have been carried out to classify VHL mutations identified in the hereditary VHL syndrome (reviewed in ref. 16). Nonsense and frameshift mutations generating VHL-null alleles are associated with ccRCC (type 1 VHL disease), whereas type 2 VHL disease is mainly characterized by missense mutations. This type is further subdivided into type 2A (with low risk of ccRCC), type 2B (with high risk of ccRCC), and type 2C which predisposes for pheochromocytoma. Missense mutations affecting pVHL's surface result in a higher risk for developing pheochromocytoma compared with substitutions altering the protein core (17). Forman and colleagues used bioinformatic tools to determine the thermodynamic change (ddG) of missense mutations and linked destabilizing mutations in the interface of HIFα and elongin B to a prevalence of ccRCC, whereas mutations interfering with the elongin C interface resulted in increased risk of pheochromocytoma (18).

Controversial data exist about the prognostic and predictive value of the VHL mutation type in sporadic ccRCC. Some groups found a correlation between “loss-of-function” mutations (nonsense and frameshift) and a worse prognosis for patients or higher HIF target-directed response rates, whereas other groups were not able to confirm these results (for review see, ref. 16). A large study at the Dana-Faber Cancer Institute revealed an increased response in ccRCC patients with “loss-of-function” mutations (nonsense, frameshift, and in frame) treated with antiangiogenic therapies (sunitinib, sorefanib, axitinib, or bevacizumab), blocking some of the downstream effects of pVHL (19). In contrast to nonsense, frameshift, and in frame mutations, the impact of missense mutations on the function of pVHL seems to be highly diversified ranging from imperceptable to complete functional loss (20–22).

The goal of our study was to functionally characterize missense mutations in sporadic ccRCC on pVHL and HIF using a combination of in silico and in vitro assays.

Two hundred and fifty-six formalin-fixed paraffin-embedded ccRCC samples were histologically reviewed by 1 pathologist (H.M.). This study was approved by the local commission of ethics (ref. number StV 38-2005). Survival time was obtained for 123 patients (48%). The mean age of patients was 64 (31–88), and the mean follow-up of patients was 47 months (0–139). Tumors were graded according to tumor stage (pT) and the Fuhrman grading system and histologically classified according to the World Health Organization classification (23). The histologicl data are listed in Supplementary Table S2.

For DNA extraction, 3 tissue cylinders (diameter 0.6 mm) were punched from each paraffin block. DNA was extracted according to the QIAGEN EZ1 DNA Tissue protocol for automated purification of DNA from tissue (Qiagen). PCR was carried out as previously described (24) with slight modifications. Only 1 PCR step with 40 cycles was carried out by using unlabelled primers. As mutations are rare in the 5′ region of exon 1, the first 162 coding basepairs of VHL were excluded from sequence analysis (25). DNA sequencing was carried out by using the BigDye Terminator v1.1 Cycle Sequencing Kit (Applied Biosystems). The obtained sequences were compared with the National Center for Biotechnology Information (NCBI) sequence AF010238 by using NCBI's Blast 2 Sequences. VHL point mutations were validated by a second separate PCR and sequencing analysis.

Two crystal structures of pVHL in complex with elongin B, elongin C, and the HIF peptides are available (26, 27) and stored in the Piccolo database of protein interaction (PDB codes 1lm8 and 1lqb; ref. 28). As input for in silico analyses pVHL was separated from the VHL/elongin C/elongin B complex stored in the piccolo database 1LM8.pdb. Swiss PDB viewer “Deep View v.4.0” (http://spdbv.vital-it.ch) was used to locate the missense mutation either to the surface of pVHL or to its core. A threshold for amino acids (aa) with a solvent accessibility of more than 9% was defined as “surface” (29). To predict the association of disease because of missense mutations, the program Site Directed Mutator (SDM) was used which was developed by the Crystallography and Bioinformatic group from the University of Cambridge (30, 31). SDM calculates the ddG of the protein after 1 missense mutation (32, 33). A ddG of more than 2 is set by the algorithm as a cutoff for a disease-associated mutation. The program Crescendo, which predicts the most essential regions and aa for protein–protein interactions because of conserved regions was used to analyze the impact of missense mutations on protein–protein interactions (34).

