Consequences of VHL Loss on Global DNA Methylome

In clear-cell renal cell carcinoma (ccRCC), loss of von Hippel-Lindau (VHL) tumour suppressor gene and reduced oxygen tension promote stabilisation of hypoxia-inducible factor (HIF) family of transcription factors, which promote changes in the expression of genes that contribute to oncogenesis. Multiple studies have demonstrated significant perturbations in DNA methylation in ccRCC via largely unclear mechanisms that modify the transcriptional output of tumour cells. Here, we show that the methylation status of the CpG dinucleotide within the consensus hypoxia-responsive element (HRE) markedly influences the binding of HIF and that the loss of VHL results in significant alterations in the DNA methylome. Surprisingly, hypoxia, which likewise promotes HIF stabilisation and activation, has relatively few effects on global DNA methylation. Gene expression analysis of ccRCC patient samples highlighted expression of a group of genes whose transcription correlated with methylation changes, including hypoxic responsive genes such as VEGF and TGF. These results suggest that the loss of VHL alters DNA methylation profile across the genome, commonly associated with and contributing to ccRCC progression.

often alongside areas of inadequate oxygen supply gives rise to a tumour environment that is characterised by high HIFα expression at a variety of oxygen tensions ranging from normoxia to severe hypoxia.
As is the case for all transcription factors, there must be adequate access to respective binding regions of DNA in order for transcriptional activation to occur 8 . HIFα binds to hypoxia response elements (HREs) containing the consensus sequence RCGTG (where R is A or G) to upregulate gene transcription 9 . The likelihood of transcription factor binding is often determined by epigenetic factors and previous studies have demonstrated that changes to epigenetic code surrounding HREs can augment HIF mediated transcription 10,11 . Interestingly, regulation in the opposite direction appears to also occur, whereby HIF stabilisation can elicit changes in the epigenetic code 12,13 . HRE sequences are of particular interest in relation to DNA methylation as there is a CpG dinucleotide within the consensus sequence. CpG dinucleotides can become epigenetically altered via DNA methylation, whereby the 5 th carbon of the cytosine residue gains a methyl group to become 5-methylcytosine (5MeC). The complexities of DNA methylation continue to be discovered 14 . However, in general terms methylation is associated with chromatin condensation and transcriptional repression, whilst loss of DNA methylation is associated with a loosening of chromatin and at gene promoters can result in upregulated gene expression 15 .
Global hypomethylation alongside promoter specific hypermethylation events are common features of most cancers. The Cancer Genome Atlas Research Network along with studies on smaller patient cohorts reported these as features of ccRCC patient samples 16 . Hypomethylation is thought to contribute to genome instability while promoter specific hypermethylation contributes to disease progression via silencing of anti-oncogenic genes 17 . Despite the common occurrence of these significant changes in the DNA methylome, little is known about the mechanisms in renal cancer cells that cause such distinct changes in DNA methylation and the repercussions of such changes to HIF mediated gene expression. This study was undertaken in order to investigate these questions in relation to ccRCC. Given that HIF stabilisation as a consequence of VHL loss and/or hypoxia are important events in the pathogenesis of ccRCC, alongside the fact that HIF has previously been implicated in modifying the epigenome, we asked whether these two events impact the DNA methylome of ccRCC.

CpG methylation within HRE prevents HIF binding and subsequent transcription. HIF binds
to the consensus HRE sequence (5′-(A/G)CGTG-3′) within promoters and enhancers to upregulate hypoxia-inducible gene transcription. The number of direct HIF target genes is intriguingly and significantly less than the number of HRE sequences found in the genome potentially available to HIF. Epigenetic modification is a potential mechanism that may prevent HIF binding to inappropriate sites. Considering that the HRE sequence contains a CpG dinucleotide capable of undergoing changes in DNA methylation, we asked whether CpG methylation at HRE would have an impact on HIF binding. The radiolabelled VEGF HRE probes containing either synthetically methylated or unmethylated CpG dinucleotide were analysed for binding to HIF1α and HIF2α via electrophoretic mobility shift assay (EMSA) (Fig. 1A). Both HIF1α and HIF2α bound robustly to radiolabelled unmethylated HRE probes, which were abolished by increasing amounts of competing unlabelled unmethylated HRE probes (Fig. 1A, lanes 2-3 and lanes 6-7). In contrast, neither HIF1α nor HIF2α was capable of binding radiolabelled methylated probe, even in the absence of competing unlabelled methylated probe (Fig. 1A, lanes 11 and 15). These results demonstrate that the affinities of both HIF1α and HIF2α for HRE sequence are markedly reduced when the CpG within HRE is methylated.
