Epigenetic aberrations are frequent in prostate cancer and could be useful for detection and prognostication. However, the underlying mechanisms and the sequence of these changes remain to be fully elucidated. The tumor suppressor gene RARRES1 (TIG1) is frequently hypermethylated in several cancers. Having noted changes in the expression of its paralogous neighbor gene LXN at 3q25.32, we used pyrosequencing to quantify DNA methylation at both genes and determine its relationship with clinicopathological parameters in 86 prostate cancer tissues from radical prostatectomies. Methylation at LXN and RARRES1 was highly correlated. Increasing methylation was associated with worse clinical features, including biochemical recurrence, and decreased expression of both genes. However, expression of three neighboring genes was unaffected. Intriguingly, RARRES1 methylation was influenced by the genotype of the rs6441224 single-nucleotide polymorphism (SNP) in its promoter. We found that this SNP is located within an ETS-family-response element and that the more strongly methylated allele confers lower activity in reporter assays. Concomitant methylation of RARRES1 and LXN in cancerous tissues was also detected in prostate cancer cell lines and was shown to be associated with repressive histone modifications and transcriptional downregulation. In conclusion, we found that genotype-associated hypermethylation of the ETS-family target gene RARRES1 influences methylation at its neighbor gene LXN and could be useful as a prognostic biomarker.
Prostate cancer (PCa) is the most common cancer among elderly men and has a relatively long natural history. Its highly variable course fuels a debate about optimal patient management and could reflect its pathogenetic and epigenetic heterogeneity.1
Epigenetic abnormalities, especially hypermethylation of CpG-rich gene promoters, are found throughout all stages of prostate and other cancers.2, 3 The gain of abnormal DNA methylation is, in general, associated with repressive histone modifications and lower transcriptional activity. Changes in DNA methylation occur particularly often at polycomb target genes and the histone methyltransferase EZH2, a component of the polycomb complex PRC2, is commonly upregulated in PCa.4, 5 In some instances, promoter hypermethylation takes place in the context of long-range epigenetic silencing characterized by the co-ordinated transcriptional repression of several adjacent genes or an entire chromosomal region.6, 7 Conversely, genome-wide hypomethylation resulting in the reduction of overall methylcytosine content and especially the methylation of endogenous retroelements arises later in tumor progression.8, 9
Among the most common genetic changes in PCa are chromosomal rearrangements involving the ETS transcription factor family, in particular formation of the fusion oncogene TMPRSS2-ERG. It is debated whether this gene fusion is associated with a more aggressive course of the disease.10 Likely, genetic and epigenetic changes in PCa interact. For instance, it was shown that ERG induces transcription of EZH2 and consequently promotes repressive epigenetic marks.11
Previous studies have suggested that hypermethylation of RARRES1 (TIG1) might be associated with the progression of PCa12, 13 although the findings are not fully concordant.14 Intriguingly, we noticed changes in the expression of its neighbor gene LXN. Therefore, we investigated the epigenetic state of both genes in benign and PCa tissue samples and cell lines. In addition, we considered the influence of frequent single-nucleotide polymorphisms (SNPs) in the RARRES1 and LXN promoters. Both genes are arranged consecutively at 3q25.32 at an approximate distance of 60 kbp (Fig. 1a) and their paralogous products Retinoic acid receptor responder protein 1 and Latexin share high-sequence homology with cystatins.15 Interestingly, the PCa aggressiveness locus 7q32 harbors an imprinted gene cluster encoding members of the carboxypeptidase A family16–18 for which Latexin is the only established inhibitor in mammals. RARRES1 has been recently described to interact with the carboxypeptidase AGBL2.19
Here we show, by a quantitative approach and multivariate Cox regression analysis, that increasing RARRES1 methylation is strongly and independently associated with biochemical recurrence. Its methylation is affected by the SNP rs6441224, which modulates the transcriptional activity of the RARRES1 promoter and is located in a binding site for ETS-family transcription factors. Hypermethylation of its paralogous neighbor LXN was strongly correlated with increasing RARRES1 methylation. Concomitant methylation of both genes results in a silenced epigenetic state, which does not encompass other neighboring genes.