Human RCC4 cells negative for pVHL, human telomerase reverse transcriptase-immortalized retinal pigment epithelial cell line (hTERT RPE-1) positive for pVHL19 (kindly provided by Wilhelm Krek, Institute of Cell Biology, ETH Zurich, Switzerland), and mouse embryonic fibroblasts (MEFs) negative for pVHL and p53 (kindly provided by Ian Frew, Institute of Physiology, University of Zurich, Switzerland) were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (FCS) in a humified incubator with 5% CO₂ at 37°C. RCC4 and hTERT RPE-1 cell lines were authenticated by short tandem repeat profiling on March 15, 2011 and April 26, 2011, respectively, which was carried out by Identicell (Department of Molecular Medicine, Aarhus University Hospital Skejby, Aarhus, Denmark). The MEF VHL null/p53 null cell line was generated after SV40 infection for transformation, adenoviral cre recombinase infection for pVHL and p53 knockout, and single cell cloning. MEFs were not authenticated. MEFs were transfected with 1 μg vector by using Fugene6 (F. Hoffmann-La Roche AG) for transient and stable transfection. Stable polyclonal transfectants were generated by selection with 0.5 mg/mL geniticin/G418 (Invitrogen AG) over 1 month. RCC4 and hTERT RPE-1 cells were transduced in the presence of polybrene (4 μg/mL) with viral supernatant produced by 293Amphopack cells (Clontech-Takara Bio Europe) transfected with VHL wild type (WT) or mutants and selected with 4 μg/mL puromycin (Sigma–Aldrich) over 1 month. Dimethyloxalyl glycine (DMOG; Cayman chemical) treatment of generated MEF cell lines was carried out over 24 hours at 1 mmol/L. DMOG experiments were independently repeated twice.

VHL mutants were generated by using the one-step PCR mutagenesis Kit (Agilent Technologies Inc.) and pcDNA3.1-VHL including a hemagglutinin (HA) as template. The HIF reporter construct pcDNA3.1-HIF1α-GFP (HIF1α Gene ID: 15251) was kindly provided by Stefanie Lehmann (Institute of Biomedical Engineering, ETH Zurich, Switzerland). pcDNA3.1-HIF2α-GFP (HIF2α Gene ID: 2034) was obtained by PCR amplification of pcDNA3.1-HIF2α and cloned into SacII and NotI restriction sites of the pcDNA3.1-GFP vector. All constructs were sequence verified.

RCC4 and hTERT RPE-1 cells stably expressing the control vector, WT VHL or VHL mutants were grown to 80% to 90% confluency in a 6-well format and then treated with 40 μg/mL cycloheximide (Invitrogen) for different time points (0, 1.5, 3, and 6 hours). Zero hours was taken as reference input for the Western blot analysis. After incubation, cells were harvested, processed, and analyzed by Western blot. Cycloheximide experiments were carried out once for RCC4 WT, RCC4 Ser68Thr, and RCC4 Leu101Pro. Experiments for hTERT RPE-1 VHL WT were carried out 3 times independently and once for hTERT RPE-1 Ser68Thr and hTERT RPE-1 Leu101Pro.

Cells were lysed in radioimmunoprecipitation assay buffer supplemented with protease inhibitors (F. Hoffmann-La Roche AG) for 30 minutes on ice. The protein concentration was determined by using the BCA Protein Assay Kit (Thermo Fisher Scientific Inc.). Equal amounts of protein were loaded on a 4% to 12% gradient Bis-Tris gel (Invitrogen) and transferred onto a nitrocellulose membrane (Whatman). Membranes were blocked for 1 hours in 5% milk powder and then incubated with mouse anti-HA (1:1,000 dilution; Invitrogen), rabbit anti-human VHL CT (1:1,000 dilution; kindly provided by Wilhelm Krek), or mouse anti–β-actin (1:1,000 dilution; Chemicon) overnight at 4°C. Subsequently, membranes were incubated with the appropriate secondary antibody for 1 hour at room temperature. Proteins were detected by using the ECL Kit (Thermo Fisher Scientific Inc.).

Cells were transiently transfected with HIF reporters and harvested after 24 hours with trypsin and washed in PBS for the fluorescence-activated cell sorting (FACS) analysis. The cells were then resuspended in FACS buffer (PBS, 2% FCS) and dispersed through a filter into a FACS tube (BD Bioscience). Cells were kept on ice in the dark until analysis with the FACS Calibur (BD Bioscience). The flow rate was adjusted to 200 to 300 events per second and 10,000 events were measured in total per sample. Data analysis was conducted by using FlowJo (Tree Star, Inc.). All experiments were conducted in triplicate with the exception of the DMOG treatment where 1 experiment was done in a single well transfection format and a second one in duplicate. The error bars in the figures are ± SD.