Consequently, the presence of methylation at HREs would have a marked impact on HIF mediated gene expression. Previous studies have also alluded to this and HIF responsive genes such as CAIX and EPO are both repressed if their respective HREs or areas in the vicinity of the HRE are methylated 11,18 . Considering the EMSA results, we next asked whether removal of CpG methylation would affect HIF mediated gene expression of known HIF responsive genes. For these studies we used human ccRCC cells stably expressing HA-tagged wild-type VHL (RCC4-VHL). Despite the fact that VHL degrades HIFα via ubiquitin-proteasome system in an oxygen-dependent manner, RCC4-VHL cells express low levels of HIFα in normoxic conditions whereas in hypoxia HIFα is dramatically stabilised. These cells were treated with the DNA methylation inhibitor 5-aza-2-deoxycytidine (decitabine) in normoxia and in hypoxia. Gene expression levels of classic hypoxic responsive genes (VEGF, uPAR and TGFα), alongside 2 genes that do not contain HREs (GUS and U1AsnRNP1), were quantified using qPCR (Fig. 1B). Interestingly hypoxic responsive gene expression was enhanced in both normoxia and hypoxia when cells were treated with decitabine whereas expression of genes without HREs remained unchanged. The increased expression in normoxia upon decitabine treatment was similar to the expression induced by growing cells in hypoxia, except for uPAR. While demethylation of tissue-specific enhancers or epigenetically silenced trans-acting transcription factors could impact transcription of HRE-containing genes, it is equally likely that that regardless of oxygen tension, a certain degree of methylation may exist at or near many of these HREs, which acts to dampen HIF mediated transcription and that removal of DNA methylation enhances HIF mediated transcription.
Loss of VHL results in significant changes to the DNA methylome. While pharmacological manipulation of DNA methylation in our cell system can alter HIF mediated transcription, we questioned whether changes at CpG sites occur in situations where HIF is known to be constitutively active and therefore capable of mediating transcription. Significant changes in DNA methylation are evident when normal renal tissue is compared to patient renal tumours 16 . We hypothesised that VHL loss, the most commonly occurring genetic alteration in ccRCC and accompanied HIF stabilisation, may contribute to the altered epigenome in ccRCC. Under these circumstances we proposed that methylation changes at HREs could effectively heighten and/or alter the HIF mediated response. In order to investigate DNA methylation differences, we employed a genome-wide approach by using Illumina human methylation 450 K arrays. These arrays offer valuable insight into the degree of methylation at a wide variety of CpG residues at various positions across the genome. We therefore chose to  Fig. 2A). Human osteosarcoma U2OS cell lines stably expressing non-targeting shRNA or shRNA against endogenous VHL were generated and 5MeC levels EMSA of in vitro-translated HIF1 and HIF2 binding to 32P-labelled unmethylated and methylated HRE probes. Competition with 250X molar excess of unlabelled/unmethylated HRE probe (lanes 3, 7,12, 16) or unlabelled control HRE-free probe (lanes 4,8,13,17). HIF1 and HIF2 complexes bound to unmethylated HRE or methylated HRE supershifted with anti-HIF1α (lanes 9 and 18) and anti-HIF2α (lanes 5 and 14). (B) RCC4-VHL cells were grown in normoxia (NT) or treated with 25 µM decitabine (5aza) or grown in hypoxia (1%) for 48 hrs and with the addition of 25 µM decitabine (1% + 5aza). QPCR was performed using primers specific to VEGF, uPAR, TGFα, GUS, U1AsnRNP1. Samples were normalised to relative expression of the housekeeping gene, ACTIN.
were measured using the same arrays. Differential methylation analysis identified 91 and 2523 loci as being significantly hyper-and hypomethylated, respectively, following VHL knockdown (Supplementary Table 3; Fig. 2A). The differentially methylated loci were identified across the genome in all three cell lines ( Fig. 2B) highlighting that VHL status has a global effect on DNA methylome.