Material and Methods
PCa tissues and adjacent benign prostate tissues were obtained from patients (median age, 66 years) without distant metastases who underwent radical prostatectomy at our institution between 1997 and 2002. Patient and sample parameters are summarized in Supporting Information Table S1. Selection and histological evaluation of the samples was performed by a qualified pathologist at the Department of Pathology. TNM classification was performed according to the rules of the International Union against Cancer from 2002. Follow-up data were available for a median period of 92 months. Biochemical recurrence defined as a total PSA of >0.2 ng/mL on two consecutive tests was considered as the surrogate endpoint for survival analysis. The study was approved by the ethics committee of the Heinrich Heine University medical faculty.
The prostate carcinoma cell lines LNCaP, 22Rv1, PC-3 and Du145 were cultured in RPMI1640 (Gibco Life Technologies, Karlsruhe, Germany), supplemented with 10% fetal calf serum (FCS) and penicillin (100 U/mL)/streptomycin (100 μg/mL). Treatment with 2 μM 5-aza-2′-deoxycytidine was performed by daily addition for 3 days. MDA PCa 2b (MDAPCa) was cultured on collagen-coated dishes in BRFF-HPC1 (AthenaES, Baltimore, MD), supplemented with 20% FCS, 2 mM L-glutamine and penicillin (100 U/mL)/streptomycin (100 μg/mL). Normal prostate epithelial cells (PrEC) purchased from Lonza (Verviers, Belgium) were cultured as recommended by the supplier. Bladder cancer cell lines were cultured in Dulbecco's Modified Eagle Medium (Gibco), supplemented with 10% FCS and penicillin/streptomycin.
Nucleic acid extraction and quantitative reverse-transcription polymerase chain reaction
DNA and RNA were extracted from the same powdered tissues. High molecular weight genomic DNA was isolated using the blood and cell culture DNA kit (Qiagen, Hilden, Germany). Leukocyte DNA was isolated from total peripheral blood. Total RNA was isolated using the RNeasy Mini Kit (Qiagen) after guanidinium/acid phenol/chloroform extraction (peqGOLD TriFast, peqLab, Erlangen, Germany). Quality of DNA and RNA was initially checked by spectrophotometry and subsequently by agarose gel or capillary electrophoresis, respectively. Only high-quality DNA and RNA preparations were used in the present study. cDNA was prepared using SuperscriptII (Invitrogen, Karlsruhe, Germany) according to the manufacturer's protocol with a mixture of random and oligo-dT primers. qPCR was performed on a 7500 Fast Real-Time PCR System (Applied Biosystems, Darmstadt, Germany) using commercially available Taqman assays or QuantiTect SybrGreen PCR Kit (Qiagen) with primers specific for the respective mRNA (Supporting Information Table S2). Relative expression was calculated by Pfaffl's ΔΔCT method.
DNA methylation analysis
DNA was bisulfite-treated using the EZ DNA Methylation Gold Kit (Zymo Research, Freiburg, Germany). Pyrosequencing was performed on a PyroMark Q24 instrument (Qiagen) according to the manufacturer's protocol. Primers were designed using the PyroMark assay design software version 2.0. For quantitation, the average methylation across CpG sites 1-5 at the LXN promoter and CpG sites 2–7 at the RARRES1 promoter was considered. For conventional bisulfite sequencing, the PCR products were cloned into pCR4-TOPO (Invitrogen, Karlsruhe, Germany) and sequenced by standard methods in the biomedical research center of the Heinrich Heine University. All primer sequences and PCR reaction details are summarized in Supporting Information Table S2.