Contingency table analysis, χ² correlation tests, Kaplan–Meier curves, and log-rank tests were calculated by using SPSS/PASW 18.0.0. GraphPad Prism 5 was used for Spearman correlation tests and 1-way ANOVA with Bonferroni's Multiple Comparison tests. P < 0.05 was considered significant.

Sequence analysis of VHL in 256 ccRCC cases resulted in 75 (29%) WT and 181 (71%) mutated tumors (Fig. 1). Six tumors had 2 mutations. There were 88 (47%) point mutations, including 1 silent mutation, 13 (7%) splice mutations, and 86 (46%) deletions/insertions. Point mutations were further subdivided into 22 (25%) nonsense and 65 (75%) missense mutations. The deletions/insertions were split into 74 (86%) frameshift and 12 (14%) in frame mutations. All mutations are listed in Supplementary Table S1.

The distributions of nuclear differentiation grade (Fuhrman), tumor stage (pT), and VHL mutation type are summarized in Supplementary Table S2. Survival analysis revealed no significant differences neither between WT and mutated VHL nor between the different mutation types (Fig. 1B). The same was true when the mutations and the mutation types were subgrouped according to their exon locations.

The distribution and number of the identified mutations and mutation types within the 3 exons of VHL are shown in Figure 2A–J. The number of VHL mutations was significantly increased in exon 1 compared with exons 2 and 3. The most frequent mutated amino acids (aa) were found at positions 65 (whole dataset and nonsense, respectively; Fig. 2A and B), 135 (frameshift, Fig. 2C), 78 (missense, Fig. 2D), and 76 (in frame, Fig. 2E). A clustering of mutations was detected from aa 65–77 (nonsense), 61–64/131–137 (frameshift), 86–89/111–119 (missense), and 66–76 (in frame).

In a first step, we characterized the VHL missense mutations identified in our set of sporadic ccRCC by using the Swiss PDB viewer “Deep View v.4.0.” This program predicts whether the locations of missense mutations are pVHL surface or core related. With a threshold of accessibility of the aa of more than 9% (defined as surface; ref. 29), we obtained the most surface mutations in exon 3 (n = 14), whereas the majority of the missense mutations in exons 1 (n = 25) and 2 (n = 16) were assigned to the core (Fig. 3A). Calculating the surface to core ratio in pVHL (aa 53–213) resulted in similar values for exons 1 (1.9) and 2 (1.7) but a higher value for exon 3 (5.6). This suggests an increased probability of having surface mutations in exon 3. Because of the crystallographic structure of the pVHL complex (pdb code 1lm8), 21 of 23 missense mutations located on the surface of pVHL could be assigned to interact with either HIF, elongin C, or elongin B (Supplementary Table S1, Fig. 3C; ref. 18). The remaining 2 missense mutations were not located in the domain of any known interaction partner. By using the software Crescendo which predicts the most essential regions and aa for protein–protein interactions, most of the missense mutations were also allocated to these conserved regions. Other binding partners (reviewed in ref. 35) listed in Supplementary Table S1 were found to be influenced by the surface missense mutations (Trp88Cys: aPKC, Tyr98Asn: aPKC/CollIV/CARD9/SP1/KIF3A; Tyr112Asp: aPKC/CollIV/CARD9/SP1/KIF3A/CCT; Thr124Ala: CCT/aPKC/TD-NEM; Gln132Pro: CCT, Leu158Val: p53, Lys159Asn: p53, Arg161Pro: p53, Arg161Gln: p53).

To predict changes of the pVHL structure because of missense mutations, we used the program SDM. The algorithm calculates the ddG caused by a mutation and output values with a ddG more than 2 are defined as “disease-associated.” By calculating the ddG values for all missense mutations, we found 48% predicting a significant impact on the protein structure (ddG > 2) and the remaining 52% having “neutral” mutations (ddG < 2). A significant increased number of missense mutations changing the structure of pVHL was detected in exon 1 followed by exons 2 and 3 (Fig. 3D). When we split surface- and core-related mutations according to their exon locations, we observed “disease-associated” mutations primarily in the core of exons 1 and 2 and on the surface of exon 3 (Fig. 3F).

