NMD and microRNA expression profiling of the HPCX1 locus reveal MAGEC1 as a candidate prostate cancer predisposition gene

Prostate cancer is the most common form of cancer affecting men in the Western world. In Finland, there were 4234 new cancer cases diagnosed in 2008, and the incidence of prostate cancer (PC) was 82.9/100.000 [1]. In addition to age, a well-established risk factor for PC is a family history of the disease. In a large Scandinavian twin study [2], it was reported that approximately 40% of the risk for PC can be explained by heritable components. This proportion is the highest ever reported for a common malignancy. Most of the genes that are involved in the causation of hereditary cancers have been identified by linkage analysis. Several linkage studies of hereditary prostate cancer (HPC) have been performed and the results have implicated many risk loci located on different chromosomes, which indicates a great heterogeneity of this disease [3]. One of the loci found by linkage analysis is HPCX1 (OMIM %300147), which is located on chromosome Xq27-q28 [4]. This locus has proven to be important in the Finnish population [5] and the region around the best linkage marker was found to be in strong linkage equilibrium [6]. However, the susceptibility gene has not yet been identified because the chromosomal region has an extremely complex genomic structure with multiple gene duplications and inversions that have hampered conventional gene cloning methods [7]. The SPANX genes and LDOC1 at Xq27 have been considered to be the best positional candidate genes for HPCX1, but no direct evidence for causative mutations in any of the genes studied have been detected [8, 9].

Mutations, especially nonsense mutations, in tumor suppressor genes (TSGs) are common in the development and progression of cancer. They give rise to in-frame premature translation termination codons within the coding regions of genes and lead to truncated protein translation products. However, the identification of TSGs by classical cancer genetics methods is difficult and slow. In addition, RNA transcripts carrying nonsense mutations are usually targeted for degradation through nonsense-mediated decay (NMD) [10]. NMD is a complex process in mammalian mRNA metabolism, and its function is to eliminate faulty transcripts and control the expression of normal genes.

A conventional strategy for the identification of disease genes is to use microarrays to compare the levels of gene-specific mRNA expression between patient and control samples. However, identification of the mutated gene can be obscured by inter-individual variation and secondary changes in gene expression caused by the disease process. Noensie and Dietz [11] reported an alternative strategy that circumvents these limitations, called GINI (Gene Identification by NMD Inhibition), in which the patient sample is compared to itself after the pharmacological inhibition of NMD. Microarrays are then used to identify potential nonsense transcripts that are increased in abundance after the loss of NMD. Emetine was used to block the pathway, but was problematic as emetine induces a stress response that results in the upregulation of additional transcripts. Ionov et al. [12] combined the emetine treatment with actinomycin D, which effectively prevents the upregulation of stress response genes while still stabilizing mutant transcripts.

Inactivation of autosomal tumor suppressor genes is a two-step process involving the mutation of the target gene and the loss of the wild type allele. In lymphoblastoid cell lines established from patient samples, the normal wild type allele can mask the effect of a mutated allele. However, because males have only one X chromosome, there is only one allele of the X chromosomal germline genes. Therefore, truncated tumor suppressor mRNAs may be identified by using NMD method applied for RNA extracted from patients' lymphoblastoid cell lines.

Cancer is fundamentally a disease of disordered gene expression. Since no causative mutations have been identified from the transcribed genes of the HPCX1 region, it is possible that the defect occurs at the regulatory level. Regulatory defects might also explain the relatively late onset of the disease, assuming a polygenic model for PC development with additive effect to the phenotype. MicroRNAs (miRNAs) are small RNA molecules that regulate gene expression post-transcriptionally [13]. They play an important role in diverse biological processes and, accordingly, altered miRNA expression is likely to contribute to human disease, including cancer. It has been shown that miRNA profiles are surprisingly informative. They become altered with the development and progression of PC [14] and have a very important role in the biology of the disease [15]. Since considerable amount of miRNAs are located within intronic regions and regulated by the host gene promoter [16], miRNA expression profiles give us one possibility to study disease related variations in non-protein coding chromosomal areas and this could lead to identification of regulatory variants especially in region with complex genomic structure like one in chromosome X.

Here, we present a study with Finnish multiplex HPCX1 linked families in which we have characterized the HPCX1 locus by NMD and miRNA microarray methods and evaluated the role of HPCX1 in the causation of familial prostate cancer.