Transfection and luciferase assays
RARRES1 promoter fragments were obtained from homozygous PCa cell lines using HotStar Taq Polymerase (Qiagen) with the indicated primer pairs (Supporting Information Table S2). The PCR products were sequenced in pCR4-TOPO and subsequently cloned into pGL3basic (Promega, Mannheim, Germany) vector. The ETS1-pCDNA3 and Mock-pCDNA3 expression vectors were kindly provided by Prof. Nicolas Wernert (University of Bonn). The ERG-pCDNA3 expression vector was kindly provided by Dr. Michèle J. Hoffmann. For transfection, Du145 and PC-3 cells grown in 24-well plates were transfected using FuGene 6 (Roche, Basel, Switzerland) according to the manufacturer's instruction. Reporter activity was measured 48 hr later using the Dual-Luciferase Reporter Assay System (Promega, Mannheim, Germany).
Chromatin immunoprecipitation and quantification by real-time PCR
ChIP was performed by the ChIP-IT Express Kit (#53008, Active Motif, Brussels, Belgium) according to the instructions of the manufacturer. In brief, intact cells were fixed with 1% formaldehyde. The crosslinked chromatin was sheared by sonication to obtain fragments in the range 200–1,500 bp. Approximately, 10 μg of sheared chromatin per reaction was immunoprecipitated overnight with protein-G-coated magnetic beads, antibodies against H3K9ac (#ab4441), H3K9me3 (#ab8898, both Abcam, Cambridge, United Kingdom), H3K27me3 (#39535, Active Motif) compared to positive (RNA Polymerase II antibody) or negative (IgG antibody) control antibodies (CHIP-IT Control Kit Human #53010, Active Motif), in the presence of proteinase inhibitor cocktail. After washing out unbound proteins, the bound chromatin was eluted from the beads, crosslinks were reversed and DNA was recovered after treatment with proteinase K. Before DNA was used for PCR analysis, it was treated with a proteinase K inhibitor (ChIP-IT Express Kit, Active Motif). The ChIP fractions were quantified by real-time PCR amplification using QuantiTect SYBR Green PCR Kit (Qiagen) and primer pairs for the promoter regions of RARRES1, LXN, GAPDH and CTCFL (Supporting Information Table S2). DNA from nonprecipitated sheared chromatin (input) was purified in parallel to the ChIP reactions and was used to create a standard curve in the qPCR analysis. The relative quantity of active histone modifications at RARRES1 and LXN was normalized versus the enrichment of these modification at the housekeeping gene GAPDH. Analogously, inactive histone modification enrichment at sequences of interest was normalized to the testis-specific gene CTCFL.
Statistical calculations were performed by the methods described in the respective figure or table legends. The cutoff for cancer-specific hypermethylation was calculated by receiver operating characteristic (ROC) analysis. Euclidean k-medians clustering was used to classify the continous methylation measurements. Prognostic parameters were compared by univariate and multivariate Cox proportional hazards models and Harrell's concordance index (c-index).20 Correlation analyses were calculated by Spearman's rho (ρ).
Concomitant DNA methylation of RARRES1 and LXN in PCa tissues and its association with prognosis
DNA methylation analyses of the LXN and RARRES1 promoters were performed by pyrosequencing of 86 bisulfite-treated DNA samples from patients who underwent radical surgery for prostatic adenocarcinoma and 21 DNA samples from adjacent benign prostatic tissues. By ROC analysis, cancer-specific hypermethylation was defined as methylation >6.5% at LXN and >8.5% at RARRES1 with a specificity of 1.0 in the investigated tissue samples. Cancer-specific hypermethylation was observed in 29 and 60% of the cancer tissues at LXN and RARRES1, respectively (Figs. 1b and 1c). Quantitative methylation at the investigated sites was well correlated (Fig. 1d, Spearman's ρ = 0.4, p = 0.000147).