To verify the predictive power of SDM on the ddG and stability of pVHL, we selected 6 missense mutations of which 4 were assigned disease-associated by protein destabilization (Ser68Thr, Gly93Glu, Leu101Pro, and Tyr112Asp), 1 was assigned disease-associated by protein stabilization (Pro86His), and 1 was neutral (Ser68Thr). Four of these missense mutations (Ser68Thr, Gly93Glu, Leu101Pro, and Tyr112Asp) were stably transduced into the RCC4 tumor cell line and 6 (Ser68Thr, Gly93Glu, Leu101Pro, Tyr112Asp, Pro86His, and Trp117Arg) into the hTERT RPE-1. As expected, transduction of the negative control mutation Leu63fsX67 showed no expression in RCC4 and hTERT RPE-1 cells (Fig. 4A and B). In both cell lines, a pVHL expression pattern was observed after transduction of WT pVHL and the neutral pVHL-mutant Ser68Thr as previously described (36). In contrast, the pVHL mutants Gly93Glu, Trp117Arg, and Leu101Pro, which had the highest destabilizing ddG values of 2.9, 4.3, and 6.1, respectively, showed much lower expression levels with bands between 20 and 30 kDa and additional abundant degradation products below 20 kDa which were not detected with the pVHL WT and the mutation Ser68Thr. The lower expression of the bands in these pVHL mutants were not because of decreased mRNA expression as verified by quantitative reverse transcriptase-PCR analysis (data not shown). The pVHL-mutant Tyr112Asp (ddG = 2.6) was only minimally affected by the mutation underscored by the detection of a faint additional band below 20 kDa compared with pVHL WT. Interestingly, the pVHL-mutant Pro86His (ddG = 2.3) expressed in the hTERT RPE-1 cell line showed an increased band intensity of full-length pVHL (30 kDa) compared with the additional bands between 20 and 30 kDa present in the WT protein which reflects the predicted stabilizing effect of the mutation by SDM.

To further investigate the stability of mutant proteins, a cycloheximide assay was carried out with 2 selected missense mutations. pVHL WT and pVHL-mutant Ser68Thr exhibited similar band intensities after 1.5, 3, and 6 hours cycloheximide treatment, whereas the pVHL-mutant Leu101Pro was clearly less stable (Fig. 4C and D). The bands between 20 and 30 kDa as well as the degradation product at 15 kDa were significantly decreased in intensity already after 1.5 hours in the cell line hTERT RPE-1.

To address the question whether the prediction of the stability of pVHL mutants influences its functionality, we set up a screening platform based on MEFs negative for pVHL and p53 (MEF^−/−). First, we transiently transfected hTERT RPE-1 and MEF^−/− with an N terminally HA-tagged VHL expressed from the pcDNA3.1 vector by using Fugene6 with a vector DNA ratio 3:1, 3:2, and 6:1. In both cell lines, the expression pattern looked very similar (Fig. 5A). Interestingly, the HA Western blot exhibited only 1 major band at approximately 30 kDa compared with the staining with anti-VHL CT suggesting an N-terminal processing of the full-length pVHL. Next, we tested a set of HA-VHL–mutant constructs for transient expression in MEF^−/−; Fig. 5B]. As expected, the VHL-mutant Leu101Pro with the highest destabilizing value (ddG = 6.1) showed the lowest expression levels similar to those shown in Figure 4A and C. To further establish our screening platform, we stably transfected MEF^−/− cells with VHL WT and selected 3 independent polyclonal pools (batches 1–3) to determine the batch specific variations (Fig. 5C). After selection, the expression levels of pVHL were similar for batches 2 and 3 and slightly lower in batch 1. Subsequently, we transiently transfected these batches with an HIFα-GFP expression vector and verified the expression by Western blot (data not shown), fluorescent microscopy (data not shown), and by quantitative measurement of expression by flow cytometry (FC; Fig. 5D). Compared with MEF^−/− that were stably transfected with a negative control vector, the 3 batches were able to decrease the green fluorescent protein (GFP) signal by 73%, 63%, and 75%, respectively (Fig. 5E). To validate the system, we added DMOG to MEF^−/− pVHL WT cells after transfection with HIFα-GFP and measured the change of the FL-1 channel in gate 1 by FC 24 hours after transfection (Fig. 5F). As a control, pmaxGFP was transfected. After addition of DMOG, the GFP signal increased in both the HIF1α-GFP– and HIF2α-GFP–transfected MEF^−/− pVHL WT cells 2.1-fold and 2.2-fold, respectively. DMOG-treated and -untreated MEF^−/− pVHL WT cells transfected with pmaxGFP showed no increased GFP signals. Next, we generated MEF^−/− stably expressing the same set of pVHL mutants (Fig. 5G) as we used for the transient expression (Fig. 5B). In contrast to the transient transfection experiments (see Fig. 5B), the MEF^−/− expressing the pVHL-mutant Leu101Pro showed similar expression levels as pVHL WT.