In order to identify genes containing inactivating mutations in the Xq27-q28 region, an NMD microarray analysis with Agilent 44 K Whole Human Genome oligonucleotide microarrays was performed in the families showing the strongest linkage to HPCX1. The candidate genes (n = 17) for subsequent sequence analysis were selected according to the microarray analysis (RBMX, CSAG2, RAP2C, SOX3, MBNL3, ZNF75, MAGEC1, MAGEA1, MAGEA11, MAGEC3, MAGED1, U66046, SSR4, VBP1, LDOC1, TKTL1, CD40LG) (Table 2). All identified stress response genes were excluded [21]. No truncating mutations were detected, but a total of 34 changes were found by direct sequencing. Eight of the changes were missense variants, six were silent changes, and twenty of the variants took place in introns, 5'UTR, or 3'UTR regions. Twenty of the changes were novel and not found in any genomic databases. A summary of the identified variants is presented in Table 3.

The most interesting variation selected for follow-up was the MAGEC1 p.Met1? (c.2T > C; Met > Thr) start codon mutation found in family 311. The frequency of that mutation was determined by sequencing among patients with unselected PC or HPC and in the control groups. The carrier frequencies for p.Met1? were 2.45%, 1.72%, 0.92%, 0.53%, 1.21%, and 0.65% in the probands with HPC, unselected PC cases, male population controls, BPH controls, PSA controls, and female population controls, respectively. The frequency of P.Met1? was found to be in Hardy-Weinberg equilibrium in controls. A statistically significant difference was observed in the carrier frequencies of the P.Met1? variant between the sample groups, and an association was seen between the variant and unselected PC and HPC (see Table 4). The association was strongest when male and female blood donors and BPH patients were used as a control group (OR = 2.35, 95% CI = 1.10-5.02 for unselected PC; OR = 3.38, 95% CI = 1.10-10.40 for HPC). The association between the frequency of the variant and the disease phenotype, including tumor WHO grade, Gleason score, T-stage, age at diagnosis, and PSA value at diagnosis, was also analyzed among the unselected PC cases as no complete data were available for the HPC cases. No significant associations were found from these studies (data not shown). Additional samples from three prostate cancer families carrying the p.Met1? variant were also analyzed, but cosegregation of the p.Met1? was incomplete (data not shown) which is quite expected for a low penetrance gene.

MicroRNA expression levels in lymphoblastoid cell lines were determined with Agilent Human miRNA arrays. The most differentially expressed miRNAs when testing against both alternatives (miRNA expression is either up- or downregulated in case vs. controls) are presented in Table 5 and 6 together with information about genomic and intronic/intergenic location. The miRanda algorithm produced 1211 different variant-miRNA combinations with a total score value above the cut-off value. From the 29 differentially expressed miRNAs between patients and healthy individuals twelve miRNAs for validation were selected based on that they supposedly had a target site in MAGEC1 gene (Table 5 and 6). In addition, miRNAs located in X chromosome were selected for validation. Validation of these selected miRNA expressions in cell lines was based on specific TaqMan MicroRNA Assays (Tables 5 and 6, Figures 1, 2 and 3).

The recognition that PC clusters within families has led to the collection of HPC families with the goal of localizing and identifying PC susceptibility genes. Initially, the PC susceptibility locus, HPCX1, was mapped to the X chromosome by using a set of high-risk PC families from the United States, Finland, and Sweden. The effect of HPCX1 was proven to be the most important in the Finnish population, where a specific haplotype in the region was also identified [5, 6]. Further, X-chromosomal inheritance coheres with the results of our segregation analysis of multifactorial recessive inheritance as the only model in the Finnish population [22]. In this study, a recently developed NMD microarray technology was utilized for the analysis of the HPCX1 region with brother pairs from HPCX1 linked families. Use of the unaffected brothers as controls might be problematic given the late age of prostate cancer diagnosis, but the unaffected brothers were chosen to be the oldest possible from every family with no history of PC. Previously, the manipulation of NMD together with expression array analysis has proven to be a powerful tool for detecting novel gene mutations in cancer cells. Huusko et al. [23] successfully identified EPHB2 gene mutations in PC. Since then, mutations in melanoma cell lines [24], colon cancer cell lines [25], and PC cell lines [26] have been identified by inhibition of NMD.

