High throughput methylation profiling platforms such as the Illumina Goldengate assay enable extensive methylation profiling of human tumours for a large number of genes. Although many of the genes represented in the assay have previously been reported as methylated in human cancers, most (33/42) of the candidate tumour suppressor genes that were methylated in ≥ 20% of all primary RCC tested had not previously been reported to be methylated in RCC (HTR1B, CALCA, IGFBP2, SOX17, COL1A2, BMP4, HS3ST2, FRZB, TAL1, MCM2, KCNK4, HOXC6, PITX2, SEPT5, IRF7, CCNA1, HOXA11, TERT, TMEFF2, EPHA3, PGF, MYOD1, MMP2, TNFRSF10C, PENK, EYA4, MYLK, IRAK3, ZNF215, SMARCB1, TWIST1, SCGB3A1, and IGFBP7). Thus our analysis suggests a large number of potential epigenetic biomarkers and/or TSGs for RCC. In addition, several of these genes have not previously been reported as methylated in any type of cancer (KCNK4, SEPT5, PENK, BMP4, TAL1, PGF, SMARCB1 (INI1), FRZB (SFRP3), IRAK3 and MCM2). Detection of methylated CpGs in urine has been investigated as a potential screening tool for RCC and other urinary tract neoplasms (reviewed in ref 11). To date, no single gene is known to be hypermethylated in all RCCs, and so successful application of promoter methylation assays as biomarkers will necessitate the study of multiple loci. Ideally the combination of loci chosen for such a screening strategy would cover all types of RCC. Such a panel might include genes that are (a) methylated in all types of RCC and (b) preferentially methylated in a particular subtype of RCC, as such a combination could provide a basis for minimally-invasive screening for diagnosis, prognostication and therapeutic targeting. The potential for such a strategy is demonstrated by the results of Euclidean cluster analysis that grouped most tumours into groups consistent with their histopathology/VHL status.
We found evidence that a small number of RCC harboured more methylated CpGs than would be expected by chance. This provides evidence for a "CpG island methylator phenotype" (CIMP+) in a subset of RCC. CIMP+ tumours were not specific to any subgroup of RCC. The concept of a CIMP+ phenotype was developed from studies of colorectal cancer and remained controversial until relatively recently. However, using an unbiased approach to defining putative CIMP-related markers, Weisenberger et al (a) provided evidence for CIMP+ colorectal cancer and (b) demonstrated that CIMP+ sporadic colorectal cancers were associated with the presence of a BRAF mutation and microsatellite instability (from MLH1 promoter methylation) (12). Although CIMP has been described in a range of other tumour types (including head and neck squamous cell carcinoma, bladder cancer, non-small cell lung cancer and malignant pleural mesothelioma) , evidence for a CIMP in RCC has not been reported previously. However, we note that Dulaimi et al  reported that a subset of RCC (~3%) were methylated for at least five out of the ten genes studied. Identification of further CIMP+ RCCs will provide a basis for determining whether tumours with extensive CpG methylation have distinct clinical characteristics and whether a subset of genes are methylated only in CIMP+ phenotype tumours. The molecular causes of CIMP are not well-understood and identification of CIMP+ tumours will facilitate further studies to investigate the molecular basis of this phenotype .
We were interested to compare the methylation profiles of cRCC and papillary RCC, as little information is available on this topic. Several genes were significantly (p < 0.025) more frequently methylated in papillary than in cRCC (and similarly methylated in both cRCC subtypes). These loci could prove to be useful candidate biomarkers for papillary RCC and might give insights into differing mechanisms of tumourigenesis in papillary and cRCC, thereby enable focusing of targeted therapeutic drugs in different disease sub-types. Recently, Matsuda et al  undertook high resolution array CGH studies in sporadic RCC and reported that whereas cRCC was characterized by frequent 3p loss and 5q gain, papillary RCC demonstrated frequent gains on chromosomes 2, 7, and 12; additionally, loss on 1q, 9, and 11q was unique to papillary RCC. However none of the loci preferentially methylated in papillary RCC mapped to 1q, 9 or 11q; and RARB, although preferentially methylated in papillary RCC, maps to 3p24. Thus loci with high frequencies of CpG methylation may occur in regions with a low frequency of allele loss. It is noteworthy that both HOXA11 and HOXC6 were preferentially methylated in papillary RCC. However, although evidence for preferential methylation of genes within a specific chromosome region has been reported in colon cancer , HOXA11 and HOXC6 map to 7p15-p14 and 12q13.3 respectively and so their preferential involvement in papillary RCC presumably reflects their related functions rather than their cytogenetic location. Three genes, CDH1, PTGS2 and TWIST1 were specifically methylated in cRCC (both p < 0.025 compared to papillary RCC). Germline mutations in CDH1 and TWIST1 may be associated with inherited cancer susceptibility. Thus, CDH1 mutations cause familial diffuse type gastric cancer  and the TWIST1 transcription factor is mutated in Saethre-Chotzen syndrome, which is characterised by developmental defects (craniosynostosis and digital anomalies) and is also reported to be associated with an increased risk of breast cancer . PTGS2 (COX2) encodes prostaglandin-endoperoxide synthase 2, and recently Costa et al  also reported that PTGS2 methylation levels were significantly higher in cRCC than papillary RCC. Thus these three genes might prove to be useful epigenetic biomarkers for the diagnosis of cRCC (e.g. in needle aspirates). We note that many of the genes that were preferentially methylated in pRCC could be linked in to TGF-β/ERK/Akt pathways. Although the role of TGF-β-related pathways have not been investigated in pRCC, we note that TFE3 may regulate TGF-β signalling  and that a subgroup of pRCC are charcaterised by chromosomal translocations involving TFE3 .