To classify the cases by DNA methylation of both genes, euclidean k-clustering was applied to the continuous quantitative methylation data. This analysis yielded two major clusters (Fig. 1d). In general, cluster 2 consisted of cases with DNA methylation above 20% at the RARRES1 promoter, encompassing the majority of cases with cancer-specific LXN hypermethylation (Fisher's exact test, p < 0.001). Cluster 2 was strongly associated with biochemical recurrence in univariate and multivariate Cox proportional hazards models and could improve the predictive accuracy (c-index)20 of established histopathological parameters up to 0.753 (Table 1 and Fig. 2a). Specifically, this prognostic potential was retained in the subsets of cancers with Gleason 7 (Fig. 2b) or with organ-confined disease (Fig. 2c) and was additionally associated with the Gleason score (χ2-test, p = 0.036) and stage (Fisher's exact test, p = 0.007). We further studied the prognostic potential of cluster 2, as well as cancer-specific hypermethylation of each gene in comparison to continuous methylation data in the entire dataset and in completely random training and test sets (Supporting Information Table S3 and Fig. S1). Clinical outcome in these sets was not associated with cancer-specific (>8.5%) hypermethylation of RARRES1, but strongly with continuous methylation. Cancer-specific hypermethylation of LXN (>6.5%) tended to be associated with biochemical recurrence, but not in a continuous manner. The most robust prognostication was achieved after dividing the samples in two groups by euclidean k-clustering (Supporting Information Table S3) as described above.
Table 1. Uni- and multivariate analyses of RARRES1 hypermethylation and established prognostic parameters
Increasing methylation of the PITX2 gene has similarly been associated with worse prognosis in PCa.21, 22 We have recently confirmed that finding in our own cohort.23 Interestingly, in these samples hypermethylation of PITX2 correlated with that of RARRES1 (Spearman's ρ = 0.544, p = 7.7 × 10−8) and LXN (Spearman's ρ = 0.351, p = 9.6 × 10−4).
The SNP rs6441224 in the RARRES1 promoter is associated with cancer-specific hypermethylation, lower transcriptional activity and is contained in an ETS-family-response element
Our pyrosequencing assay for RARRES1 covers the SNP rs6441224 near the transcriptional start site. Overall, the frequencies of the genotypes were measured as 34.9% (30/86) for C/C, 45.5% (39/86) for C/T and 19.8% (17/86) for T/T with a calculated minor allele frequency (MAF) of 0.424 (as compared to SNP database MAF T-allele = 0.448). Intriguingly, DNA hypermethylation was significantly more frequently observed in alleles containing a T rather than a C at this position (Fig. 3a). Homozygous patients had a slightly worse clinical outcome (Supporting Information Fig. S1). In contrast, the SNP rs9841 in the 5′-UTR of the LXN promoter, located within a CTCF binding site, was not associated with methylation.
To understand the mechanism underlying the preferential methylation of the rs6441224 T-allele, we investigated the sequence for putative transcription factors binding sites using TRAP.24 The studied SNP rs6441224 was predicted to be located within an ETS-family-response element and to modulate the binding affinity of further transcription factors, with higher affinities predicted for the C-allele (Supporting Information Table S4).
These predictions were experimentally investigated by luciferase reporter assays. As predicted, the RARRES1 promoter containing the C-allele displayed a significantly higher activity than its T-allele counterpart (Figs. 3b and 3c). Cotransfection of ETS1 or ERG augmented promoter activity in Du145 and PC-3 cell lines. Interestingly, basal activities of the RARRES1 promoter were lower in the RARRES1 methylated PC-3 than in the unmethylated Du145 cells as compared to the cloning vector pGL3basic.