Given the significant changes in DNA methylation between VHL+ve and isogenically matched VHL−ve cells, we next investigated what CpG sites were being altered in these cells. We focused our analysis on RCC4 and 786 cells as both of these cell lines are derived from ccRCC harbouring a loss of endogenous VHL and thus are more relevant for studying functional consequences of VHL loss in ccRCC pathogenesis. Our preceding analyses of 786 and RCC4 cells uncovered many thousands of sites that showed significant changes in their methylation upon re-expression of VHL (Supplementary Tables 1 and 2). When combining data of 786 and RCC4 cell lines, we identified 20,834 loci that exhibited similar changes in these two cell lines upon ectopic expression of VHL; 16,902 with hyper-and 3,932 with hypomethylation. To investigate the potential contribution of DNA methylation of these loci to the hypoxia-mediated transcription, we examined the coincidence of these loci with canonical HRE and observed 2-fold enrichment of hypermethylated loci in HREs (p = 2.2 × 10 −16 , Fisher's exact test) compared to hypomethylated loci (Fig. 2C). These results suggest that VHL-dependent DNA hypermethylation may affect HIF-mediated gene expression by altering epigenetic status of HREs, thereby modulating HIF binding to HREs.

Hypoxia has minimal effects on global 5MeC that is likely governed by repression of DNMT1.
As we recorded dramatic changes in DNA methylation between cells with and without VHL, we next performed similar experimental procedure to investigate the effect of hypoxia on DNA methylation patterns. Hypoxia occurs in a variety of situations in both health and disease, including cancer. It has previously been shown that vast changes in the chromatin signature are required to elicit a complete hypoxic response 10,19 ; we therefore hypothesised that this process would likely involve changes in DNA methylation marks, some of which may occur at putative and/or recognised HREs. Surprisingly the results from these experiments revealed striking similarities between the DNA methylomes of RCC4-VHL cells grown for 72 hours in 1% O 2 and 21% O 2 , (Fig. 3A); there were only 7 loci showing significant differential methylation between them. Considering this lack of change in 5MeC between normoxic and hypoxic cells, we questioned whether the results that were recorded in RCC4-VHL cells were simply an anomaly specific to that cell line. In this regard, we performed the same hypoxic experiment using human fetal lung fibroblast cell line, MRC5. Consistent with the results obtained in the ccRCC cell line, we identified only 193 (1 hyper-and 192 hypomethylated) loci with differential methylation levels between MRC5 cells grown in hypoxic and normoxic conditions. Initially we considered that perhaps 72 hrs was not sufficient time for the 5MeC levels to significantly change. To investigate this possibility, RCC4-VHL cells were grown in 1% O 2 for 168 hours (7 days) and DNA methylation was measured. However, no significant changes between normoxic and chronically hypoxic cells were apparent (total of 33 CpG sites exhibited differential methylation; Fig. 3A). Given that TET enzymes, as well as other epigenetic modifiers, are oxygen dependent, we next hypothesised that perhaps hypoxic cells required a higher level of oxygen in order to maintain activity of all DNA methylation machinery. Hence, ccRCC cells were additionally grown in 3% and 5% O 2 . In these experimental conditions HIFα remains stabilised. However, our genome-wide analysis revealed slight to moderate changes between normoxic and hypoxic cells grown in these O 2 levels (3,384 and 1,231 differentially methylated loci when normoxic condition was compared to 3% and 5% O 2 , respectively, Fig. 3A; Supplementary Tables 4 and 5). These results suggest that hypoxia does not have a profound effect on 5MeC status while loss of VHL in the same cell type promotes dramatic changes in global DNA methylation ( Fig. 2A).