Sequencing of 17 genes from the HPCX1 region did not reveal any truncating nonsense mutations. The most interesting variation for follow-up was in the start codon (p.Met1?) of the MAGEC1 gene. An association was later seen between the p.Met1? variant and unselected PC and HPC. Interestingly, the association was strongest when "supernormal" PSA controls were excluded from the control group. The PSA controls consisted of men with PSA levels < 1.0 ng/ml and a mean age of 67.5 years, which is lower than the mean age of BPH patients (73 years). As HPCX1 is suggested to be a late-onset disease in the Finnish population [5], as well as in some other populations [27], it is possible that the younger PSA control group is actually more saturated with the p.Met1? individuals who are at risk of developing late-onset HPCX. MAGEC1 is a member of the melanoma antigen gene (MAGE) family [28]. The proteins of this family are tumor-specific antigens that can be recognized by autologous cytolytic T lymphocytes. MAGEC1 is composed of four exons and encodes a protein of 1142 amino acids. It is approximately 800 residues longer than other MAGE proteins due to the insertion of a large number of short repetitive sequences in front of the MAGE-homologous sequence. MAGEC1 is expressed in a significant proportion of tumors of various histological types, but is silent in normal tissues, excluding the testis. Alternative start codons, mainly GTG and TTG, are used in prokaryotes and, very rarely, in higher organisms. One example is the vitamin-D receptor (VDR) gene start codon polymorphism, where a T/C polymorphism in the first of the two potential start (ATG) codons results in two alleles that can be distinguished by RFLP using the endonuclease FokI [29]. The biological function of the identified MAGEC1 start codon variant in this case is difficult to assess since we were not able to investigate the expression of MAGEC1 in individuals carrying the mutated allele in the absence of other tissue material. Considering the conserved structure and similar functions of MAGE proteins, it might be possible that other members of the gene family can partly compensate for the functions of MAGEC1 if the start codon mutation totally blocks the translation of the gene.

In this NMD array analysis, the false positive rate was evidently high since no truncating mutations were found. Emetine treatment followed by actinomycin D treatment was used, but it has been observed that treatment with actinomycin D after emetine incubation does not have a significant effect on treated cells, suggesting that the combination of the drugs is not the best possible method for this type of study [30]. A novel improvement to the NMD protocol includes a combination of emetine and caffeine treatment [25], which leads to a more efficient identification of false positives produced by cell stress. On the other hand, we might have missed genes that actually carry truncating mutations. If the genes were mutated both in the seemingly healthy males with a normal clinical phenotype and their already affected siblings, they would have had the same profile as the stress response genes and would have, therefore, been excluded. In addition, relevant variants might have been missed by being limited to use lymphoblastoid cell lines, as they may not resemble the whole set of active genes in prostate tissue. Although there is substantial amount of evidence that lymphoblastoid cells encompasses a variety of metabolic pathways that are specific to individuals where the cells originated, making these cell lines suitable for molecular and functional studies [31].

Despite its association with PC, the segregation of the P.Met1? variant in studied families was incomplete. This finding can also reflect the high phenocopy rate of the disease, and as a consequence, even the most extreme families may have a few sporadic cases. Currently, there are no clinical or molecular tools to distinguish "true" HPC cases from sporadic ones. Therefore, the incomplete segregation does not necessarily diminish the importance of p.Met1? as a HPCX1 risk variant.

The number of publications reporting differential miRNA expression in prostate cancer is constantly growing. Analyses have been performed for example in clinical prostate specimens [32], and in PC cell lines, xenografts, BPH vs. PC samples [33]. Most recently, miRNA expression profiles of androgen-responsive and castrate-resistant PC cell lines were compared [34]. In all of these studies, a set of 20-30 miRNAs unequivocally differentiated the PC samples from the normal samples and non-malignant precursor lesions, and specific miRNAs including miR-125b, miR-145, and let-7c were repeatedly detected differentially expressed in different studies.

In the current study, we hypothesized that miRNA profiling could be used as a tool for discovering variants from non-protein coding regions which could explain the "dark" inheritance behind HPCX. Altered miRNA expression in patients lymphoblastoid cells could lead to the identification of germline variants in promoter or other regulatory regions of protein coding genes since considerable amount of miRNA expression is correlated to host and target gene expression [35]. Possible explanation of differences in miRNA levels could be their role in tumor development and initiation. Already, several studies have shown that miRNAs in serum can be considered as biomarkers [36, 37] and serum miRNAs are mainly derived from circular blood cells [38], the same way as lymphoblastoid cells. We were able to identify 29 significantly differently expressed miRNAs between patients and their healthy brothers. Validation of the expressions of the selected 16 miRNAs with TaqMan assays were successful in 14/16 cases, including all four miRNAs located in the X chromosome.

In conclusion, we suggest a role for MAGEC1 in genetic PC susceptibility, especially in the HPCX1-linked form of the disease. The start codon missense variation in the MAGEC1 gene showed a borderline association between the variant and both HPC and unselected PC, and therefore additional research is warranted. In addition, the role of certain miRNAs needs further study, especially since MAGEC1 was predicted to be one of their targets.