Epigenetic patterns and transcriptional regulation at 3q25.32 in PCa cell lines
DNA methylation of RARRES1 and LXN was further analyzed in the commonly used PCa cell lines LNCaP, 22Rv1, PC-3, MDAPCa, Du145, normal prostate epithelial cells (PrEC) (Fig. 4a), several bladder cancer cell lines and leukocytes from peripheral blood (Supporting Information Fig. S2). Concomitant RARRES1 and LXN hypermethylation were detected in LNCaP, 22Rv1 and the bladder cancer cell line HT1376, whereas PC-3, MDAPCa and EJ presented RARRES1 hypermethylation only. The Du145 cell line represented the rare case of an exclusive hypermethylation of the LXN promoter, which was also detected in individual PCa tissue samples (Fig. 4a, compare Fig. 1d). Intriguingly, in this cell line, four alleles each were fully methylated or completely unmethylated, a situation suggestive of allele-specific methylation. The methylation patterns observed in the cancer tissues were therefore also represented in PCa cell lines (Figs. 1d and 4a).
In the cell lines, concomitant methylation of RARRES1 and LXN was associated with decreasing expression, as compared to benign prostatic tissue. Treatment with the DNA methyltransferase inhibitor 5-aza-2′-deoxycytidine induced LXN and RARRES1 expression. In Du145 cells, which express RARRES1, only LXN expression was induced (Fig. 4c).
As the various DNA hypermethylation states in the tissues were also detected in specific PCa cell lines, these could be used to conveniently analyze the association of DNA hypermethylation with specific histone modifications. Repression of RARRES1 in the cell lines was associated with a switch from H3K9 acetylation (Fig. 5a) to trimethylation (Fig. 5b) and an increase in H3K27 trimethylation (Fig. 5c) at its promoter. The same changes accompanied silencing of LXN in LNCaP cells. In contrast, the unique methylation pattern at the LXN promoter in Du145 (Fig. 4a) was associated with strong acetylation of H3K9 at the RARRES1 promoter, which is in line with the transcriptional activity and the DNA-unmethylated state of RARRES1 in this cell line.
Transcriptional regulation at 3q25.32 in benign and cancerous prostatic tissues
Cancer-specific hypermethylation of LXN was associated with its transcriptional downregulation (Supporting Information Fig. S3). Presumably because of the moderate fraction of cases with LXN hypermethylation, expression of this gene overall was not significantly diminished in carcinomas compared to benign prostatic tissues (Supporting Information Fig. S4). Expression of RARRES1 tended to be downregulated in cancer tissues overall. However, hypermethylation of RARRES1 itself was not significantly associated with transcriptional downregulation. Instead, RARRES1 downregulation was most pronounced in cases with hypermethylation of LXN (Supporting Information Fig. S3). RARRES1 and LXN are paralogous and neighboring genes. Indeed, expression of the two genes was strongly correlated in benign prostatic tissues (Spearman's ρ = 0.879, p < 0.00001) and to a lesser extent in cancers. To investigate the influence of the epigenetic changes on further genes at 3q.25.32 (Fig. 1a), we measured the expression of the adjacent or overlapping genes GFM1, MLF1 and MFSD1, which are transcribed from the opposite strand. Expression of GFM1 and MFSD1 was unchanged between cancerous and benign tissues, whereas MLF1 (myeloid leukemia factor) tended to be upregulated (Supporting Information Fig. S4). Expression of none of these genes was correlated with that of LXN or RARRES1.