QPCR analysis of the DNMTs as well as expression of the TETs were analysed in normoxic and hypoxic RCC4-VHL cells (Fig. 3B-C). At an mRNA level, DNMT1 transcription was significantly repressed in hypoxic cells (Fig. 3C). Protein expression analysis revealed that DNMT1 expression was markedly attenuated during 48-hour growth in hypoxia in a wide variety of cells of human, rat and mouse origin (Fig. 3D). DNA methylation can be lost passively via suppression of DNA methyltransferases. Therefore it is likely that the decrease in DNMT1 mediates many of the changes seen in hypoxic cells. To address this notion, RCC4-VHL cells were transfected with non-targeting scrambled siRNA (siSCR) or DNMT1-targeting siRNA (siDNMT1) and simultaneously grown in normoxia. Illumina 450k arrays were used to interrogate changes in DNA methylation. There were slight changes in DNA methylation (999 loci) upon DNMT1 knockdown in RCC4-VHL cells maintained in normoxia (Fig. 3E), which was relatively similar in number to the changes recorded in hypoxic cells. Although the numbers of CpG sites were similar, 2% of these sites were the same in hypoxic cells and siDNMT1 cells. While the results suggest that the changes in DNMT1 level in hypoxia may account for the modest loss of DNA methylation recorded in these cells, the results also indicate that the location of these changes appear to be more random.

Changes in 5MeC at various regions in primary ccRCC include HIF target genes and putative
HREs. We next compared the results from methylation profiling of ccRCC cell lines to the methylome of human primary ccRCC specimens, which are characterised by inactivation of VHL due to somatic mutations or promoter hypermethylation. We used differential methylation data between ccRCC samples and matched normal kidney tissue from 100 individuals, which were profiled with the same arrays. This analysis showed that VHL re-expression in cell lines can change methylation of 2197 out of 34598 loci with ccRCC-associated methylation to the patterns observed in normal kidney tissue, suggesting that VHL loss contributes to the abnormal methylation of 6.3% of affected loci. It is notable that this fraction represents CpG loci with high-confidence methylation changes observed recurrently across multiple tumors. Therefore, the extent of VHL-dependent DNA methylation changes across the genome may well be much broader. However, among 18,191 CpGs that were hypomethylated in ccRCC as compared to patient-matched normal kidney tissues, 2,037 loci (11.2%) become hypermethylated upon ectopic expression of VHL in ccRCC cell lines. In contrast, among 16,407 CpG sites that exhibited hypermethylation in ccRCC tumors, only 160 loci (~1%) showed hypomethylation in VHL-reconstituted cell lines. These findings suggest that, in regards to epigenetic aberrations, VHL deficiency contributes predominantly to the loss of DNA methylation in ccRCC tumors (Fig. 4A, Supplementary Table 6) We further investigated possible correlation between abnormal methylation of these loci and cancer associated expression of the genes associated with them. Given the well-known negative regulation of gene expression by DNA methylation, we focused our analysis on genes that were associated with hypomethylated loci and were upregulated in ccRCC tumours compared to normal kidney samples. Our analysis identified 83 genes that displayed significant negative correlation between methylation status of their associated CpG loci and expression of their mRNA across sample pairs (Pearson correlation <−0.3, p < 0.05; Supplementary Table 7). Pathway analysis of these genes showed enrichment in various pathways including many immune system and cell signalling related pathways. Furthermore, several genes implicated in ccRCC development or progression were distinctly altered in both tumour samples and cells. Intriguingly, and in support of our previous findings, these genes included HIF responsive genes such as VEGFA, TGFBI 20 . While the loss of 5MeC did not occur in VEGF HRE, significant alterations did take place in the promoter (Fig. 4B). In addition many other genes with putative HREs were also affected, including TRIB3, INSR and CD48. Similar analysis with genes associated with loci that showed hypomethylatation upon VHL expression in cell lines and hypermethylatation in ccRCC tumors returned only 3 genes whose expression showed negative correlation to the CpG methylation; HOXB8, FAM89A and PPP2R2D, of which none were among the HIF-target genes.
According to the role of DNA methylation of promoter CpG islands on the regulation of gene expression, we also investigated associations between changes of DNA methylation of promoter CpG islands and alterations in expression of nearby genes in ccRCC, and the potential involvement of VHL. To this end, we first identified cluster of CpGs, which are annotated to the same promoter CpG islands, and represented similar changes (hyper or hypomethylation) in ccRCC compared to matched normal samples, and then examined association between weighted average methylation value of these cluster and expression of the associated gene (see Methods for details). This analysis resulted in 42 genes whose expression changes showed a negative correlation with changes of DNA methylation of their promoter CpG clusters, which consisted of minimum of three CpGs (Supplementary  Table 8). We observed that methylation of clusters associated with fourteen of these 42 genes is changed upon expression of VHL in either 786-O or RCC-4 cell lines (Supplementary Table 8). These include VEGFA, which was also identified by our single CpG analysis.