By using our screening method, we established MEF^−/− expressing 29 VHL missense mutations, of which 14 have already been described and 15 are not yet registered in the UMD database (13). Additionally, we included 5 well-characterized controls, that is, pVHL WT, Leu63fsX67, Tyr98His, Tyr112His, and Tyr112Asn. MEF^−/− Leu63fsX67 was used as negative control and was set to 100% GFP signal for both the HIF1α-GFP and the HIF2α-GFP transfection experiments (Fig 6A and B). pVHL WT was capable of reducing the signal down to 40%. A comparable reduction was obtained with the VHL missense mutation Ser68Thr. The ddG of pVHL caused by this mutation was neutral (ddG = 0.44) and not destabilizing. In addition, cycloheximide assays showed that the stability of this mutant was similar to that of the WT pVHL, thus confirming our in silico data.

A good correlation was seen between our in silico and in vitro findings when investigating other missense mutations in our dataset. For example, the pVHL destabilizing missense mutation Leu101Pro (ddG = 6.1) showed a severe loss-of-function represented by a GFP signal of 101% for HIF1α and 95% for HIF2α. It is of note, that the MEF cell line stably expressing pVHL Leu101Pro showed similar expression levels in western blots as pVHL WT (see Fig. 5G). Missense mutations described in VHL disease, such as Asn78Ser (type 1: 102% for HIF1α and 86% for HIF2α), Tyr98His (type 2A: 62% for HIF1α and 55% for HIF2α), Tyr98Asn (type 2B: 77% for HIF1α and 74% for HIF2α), and Ser80Asn (type 2C: 40% for HIF1α and 46% for HIF2α) yielded expected results.

As an additional validation, we also analyzed the impact of VHL frameshift mutations on the ability to degrade the 2 HIFα isoforms. For each of the 3 exons we generated 2 frameshift mutations which were identified in our ccRCC patient set. All frameshift mutations except Glu204fsX44, which is located at the very end of exon 3, failed to destabilize HIF1α and HIF2α (Fig. 6C and D). As expected, the frameshift mutations in all 3 exons affected pVHL ability to degrade both HIFα isoforms in the same way (Fig. 6E).

Most of the missense mutations affected HIF1α and HIF2α degradation equally (r = 0.9208, P < 0.0001). Notably, 3 VHL missense mutations showed a selectivity for either HIF1α (Asp121Tyr and Val130Phe) or HIF2α (Asn78Ser) in the range of 10% to 20% (Fig. 6F). Twelve missense mutations identified in our sporadic ccRCC set are also known to be VHL disease related. The analysis of their effects on HIFα destabilization revealed a classification system of VHL mutation types that is similar to that described for VHL disease. In sporadic ccRCC, this resulted in missense mutation type 1: HIFα-GFP expression mean 94% (±11%), type 2A: 58% (±3%), type 2B: 83% (±19%), and type 2C: 64% (±24%).

Finally, we correlated the data obtained from pVHL stability prediction and functionality. Spearman correlation coefficient for HIF1α and HIF2α were: r = 0.5741, P = 0.0006 and r = 0.5494, P = 0.0011, respectively. A cutoff of 58% of the GFP signal for HIF1α and 54% for HIF2α resulted in 13 disease-associated missense mutations with a ddG more than 2 for each of the HIF isoforms (Fig. 6A and B, boxes A). Eleven (34%) and 16 (50%) mutations were predicted to be “neutral” but were compromised in their ability to ubiquitinate HIF1α and HIF2α, respectively (Fig. 7A and B, boxes B). Only 8 (25%) and 3 (9%) missense mutations were identified with similar functionality on HIF1α and HIF2α destabilization, respectively, as pVHL WT (Fig. 7A and B, boxes C).