Despite many reports of hypermethylated genes in PCa, it has proven difficult to develop classifications of the disease based on these epigenetic changes. One reason for this difficulty may be that hypermethylation of a substantial number of genes, exemplified by GSTP1,25 is so common that it cannot serve to distinguish subgroups of the disease. Hypermethylation of RARRES1 (TIG1) has been observed by several groups to occur in around 60% of PCas13, 14, 26 and hypermethylation of this gene per se may remain too frequent to distinguish subgroups of cancers with better or worse prognosis. In theory, prognostic classifications of the disease might be obtained from investigations of hypermethylation events at genes that are methylated in a smaller proportion of the cases or by determining methylation in a quantitative manner, based on the hypothesis that stronger methylation is associated with more aggressive disease. Indeed, quantitative estimates of hypermethylation at the PITX2 locus have been shown to prognosticate a worse outcome after radical prostatectomy.21, 22
We have now investigated methylation of RARRES1 alongside its paralogous neighbor gene LXN at 3q25.32 in PCas by quantitative pyrosequencing. Hypermethylation of LXN was relatively common in PCa tissues (25/86 cases). Although ours is the first investigation on PCa, LXN hypermethylation has been previously observed in medulloblastoma, gastric carcinoma and melanoma cell lines.27–29 Our exemplary investigation of bladder cell lines suggests that it also occurs in urothelial cancers. In our study, LXN hypermethylation was essentially restricted to cancers with RARRES1 hypermethylation, especially to those with stronger RARRES1 methylation. The observation that the cases with LXN hypermethylation are a subset of those with RARRES1 hypermethylation and that higher levels of RARRES1 methylation are associated with hypermethylation of the adjacent LXN promoter suggests an expansion of a repressed epigenetic state from RARRES1 to LXN in a subset of PCas. These findings and the strong correlation of the expression of the two paralogous genes in tissues point to a common regulation. The common regulation of specifically these two genes may also explain why changes in gene expression are not found at the neighboring genes GFM1, MLF1 and MFSD1. Insofar, the situation at the LXN/RARRES1 twin locus differs from other instances of “regional” epigenetic silencing in cancer.6, 7 Importantly, increasing methylation at 3q25.32 also characterizes cases with higher Gleason score and tumor stage as well as worse clinical outcome. Notably, cancer-specific methylation changes at RARRES1 alone were not associated with such properties and methylation changes at LXN on their own less strongly.
The epigenetic states observed in cancer tissues were also represented in commonly used PCa cell lines. Chromatin immunoprecipitation in representative cell lines revealed the canonical close association between loss of acetylation and gain of trimethylation at H3K9 in concert with transcriptional silencing and increased DNA methylation. Interestingly, restoration of transcription by treatment with 5-aza-2′-deoxycytidine appeared to be more limited in cell lines in which methylation had expanded to LXN. This could be explained by assuming that epigenetic silencing would become more intense as methylation expands from RARRES1 to LXN.
A further finding in our study was an association between the genotype at the SNP rs6441224 and the methylation of RARRES1. A few such instances have been reported previously. In colorectal and lung cancers, the SNP rs16906252 is associated with hypermethylation susceptibility of MGMT30, 31 and rs1800734 with MLH1 hypermethylation,32 but the molecular basis of these associations remains to be determined. A polymorphism encompassing insertion/deletion of several base pairs in the RIL gene promoter influences binding of the Sp1 or Sp3 transcriptional activators and is associated with differential susceptibility to hypermethylation in various cancers.33 We have now demonstrated that the T-allele of RARRES1 exhibiting higher methylation in PCa tissues conferred weaker transcriptional activity. Similarly, SNP rs16906252 influences both the epigenetic state and the transcriptional activity of the MGMT promoter.34 Furthermore, we discovered that the SNP rs6441224 is located within a predicted binding site for ETS-family transcription factors and that the RARRES1 promoter responds strongly to ETS1 or ERG.
In conclusion, we propose that intensification of an aberrant epigenetic state at two adjacent genes located at 3q25.32 may characterize a particularly aggressive subset of PCas. Quantitative methylation assays for this region should be included into methylation biomarker studies aiming at individual prognostication of PCa. Our data suggest that the mechanism underlying this particular epigenetic aberration depends on the genotype of each patient.
The authors thank Prof. Rainer Engers for histopathological evaluation of their samples, Marc Ingenwerth for help with collecting the clinical data, Dr. Michèle Hoffmann, Dr. Zaki Sheikhibrahim and Prof. Nicolas Wernert for kindly providing plasmid constructs and Ms. Christiane Hader for experimental support.