Discussion
VHL is a multifunctional protein. To date the best described function of VHL is its role as the substrate recognition subunit of the ECV complex that participates in the degradation of HIF. HIF independent functions for VHL have also been reported including its involvement in the degradation of other target proteins 21 as suggested roles independent of its E3 ubiquitin ligase function 23 . Here we report a potential role for VHL in the regulation of the DNA methylation profile of cells. The solitary loss of VHL resulted in dramatic changes in DNA methylation in different cell types. Despite the fact that ectopic VHL expression in the cell lines tested could result in promoting supraphysiological effects on the DNA methylome, many of the changes we recorded in the cell lines were evident in ccRCC patient samples where loss or mutation of VHL is the most common and earliest genetic perturbation.
In contrast to the results observed upon manipulation of VHL, this study demonstrates that in hypoxia where HIF is also stabilised, there is relatively little change in the global DNA methylome. Given that in both situations HIF is activated, it is somewhat surprising that there is such a vast difference in the methylome between these conditions, i.e., hypoxia versus VHL loss. This result is particularly intruiging as previous investigations have reported an association between HIF stabilisation and promotion of epigenetic modifications including DNA methylation. Many epigenetic enzymes contain putative HREs and have been proposed to be HIF target genes. These epigenetic modifiers, like the PHDs, also require oxygen for their enzymatic function. When TET enzymes are active, they oxidize conversion of 5mC to 5hmC which through a series of TET mediated reactions is converted to 5fC or 5caC. 5MeC will eventually be returned to cytosine, the net result being demethylation. Although they are robustly expressed in the ccRCC cells used in this study, TET mRNA expression was unchanged in hypoxic RCC4-VHL cells. This finding is in support of a recent study that quantified TET mRNA expression in a variety of hypoxic cells. While mRNA expression varied, activity of these enzymes was inhibited providing an effective mechanism to incur hypermethylation events 24 . Other studies have also interrogated the relationship between hypoxia and TETs and demonstrated that HIF1α mediated induction of TET1 promotes increased global 5hMeC in neuroblastoma cells but not in other cell types tested 25 . Interestingly in this study, despite manipulation of hypoxic time point to 7 days as well as the degree of hypoxia, there was very little effect on global 5MeC, suggesting the unlikely suppression of TET activity in this system. In fact the small degree of demethylation in hypoxia could be accounted for by the loss of DNMT1. Loss of DNMT1 results in passive removal of 5MeC during cell replication although other reports implicate that loss of DNMT1 may also have consequences independent of its catalytic activity 26 .
While a reduction in DNMT1 may account for most of the changes evident in the methylome in hypoxia, its loss does not provide a dynamic enough pathway to account for the vast changes evident in ccRCC cells lacking VHL. Loss of VHL leads to significant hyper and hypomethylation changes. Given previous reports in the literature regarding HIF mediated regulation of epigenetic machinery it is possible that altered expression of the enzymes that regulate DNA methylation may occur in a HIF mediated fashion upon VHL loss. However, changes exclusive to DNA methylation enzymes may not account for all the modifications in 5MeC upon VHL loss. It is also conceivable that other yet-identified or -confirmed mechanisms contribute to the substantial changes recorded throughout the DNA methylome. For example, many ccRCC tumors harbour mutations, predominantly of epigenetic regulators such as PBRM1 and SETD2 16,27 . There is growing evidence supporting the influence and interplay of histone modifications on DNA methylation. Recently it has been demonstrated that SETD2 inactivation results in global redistribution of 5MeC in renal cells 28 . These are likely contributory factors to changes in DNA methylation, which alongside VHL may account for the diverse alterations evident in ccRCC samples. Although the precise mechanism(s) by which VHL regulates the epigenome remains an outstanding question, it is conceivable that VHL itself directly influences chromatin organisation. In support of such notion, VHL was observed to bind to heterochromatin associated protein HP1 29 . Interestingly, this interaction was not related to the degradation function of VHL instead speculating that HP1 recruits VHL to chromatin. Little is known about the functional significance of the VHL-HP1 interaction, although it is tempting to speculate that perhaps such an interaction assist in maintaining repressive chromatin structure where high levels of DNA methylation also exist.