In this study, we analyzed VHL missense mutations in sporadic ccRCC by determining their impact on pVHL's functionality and HIFα stability using in silico and in vitro assays. The frequencies of the different VHL mutation types as well as their distribution over the 3 exons were highly comparable with the results described in previous publications (2, 37).

Our in silico approach enabled us to locate missense mutations to the surface or the core of pVHL. Most missense mutations (41 of 64) were located in the core of pVHL. The remaining 23 missense mutations were assigned to the surface of pVHL from which 21 were localized within the binding domains of either HIFα, elongin B, or elongin C. Most missense mutations were located in the elongin C interphase. Interestingly, mutations in the HIFα and elongin B binding sites favoured the formation of ccRCC in VHL syndrome patients, whereas mutations in the elongin C interphase were predisposed to developing pheochromocytoma (18). Surface and core missense mutations were equally distributed in hereditary VHL syndrome patients, whereas in our sporadic tumors 64% of the missense mutations were assigned to the core and 36% to the surface. Notably, over two-thirds of the surface missense mutations analyzed in hereditary RCC were located in the interphase of elongin C (18) which is in agreement with our findings in sporadic ccRCC. We detected an accumulation of mutations in exon 1 where the binding site of HIF is located. In contrast, most mutations of hereditary VHL syndrome patients with ccRCC, retinal angioma, central nervous system hemangioblastoma, and pheochromocytoma were located in exon 3. Therefore, loss of pVHL–HIF interaction is more common in sporadic ccRCC compared with its hereditary form.

To resolve the question whether a missense mutation represents a driver mutation, which generates a growth advantage and thus potentially contributes to tumor formation, Carter and colleagues established a computational method termed Cancer-specific High-throughput Annotation of somatic mutations (CHASM; ref. 38). This approach is a powerful tool to analyze missense mutations in multiple genes especially when structures of their gene products are not available. Because of the fact that the crystal structure of the pVHL/elongin B/elongin C complex is known, we used an algorithm which calculates the structural change caused by a specific missense mutation in this complex. Assuming that loss of pVHL function increases HIF transcriptional activity leading to tumor formation and progression, we have considered defining mutations with high structural changes as driver mutations. By calculating the thermodynamic stability of pVHL, we obtained 30 destabilizing driver (“disease-associated”) and 32 (excl. 1 silent mutation) passenger (“neutral”) mutations. Most of the driver missense mutations were located in exon 1 suggesting that the majority of VHL missense mutations in exon 1 predispose to ccRCC by affecting whole protein stability rather than disrupting binding to interaction partners. Only 38% of missense mutations in exon 3 were predicted to destabilize the whole protein structure. However, previous publications reported that loss of elongin binding to pVHL also leads to instability and rapid degradation of the protein (39–41). As all passenger surface-specific missense mutations in exon 3 localized to either elongin C or B, it cannot be excluded that at least some of these mutations exert compromising effects to the regulation of HIF. Interestingly, a recent publication showed that although the missense mutation Arg167Gln causes loss of pVHL–elongin interaction, the HIFα ubiquitination was still functional (22). The authors suggested the formation of a remnant protein E3 ligase complex which is still partially capable of regulating HIF.

To analyze the impact of VHL missense mutations on the thermodynamic stability and HIF regulation in more detail, we established an in vitro screening platform based on MEFs negative for pVHL and generated polyclonal stable cell lines expressing various pVHL mutants. We used MEFs because higher transient transfection efficiencies are obtained in comparison with pVHL-negative RCC4 or 786-O cell lines. We carried out all experiments by using polyclonal batches with similar levels of pVHL rather than individually picked clones to avoid any possible selection for a specific cellular background. Our in vitro platforms verified the predicted ddG of more than 2 for the 4 selected mutations which resulted in both pVHL degradation and HIF stabilization. Additionally, the bioinformatic approach could also correctly predict the impact of different aa exchanges at the same position. For example, the missense mutation at aa Arg161Pro (ddG = −2.86) was completely impaired and could not degrade HIFα, as expected because of structural loss of pVHL, whereas the passenger mutation Arg161Gln (ddG = −0.54) retained 56% functionality.