Although most sporadic cRCC have evidence of VHL inactivation, little is known about whether VHL-inactivated and VHL-wt RCC have differing or similar (for VHL-independent events) mechanisms of tumourigenesis. Also, there is little information on the somatic changes (other than VHL inactivation) associated with RCC from VHL patients. To our knowledge, this is the first investigation of epigenetic alterations in VHL-null and VHL-wt RCC. We found that a number of loci, including RASSF1, PITX2, CDH13, HS3ST2, TWIST1, TAL1, TUSC3 and DCC were significantly more frequently methylated in VHL-wt sporadic RCC than in VHL RCC. It might be hypothesised that genes whose functions overlap with that of VHL-related pathways might be preferentially inactivated in RCC without VHL inactivation. Therefore it is interesting that both VHL and RASSF1A downregulate Cyclin D1 [23–25]. However, although CDH1 expression is downregulated by VHL inactivation, and DAPK1 by RASSF1A inactivation, CDH1 and DAPK1 methylation frequencies were similar in wtVHL sporadic RCC and VHL RCC. Also we note that all of the genes that were preferentially methylated in wtVHL sporadic RCC mapped to different chromosomal locations, and there was no evidence that a specific chromosomal domain was more frequently methylated in a specific tumour type. 5 genes that were more frequently methylated in sporadic wtVHL-cRCC than VHL RCC were included in a network that related to ERK/Akt signalling. Both the Raf-extracellular signal-regulated protein kinase (Erk)1/2 and Akt-mTOR pathways have been implicated in the pathogenesis of familial and sporadic RCC [26, 27]. VHL-inactivation may be associated with activation of the epidermal growth factor receptor/phosphatidylinositol-3-OH kinase/protein kinase B (AKT)/IkappaB-kinase alpha/NF-kappaB signalling cascade , and so RCC without VHL inactivation may be predisposed to dysregulate these key signalling pathways by preferential methylation of other regulators. Although it might be predicted that the pathways of tumourigenesis in VHL mutated sporadic clear cell RCC will be similar to those in RCC from VHL patients, further studies (using the methods utilised in this study) are indicated to investigate this point.
It is generally thought that CpG island methylation in cancer is non-random, and many genes that are frequently methylated have been shown to have a tumour suppressor or DNA repair function. TSGs may be targeted by epigenetic silencing and/or inactivating mutations during tumourigenesis. However, whereas the repertoire of inactivating mutations in a typical tumour suppressor gene is extensive, patterns of CpG island methylation causing epigenetic silencing are much more restricted and hence are easier to detect. The relative frequency of CpG island methylation and inactivation mutations for individual TSGs is variable. For example, although VHL is primarily inactivated by somatic mutations (promoter methylation occurs in ~15% of sporadic cRCC), inactivation of the RASSF1A TSG in RCC (and in other tumour types) most commonly results from promoter methylation, whilst intragenic mutations are rare. According to the Catalogue of Somatic Mutations in Cancer (COSMIC; http://www.sanger.ac.uk/perl/genetics/CGP/cosmic) only three genes are mutated in >5% of RCC tested (VHL = 42%, CDKN2A = 12% and KIT = 8%). Thus, with the exception of VHL, epigenetic biomarkers are likely to provide a better approach for novel prognostic and diagnostic strategies than genetic markers . This is substantiated by the observation that, of the 43 genes that were methylated in >20% of RCC tested, RCC mutation analysis data is available for 21, of which none are subject to somatic mutations in RCC. Further characterisation of the genes and pathways that are epigenetically altered in RCC of different subtypes may thus lead to the development of novel minimally-invasive diagnostic and prognostic tools for kidney cancer, and, in the longer term, may enable more focused treatments for individual tumours.