The findings that VHL loss and hypoxia impact the DNA methylation profiles of cells differently may have very interesting implications for the development and progression of ccRCC. It is widely acknowledged that loss or mutation of VHL occurs early in tumourigenesis, while hypoxic regions of tumours are associated with later disease stages when avascular or dysregulated angiogenesis, primarily mediated by VEGF, are commonplace. Based on the observations of this study it is plausible that VHL loss early in disease progression, at a time when the tissue is well oxygenated, may significantly alter the DNA methylome. These changes could include removal of methylation at gene promoters, including HIF binding sites. As a consequence cancer cells may be capable of exaggerated expression of HIF target genes such as VEGF and TGF, as well as others. Meanwhile in hypoxia, although expression of HIF target genes occurs, our results would suggest that the amplitude is capped as DNA methylation remains and prevents an exaggerated expression capacity.
Taken together, we demonstrate that loss of VHL dramatically impacts the DNA methylation landscape of renal cancer cells while in contrast there are relatively few changes to 5MeC in hypoxia. These findings are in support of previous studies that have implicated an association between VHL loss and epigenetic modifications that promote cancer progression 30,31 . The results support the notion that VHL influences DNA methylation early in disease and likely accounts for a significant proportion of the perturbations that are evident in the DNA methylome of tumours.

Materials and Methods
Patient samples. Patients  , as well as from local ethics committee for recruiting centres in Czech Republic, Romania and Russia. All sampling and clinical data collection was undertaken according to predefined standard operating procedures following guidelines from the International Cancer Genome Consortium. Further information about the sampling procedure, isolation of nucleic acids as well as genome and transcriptome sequencing have been described in our previous study 32 . We included DNA methylation data of ccRCC and patientmatched normal kidney tissue from 100 patients in our analysis. For correlation analysis between changes of DNA methylation and gene expression levels we focused our analysis on tumor and normal samples pairs of 40 out of the 100 patients for whom both DNA methylation and gene expression data were available. 450K DNA methylation arrays and analysis. RCC4-VHL, RCC4-mock, 786-mock, 786-VHL, U2OSshSCR, U2OSshVHL, MRC5s were grown in normoxia, hypoxia or treated with siDNMT1. Protein (method as previously described) was extracted from each sample and tested using western blot to validate presence of expected markers and ensure the experimental conditions were effective. Genomic DNA was extracted from the same samples using an extraction kit (Qiagen;) according to the manufacturer's instructions. 1.5 µg of each sample was then bisulfite treated with a kit (EZ DNA methylation Kit; Zymo Research) according to manufacturer's instructions. Bisufite treated samples were then loaded onto the Illumina 450 K array platform.
Bioinformatics and statistical analysis. Processing and statistical analysis of the HumanMethylation450K array data was performed using the R statistical language version 3.2.2 and various packages of the Bioconductor project 33 . Specifically, minfi 34 was used to load, apply SWAN normalization 35 , annotate probes and filter probes mapping to known SNPs. Differential methylation of loci between cell lines was tested using the Limma 36 on the M-values 37 . Statistical significance for differential methylation was reached at 5% Benjamini-Hochberg false discovery rate and an absolute difference of M-values of 1. CpG loci were annotated as putative hypoxia-response elements (HREs) if they were part of the RCGTG motif.
Correlation between DNA methylation and gene expression was performed on differences of DNA methylation between tumor and normal of each patients and the differences of gene expression levels of the same samples using Pearson's correlation. We compared differences in mRNA abundance of a given gene between patient-matched tumor and normal samples to the difference of DNA methylation levels of loci associated with that gene between the same specimens.
WGCNA package [PMID:19114008] was used to identify cluster of CpGs associated with the same gene that show similar changes of DNA methylation patterns. CpG clustering was only performed on genes which have at least five cis CpGs. The eigengene, which represents the weighted average value of methylation level of the cluster was extracted, and then a regression model was fit between each eigengene and expression of its associated gene.