In the majority of the cases in our study, pVHL mutations affected both HIFα isoforms equally. Only 3 missense mutations seem to selectively influence HIF1α and HIF2α stability. However, whether these relatively modest differences are sufficient to significantly influence the tumor behavior remains elusive. A similar result was reported with the mutation Arg167Gln. Hypoxia experiments with murine embryonic stem cells showed that this mutation apparently was able to regulate HIF1α comparable with the WT but not HIF2α (42). This finding was validated by picking several VHL^−/− clones transfected with the indicated mutant. In contrast, human 786-O cells stably expressing the same pVHL mutant were able to regulate HIF2α expression upon CoCl₂ treatment (22). Similar results were obtained by Bangiyeva and colleagues, who saw comparable HIF2α levels in Arg167Gln, Arg167Trp, and pVHL WT expressing RCC10 cells (21). A contrasting result was published by Clifford and colleagues who detected impaired HIF regulation of both HIF isoforms in RCC4 cells expressing Arg167Gln (43). These different findings may be explained by polyclonal and single cell clone selection, varying levels of exogenous pVHL expression, or by a cell line-specific recruitment of a remnant protein E3 ligase complex.

The majority of the identified VHL missense mutations affected both HIFα isoforms. Therefore, it is tempting to speculate that additional factors on the activity (44), translational (45), and stability level selectively influence the expression levels of the HIFα isoforms (7, 10, 11, 46). One prominent class in the stability regulatory network are the oxygen-dependent prolyl 4-hydroxylases (PHD; ref. 47). PHDs hydroxylate both HIFα isoforms at normoxic levels and allow the E3 ubiquitin ligase complex to interact with its substrate. It was suggested that different PHDs influence the levels of HIF1α or HIF2α depending on the cell type, the tissue, and the degree of oxygenation (48–51) which is independent of the VHL mutation status.

Our in silico tool was able to group VHL missense mutations. The used algorithm proved to be a good predictor of pVHL functionality in terms of thermodynamic stability changes. By combining the results from our in silico and in vitro analyses, we obtained 3 different groups of missense mutations which (i) lead to a severe destabilization of pVHL; (ii) have no destabilizing effects on pVHL but affect the interaction with HIFα, elongin B, and elongin C; and (iii) have functionalities comparable with the WT protein. The first 2 groups represent driver mutations, whereas the latter one consists of passenger mutations. It is to note that 11 and 16 of 32 missense mutations predicted in silico to be “neutral” were compromised in their ability to ubiquitinate HIF1α and HIF2α, respectively. The higher number of missense mutations leading to HIF2α stabilization suggests a more specific binding of pVHL and HIF2α. Taken together, these results imply a gradient effect of VHL missense mutations on HIF regulation which may be caused by a combination of structural changes and alterations of the binding capability to interaction partners. Our systematic approach enabled us to group these VHL mutations according to their effects on HIF which may have important implications for tumor behavior or response to therapy. However, the translation of the HIF (de)stabilizing impact of a VHL missense mutation into clinical application needs further investigation.

Survival analysis of VHL WT, VHL mutation types, and subgrouping of VHL missense mutations into predicted destabilizing and neutral missense mutations, showed no differing effects on patient outcome. Our findings are in line with those described in a previous study (2) in which the impact of VHL mutation types on patient survival was also investigated. Based on these results, it is tempting to speculate that the analysis of VHL mutations in ccRCC patients may serve as predictive rather than as prognostic tool.

In summary, the DNA sequence analysis of VHL in sporadic ccRCC identified known and not yet described VHL mutations. In addition, our in silico and in vitro analyses enabled us to classify VHL missense mutations into 3 different functional subgroups. The knowledge of the impact of a VHL missense mutation on pVHL and HIF stability helps to distinguish between driver and passenger mutations. In this scenario, assessment of the pVHL inactivation status may be useful in predicting response to HIF-targeted drugs.

No potential conflicts of interest were disclosed.

We thank Sonja Brun-Schmid and Roger Santimaria (Institute of Surgical Pathology, University Hospital Zurich, Switzerland) for technical assistance and Carsten Danzer (Cell Biology, ETHZ, Switzerland) for helpful discussions.

This work was funded by the Swiss National Science Foundation and the Zurich Cancer League.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.